lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/osd/osd_handler.c
  32  *
  33  * Top-level entry points into osd module
  34  *
  35  * Author: Nikita Danilov <nikita@clusterfs.com>
  36  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_OSD
  40
  41 #include <linux/fs_struct.h>
  42 #include <linux/module.h>
  43 #include <linux/user_namespace.h>
  44 #include <linux/uidgid.h>
  45
  46 /* prerequisite for linux/xattr.h */
  47 #include <linux/types.h>
  48 /* prerequisite for linux/xattr.h */
  49 #include <linux/fs.h>
  50 /* XATTR_{REPLACE,CREATE} */
  51 #include <linux/xattr.h>
  52
  53 #include <ldiskfs/ldiskfs.h>
  54 #include <ldiskfs/xattr.h>
  55 #include <ldiskfs/ldiskfs_extents.h>
  56 #undef ENTRY
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <uapi/linux/lustre/lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <lustre_linkea.h>
  76
  77 /* encoding routines */
  78 #include <lustre_crypto.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC_PTR_ARRAY_LARGE(idc, i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE_PTR_ARRAY_LARGE(oti->oti_ins_cache,
 221                                            oti->oti_ins_cache_used);
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 static void osd_idc_dump_lma(const struct lu_env *env,
 290                                 struct osd_device *osd,
 291                                 unsigned long ino,
 292                                 bool check_in_oi)
 293 {
 294         struct osd_thread_info *info = osd_oti_get(env);
 295         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 296         const struct lu_fid *fid;
 297         struct osd_inode_id lid;
 298         struct inode *inode;
 299         int rc;
 300
 301         inode = osd_ldiskfs_iget(osd_sb(osd), ino);
 302         if (IS_ERR(inode)) {
 303                 CERROR("%s: can't get inode %lu: rc = %d\n",
 304                        osd->od_svname, ino, (int)PTR_ERR(inode));
 305                 return;
 306         }
 307         if (is_bad_inode(inode)) {
 308                 CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
 309                 goto put;
 310         }
 311         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 312         if (rc) {
 313                 CERROR("%s: can't get LMA for %lu: rc = %d\n",
 314                        osd->od_svname, ino, rc);
 315                 goto put;
 316         }
 317         fid = &loa->loa_lma.lma_self_fid;
 318         LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
 319                       PFID(fid), ino, (unsigned)inode->i_generation);
 320         if (!check_in_oi)
 321                 goto put;
 322         rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
 323         if (rc) {
 324                 CERROR("%s: can't lookup "DFID": rc = %d\n",
 325                        osd->od_svname, PFID(fid), rc);
 326                 goto put;
 327         }
 328         LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
 329                       PFID(fid), lid.oii_ino, lid.oii_gen);
 330 put:
 331         iput(inode);
 332 }
 333
 334 static void osd_idc_dump_debug(const struct lu_env *env,
 335                                 struct osd_device *osd,
 336                                 const struct lu_fid *fid,
 337                                 unsigned long ino1,
 338                                 unsigned long ino2)
 339 {
 340         struct osd_inode_id lid;
 341
 342         int rc;
 343
 344         rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
 345         if (!rc) {
 346                 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
 347                         osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
 348                 osd_idc_dump_lma(env, osd, lid.oii_ino, false);
 349         } else {
 350                 CERROR("%s: can't lookup "DFID": rc = %d\n",
 351                        osd->od_svname, PFID(fid), rc);
 352         }
 353         if (ino1)
 354                 osd_idc_dump_lma(env, osd, ino1, true);
 355         if (ino2)
 356                 osd_idc_dump_lma(env, osd, ino2, true);
 357 }
 358
 359 /*
 360  * lookup mapping for given FID and fill it from the given object.
 361  * the object is lolcal by definition.
 362  */
 363 static int osd_idc_find_and_init(const struct lu_env *env,
 364                                  struct osd_device *osd,
 365                                  struct osd_object *obj)
 366 {
 367         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 368         struct osd_idmap_cache *idc;
 369
 370         idc = osd_idc_find(env, osd, fid);
 371         LASSERT(!IS_ERR(idc));
 372         if (idc != NULL) {
 373                 if (obj->oo_inode == NULL)
 374                         return 0;
 375                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 376                         if (idc->oic_lid.oii_ino) {
 377                                 osd_idc_dump_debug(env, osd, fid,
 378                                                    idc->oic_lid.oii_ino,
 379                                                    obj->oo_inode->i_ino);
 380                                 return -EINVAL;
 381                         }
 382                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 383                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 384                 }
 385                 return 0;
 386         }
 387
 388         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 389                osd->od_svname, PFID(fid));
 390
 391         /* new mapping is needed */
 392         idc = osd_idc_add(env, osd, fid);
 393         if (IS_ERR(idc)) {
 394                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 395                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 396                 return PTR_ERR(idc);
 397         }
 398
 399         if (obj->oo_inode != NULL) {
 400                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 401                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 402         }
 403         return 0;
 404 }
 405
 406 /*
 407  * OSD object methods.
 408  */
 409
 410 /*
 411  * Concurrency: no concurrent access is possible that early in object
 412  * life-cycle.
 413  */
 414 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 415                                           const struct lu_object_header *hdr,
 416                                           struct lu_device *d)
 417 {
 418         struct osd_object *mo;
 419
 420         OBD_ALLOC_PTR(mo);
 421         if (mo != NULL) {
 422                 struct lu_object *l;
 423                 struct lu_object_header *h;
 424                 struct osd_device *o = osd_dev(d);
 425
 426                 l = &mo->oo_dt.do_lu;
 427                 if (unlikely(o->od_in_init)) {
 428                         OBD_ALLOC_PTR(h);
 429                         if (!h) {
 430                                 OBD_FREE_PTR(mo);
 431                                 return NULL;
 432                         }
 433
 434                         lu_object_header_init(h);
 435                         lu_object_init(l, h, d);
 436                         lu_object_add_top(h, l);
 437                         mo->oo_header = h;
 438                 } else {
 439                         dt_object_init(&mo->oo_dt, NULL, d);
 440                         mo->oo_header = NULL;
 441                 }
 442
 443                 mo->oo_dt.do_ops = &osd_obj_ops;
 444                 l->lo_ops = &osd_lu_obj_ops;
 445                 init_rwsem(&mo->oo_sem);
 446                 init_rwsem(&mo->oo_ext_idx_sem);
 447                 spin_lock_init(&mo->oo_guard);
 448                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 449                 return l;
 450         }
 451         return NULL;
 452 }
 453
 454 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 455                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 456 {
 457         int rc;
 458
 459         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 460                              (void *)loa, sizeof(*loa));
 461         if (rc > 0) {
 462                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 463
 464                 if (rc < sizeof(*lma))
 465                         return -EINVAL;
 466
 467                 rc = 0;
 468                 lustre_loa_swab(loa, true);
 469                 /* Check LMA compatibility */
 470                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 471                         rc = -EOPNOTSUPP;
 472                         CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
 473                               osd_ino2name(inode),
 474                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 475                               PFID(&lma->lma_self_fid), inode->i_ino, rc);
 476                 }
 477         } else if (rc == 0) {
 478                 rc = -ENODATA;
 479         }
 480
 481         return rc;
 482 }
 483
 484 /*
 485  * retrieve object from backend ext fs.
 486  **/
 487 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 488                        struct osd_inode_id *id)
 489 {
 490         int rc;
 491         struct inode *inode = NULL;
 492
 493         /*
 494          * if we look for an inode withing a running
 495          * transaction, then we risk to deadlock
 496          * osd_dirent_check_repair() breaks this
 497          */
 498          /* LASSERT(current->journal_info == NULL); */
 499
 500         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 501         if (IS_ERR(inode)) {
 502                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 503                        id->oii_ino, PTR_ERR(inode));
 504         } else if (id->oii_gen != OSD_OII_NOGEN &&
 505                    inode->i_generation != id->oii_gen) {
 506                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 507                        "i_generation = %u\n",
 508                        id->oii_ino, id->oii_gen, inode->i_generation);
 509                 iput(inode);
 510                 inode = ERR_PTR(-ESTALE);
 511         } else if (inode->i_nlink == 0) {
 512                 /*
 513                  * due to parallel readdir and unlink,
 514                  * we can have dead inode here.
 515                  */
 516                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 517                 iput(inode);
 518                 inode = ERR_PTR(-ESTALE);
 519         } else if (is_bad_inode(inode)) {
 520                 CWARN("%s: bad inode: ino = %u: rc = %d\n",
 521                       osd_dev2name(dev), id->oii_ino, -ENOENT);
 522                 iput(inode);
 523                 inode = ERR_PTR(-ENOENT);
 524         } else  if (osd_is_ea_inode(inode)) {
 525                 /*
 526                  * EA inode is internal ldiskfs object, should don't visible
 527                  * on osd
 528                  */
 529                 CDEBUG(D_INODE, "EA inode: ino = %u\n", id->oii_ino);
 530                 iput(inode);
 531                 inode = ERR_PTR(-ENOENT);
 532         } else if ((rc = osd_attach_jinode(inode))) {
 533                 iput(inode);
 534                 inode = ERR_PTR(rc);
 535         } else {
 536                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 537                 if (id->oii_gen == OSD_OII_NOGEN)
 538                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 539
 540                 /*
 541                  * Do not update file c/mtime in ldiskfs.
 542                  * NB: we don't have any lock to protect this because we don't
 543                  * have reference on osd_object now, but contention with
 544                  * another lookup + attr_set can't happen in the tiny window
 545                  * between if (...) and set S_NOCMTIME.
 546                  */
 547                 if (!(inode->i_flags & S_NOCMTIME))
 548                         inode->i_flags |= S_NOCMTIME;
 549         }
 550         return inode;
 551 }
 552
 553 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 554                           handle_t *handle, struct dentry *child,
 555                           struct inode *inode, struct htree_lock *hlock)
 556 {
 557         int rc, rc2;
 558
 559         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 560         if (rc == -ENOBUFS || rc == -ENOSPC) {
 561                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 562                 struct inode *parent = child->d_parent->d_inode;
 563                 struct lu_fid *fid = NULL;
 564                 char fidstr[FID_LEN + 1] = "unknown";
 565
 566                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 567                 if (!rc2) {
 568                         fid = &loa->loa_lma.lma_self_fid;
 569                 } else if (rc2 == -ENODATA) {
 570                         if (unlikely(is_root_inode(parent))) {
 571                                 fid = &info->oti_fid3;
 572                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 573                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 574                                 fid = &info->oti_fid3;
 575                                 lu_igif_build(fid, parent->i_ino,
 576                                               parent->i_generation);
 577                         }
 578                 }
 579
 580                 if (fid != NULL)
 581                         snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
 582
 583                 /* below message is checked in sanity.sh test_129 */
 584                 if (rc == -ENOSPC) {
 585                         CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
 586                               osd_name(osd), parent->i_ino, fidstr);
 587                 } else {
 588                         rc = 0; /* ignore such error now */
 589                         CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
 590                               osd_name(osd), parent->i_ino, fidstr);
 591                 }
 592
 593         }
 594
 595         return rc;
 596 }
 597
 598
 599 struct inode *
 600 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 601              struct osd_inode_id *id, struct lu_fid *fid)
 602 {
 603         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 604         struct inode *inode;
 605         int rc;
 606
 607         inode = osd_iget(info, dev, id);
 608         if (IS_ERR(inode))
 609                 return inode;
 610
 611         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 612         if (!rc) {
 613                 *fid = loa->loa_lma.lma_self_fid;
 614         } else if (rc == -ENODATA) {
 615                 if (unlikely(is_root_inode(inode)))
 616                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 617                 else
 618                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 619         } else {
 620                 iput(inode);
 621                 inode = ERR_PTR(rc);
 622         }
 623         return inode;
 624 }
 625
 626 static struct inode *osd_iget_check(struct osd_thread_info *info,
 627                                     struct osd_device *dev,
 628                                     const struct lu_fid *fid,
 629                                     struct osd_inode_id *id,
 630                                     bool trusted)
 631 {
 632         struct inode *inode;
 633         int rc = 0;
 634
 635         ENTRY;
 636
 637         /*
 638          * The cached OI mapping is trustable. If we cannot locate the inode
 639          * via the cached OI mapping, then return the failure to the caller
 640          * directly without further OI checking.
 641          */
 642
 643 again:
 644         inode = osd_iget(info, dev, id);
 645         if (IS_ERR(inode)) {
 646                 rc = PTR_ERR(inode);
 647                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 648                         goto check_oi;
 649
 650                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 651                        PFID(fid), id->oii_ino, rc);
 652                 GOTO(put, rc);
 653         }
 654
 655 check_oi:
 656         if (rc != 0) {
 657                 __u32 saved_ino = id->oii_ino;
 658                 __u32 saved_gen = id->oii_gen;
 659
 660                 LASSERT(!trusted);
 661                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 662
 663                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 664                 /*
 665                  * XXX: There are four possible cases:
 666                  *      1. rc = 0.
 667                  *         Backup/restore caused the OI invalid.
 668                  *      2. rc = 0.
 669                  *         Someone unlinked the object but NOT removed
 670                  *         the OI mapping, such as mount target device
 671                  *         as ldiskfs, and modify something directly.
 672                  *      3. rc = -ENOENT.
 673                  *         Someone just removed the object between the
 674                  *         former oi_lookup and the iget. It is normal.
 675                  *      4. Other failure cases.
 676                  *
 677                  *      Generally, when the device is mounted, it will
 678                  *      auto check whether the system is restored from
 679                  *      file-level backup or not. We trust such detect
 680                  *      to distinguish the 1st case from the 2nd case:
 681                  *      if the OI files are consistent but may contain
 682                  *      stale OI mappings because of case 2, if iget()
 683                  *      returns -ENOENT or -ESTALE, then it should be
 684                  *      the case 2.
 685                  */
 686                 if (rc != 0)
 687                         /*
 688                          * If the OI mapping was in OI file before the
 689                          * osd_iget_check(), but now, it is disappear,
 690                          * then it must be removed by race. That is a
 691                          * normal race case.
 692                          */
 693                         GOTO(put, rc);
 694
 695                 /*
 696                  * It is the OI scrub updated the OI mapping by race.
 697                  * The new OI mapping must be valid.
 698                  */
 699                 if (saved_ino != id->oii_ino ||
 700                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 701                         if (!IS_ERR(inode))
 702                                 iput(inode);
 703
 704                         trusted = true;
 705                         goto again;
 706                 }
 707
 708                 if (IS_ERR(inode)) {
 709                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 710                             SF_INCONSISTENT)
 711                                 /*
 712                                  * It still can be the case 2, but we cannot
 713                                  * distinguish it from the case 1. So return
 714                                  * -EREMCHG to block current operation until
 715                                  *  OI scrub rebuilt the OI mappings.
 716                                  */
 717                                 rc = -EREMCHG;
 718                         else
 719                                 rc = -ENOENT;
 720
 721                         GOTO(put, rc);
 722                 }
 723
 724                 if (inode->i_generation == id->oii_gen)
 725                         rc = -ENOENT;
 726                 else
 727                         rc = -EREMCHG;
 728         }
 729
 730         GOTO(put, rc);
 731
 732 put:
 733         if (rc != 0) {
 734                 if (!IS_ERR(inode))
 735                         iput(inode);
 736
 737                 inode = ERR_PTR(rc);
 738         }
 739
 740         return inode;
 741 }
 742
 743 /**
 744  * \retval +v: new filter_fid does not contain self-fid
 745  * \retval 0:  filter_fid_18_23, contains self-fid
 746  * \retval -v: other failure cases
 747  */
 748 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 749                  struct dentry *dentry, struct lu_fid *fid)
 750 {
 751         struct filter_fid *ff = &info->oti_ff;
 752         struct ost_id *ostid = &info->oti_ostid;
 753         int rc;
 754
 755         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 756         if (rc == sizeof(struct filter_fid_18_23)) {
 757                 struct filter_fid_18_23 *ff_old = (void *)ff;
 758
 759                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 760                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 761                 /*
 762                  * XXX: use 0 as the index for compatibility, the caller will
 763                  * handle index related issues when necessary.
 764                  */
 765                 if (!rc)
 766                         ostid_to_fid(fid, ostid, 0);
 767         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 768                 rc = 1;
 769         } else if (rc >= 0) {
 770                 rc = -EINVAL;
 771         }
 772
 773         return rc;
 774 }
 775
 776 static int osd_lma_self_repair(struct osd_thread_info *info,
 777                                struct osd_device *osd, struct inode *inode,
 778                                const struct lu_fid *fid, __u32 compat)
 779 {
 780         handle_t *jh;
 781         int rc;
 782
 783         LASSERT(current->journal_info == NULL);
 784
 785         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 786                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 787         if (IS_ERR(jh)) {
 788                 rc = PTR_ERR(jh);
 789                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 790                       osd_name(osd), rc);
 791                 return rc;
 792         }
 793
 794         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 795         if (rc != 0)
 796                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 797                       osd_name(osd), rc);
 798         ldiskfs_journal_stop(jh);
 799         return rc;
 800 }
 801
 802 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 803 {
 804         struct osd_thread_info *info = osd_oti_get(env);
 805         struct osd_device *osd = osd_obj2dev(obj);
 806         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 807         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 808         struct inode *inode = obj->oo_inode;
 809         struct dentry *dentry = &info->oti_obj_dentry;
 810         struct lu_fid *fid = NULL;
 811         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 812         int rc;
 813
 814         ENTRY;
 815
 816         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 817                              (void *)loa, sizeof(*loa));
 818         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 819                 fid = &lma->lma_self_fid;
 820                 rc = osd_get_idif(info, inode, dentry, fid);
 821                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 822                         /*
 823                          * For the given OST-object, if it has neither LMA nor
 824                          * FID in XATTR_NAME_FID, then the given FID (which is
 825                          * contained in the @obj, from client RPC for locating
 826                          * the OST-object) is trusted. We use it to generate
 827                          * the LMA.
 828                          */
 829                         osd_lma_self_repair(info, osd, inode, rfid,
 830                                             LMAC_FID_ON_OST);
 831                         RETURN(0);
 832                 }
 833         }
 834
 835         if (rc < 0)
 836                 RETURN(rc);
 837
 838         if (rc > 0) {
 839                 rc = 0;
 840                 lustre_lma_swab(lma);
 841                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 842                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 843                               S_ISREG(inode->i_mode)))) {
 844                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 845                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 846                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 847                               PFID(rfid), inode->i_ino);
 848                         rc = -EOPNOTSUPP;
 849                 } else {
 850                         fid = &lma->lma_self_fid;
 851                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 852                             osd->od_is_ost)
 853                                 obj->oo_pfid_in_lma = 1;
 854                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 855                             !osd->od_is_ost)
 856                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 857                 }
 858         }
 859
 860         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 861                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 862                         struct ost_id   *oi   = &info->oti_ostid;
 863                         struct lu_fid   *fid1 = &info->oti_fid3;
 864                         __u32            idx  = fid_idif_ost_idx(rfid);
 865
 866                         /*
 867                          * For old IDIF, the OST index is not part of the IDIF,
 868                          * Means that different OSTs may have the same IDIFs.
 869                          * Under such case, we need to make some compatible
 870                          * check to make sure to trigger OI scrub properly.
 871                          */
 872                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 873                                 /* Given @rfid is new, LMA is old. */
 874                                 fid_to_ostid(fid, oi);
 875                                 ostid_to_fid(fid1, oi, idx);
 876                                 if (lu_fid_eq(fid1, rfid)) {
 877                                         if (osd->od_index_in_idif)
 878                                                 osd_lma_self_repair(info, osd,
 879                                                         inode, rfid,
 880                                                         LMAC_FID_ON_OST);
 881                                         RETURN(0);
 882                                 }
 883                         }
 884                 }
 885
 886                 rc = -EREMCHG;
 887         }
 888
 889         RETURN(rc);
 890 }
 891
 892 struct osd_check_lmv_buf {
 893         /* please keep it as first member */
 894         struct dir_context ctx;
 895         struct osd_thread_info *oclb_info;
 896         struct osd_device *oclb_dev;
 897         int oclb_items;
 898         bool oclb_found;
 899 };
 900
 901 /**
 902  * It is called internally by ->iterate*() to filter out the
 903  * local slave object's FID of the striped directory.
 904  *
 905  * \retval      1 found the local slave's FID
 906  * \retval      0 continue to check next item
 907  * \retval      -ve for failure
 908  */
 909 #ifdef HAVE_FILLDIR_USE_CTX
 910 static int osd_stripe_dir_filldir(struct dir_context *buf,
 911 #else
 912 static int osd_stripe_dir_filldir(void *buf,
 913 #endif
 914                                   const char *name, int namelen,
 915                                   loff_t offset, __u64 ino, unsigned int d_type)
 916 {
 917         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 918         struct osd_thread_info *oti = oclb->oclb_info;
 919         struct lu_fid *fid = &oti->oti_fid3;
 920         struct osd_inode_id *id = &oti->oti_id3;
 921         struct osd_device *dev = oclb->oclb_dev;
 922         struct inode *inode;
 923
 924         oclb->oclb_items++;
 925
 926         if (name[0] == '.')
 927                 return 0;
 928
 929         fid_zero(fid);
 930         sscanf(name + 1, SFID, RFID(fid));
 931         if (!fid_is_sane(fid))
 932                 return 0;
 933
 934         if (osd_remote_fid(oti->oti_env, dev, fid))
 935                 return 0;
 936
 937         osd_id_gen(id, ino, OSD_OII_NOGEN);
 938         inode = osd_iget(oti, dev, id);
 939         if (IS_ERR(inode))
 940                 return PTR_ERR(inode);
 941
 942         iput(inode);
 943         osd_add_oi_cache(oti, dev, id, fid);
 944         osd_oii_insert(dev, fid, id, true);
 945         oclb->oclb_found = true;
 946
 947         return 1;
 948 }
 949
 950 /*
 951  * When lookup item under striped directory, we need to locate the master
 952  * MDT-object of the striped directory firstly, then the client will send
 953  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 954  * and the item's name. If the system is restored from MDT file level backup,
 955  * then before the OI scrub completely built the OI files, the OI mappings of
 956  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 957  * not a problem for the master MDT-object. Because when locate the master
 958  * MDT-object, we will do name based lookup (for the striped directory itself)
 959  * firstly, during such process we can setup the correct OI mapping for the
 960  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 961  * the client will not trigger name based lookup on the MDT to locate the slave
 962  * MDT-object before locating item under the striped directory, then when
 963  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 964  * is invalid and does not know what the right OI mapping is, then the MDT has
 965  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 966  * the OI file, related OI mapping is unknown yet, please try again later. And
 967  * then client will re-try the RPC again and again until related OI mapping has
 968  * been updated. That is quite inefficient.
 969  *
 970  * To resolve above trouble, we will handle it as the following two cases:
 971  *
 972  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 973  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 974  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 975  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 976  *    directly. Please check osd_fid_lookup().
 977  *
 978  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 979  *    Under such case, during lookup the master MDT-object, we will lookup the
 980  *    slave MDT-object via readdir against the master MDT-object, because the
 981  *    slave MDT-objects information are stored as sub-directories with the name
 982  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 983  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 984  *    the correct OI mapping for the slave MDT-object.
 985  */
 986 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 987                          struct inode *inode)
 988 {
 989         struct lu_buf *buf = &oti->oti_big_buf;
 990         struct file *filp;
 991         struct lmv_mds_md_v1 *lmv1;
 992         struct osd_check_lmv_buf oclb = {
 993                 .ctx.actor = osd_stripe_dir_filldir,
 994                 .oclb_info = oti,
 995                 .oclb_dev = dev,
 996                 .oclb_found = false,
 997         };
 998         int rc = 0;
 999
1000         ENTRY;
1001         /* We should use the VFS layer to create a real dentry. */
1002         oti->oti_obj_dentry.d_inode = inode;
1003         oti->oti_obj_dentry.d_sb = inode->i_sb;
1004
1005         filp = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
1006                                  inode->i_fop);
1007         if (IS_ERR(filp))
1008                 RETURN(-ENOMEM);
1009
1010         filp->f_mode |= FMODE_64BITHASH;
1011         filp->f_pos = 0;
1012         ihold(inode);
1013 again:
1014         rc = __osd_xattr_get(inode, filp->f_path.dentry, XATTR_NAME_LMV,
1015                              buf->lb_buf, buf->lb_len);
1016         if (rc == -ERANGE) {
1017                 rc = __osd_xattr_get(inode, filp->f_path.dentry,
1018                                      XATTR_NAME_LMV, NULL, 0);
1019                 if (rc > 0) {
1020                         lu_buf_realloc(buf, rc);
1021                         if (buf->lb_buf == NULL)
1022                                 GOTO(out, rc = -ENOMEM);
1023
1024                         goto again;
1025                 }
1026         }
1027
1028         if (unlikely(rc == 0 || rc == -ENODATA))
1029                 GOTO(out, rc = 0);
1030
1031         if (rc < 0)
1032                 GOTO(out, rc);
1033
1034         if (unlikely(buf->lb_buf == NULL)) {
1035                 lu_buf_realloc(buf, rc);
1036                 if (buf->lb_buf == NULL)
1037                         GOTO(out, rc = -ENOMEM);
1038
1039                 goto again;
1040         }
1041
1042         lmv1 = buf->lb_buf;
1043         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1044                 GOTO(out, rc = 0);
1045
1046         do {
1047                 oclb.oclb_items = 0;
1048                 rc = iterate_dir(filp, &oclb.ctx);
1049         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1050                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1051 out:
1052         fput(filp);
1053         if (rc < 0)
1054                 CDEBUG(D_LFSCK,
1055                        "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",
1056                        osd_ino2name(inode), inode->i_ino, inode->i_generation,
1057                        rc);
1058         else
1059                 rc = 0;
1060
1061         RETURN(rc);
1062 }
1063
1064 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1065                           const struct lu_fid *fid,
1066                           const struct lu_object_conf *conf)
1067 {
1068         struct osd_thread_info *info;
1069         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1070         struct osd_device *dev;
1071         struct osd_idmap_cache *oic;
1072         struct osd_inode_id *id;
1073         struct inode *inode = NULL;
1074         struct lustre_scrub *scrub;
1075         struct scrub_file *sf;
1076         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1077         __u32 saved_ino;
1078         __u32 saved_gen;
1079         int result = 0;
1080         int rc1 = 0;
1081         bool remote = false;
1082         bool trusted = true;
1083         bool updated = false;
1084         bool checked = false;
1085
1086         ENTRY;
1087
1088         LINVRNT(osd_invariant(obj));
1089         LASSERT(obj->oo_inode == NULL);
1090
1091         if (fid_is_sane(fid) == 0) {
1092                 CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
1093                        PFID(fid));
1094                 dump_stack();
1095                 RETURN(-EINVAL);
1096         }
1097
1098         dev = osd_dev(ldev);
1099         scrub = &dev->od_scrub.os_scrub;
1100         sf = &scrub->os_file;
1101         info = osd_oti_get(env);
1102         LASSERT(info);
1103         oic = &info->oti_cache;
1104
1105         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1106                 RETURN(-ENOENT);
1107
1108         /*
1109          * For the object is created as locking anchor, or for the object to
1110          * be created on disk. No need to osd_oi_lookup() at here because FID
1111          * shouldn't never be re-used, if it's really a duplicate FID from
1112          * unexpected reason, we should be able to detect it later by calling
1113          * do_create->osd_oi_insert().
1114          */
1115         if (conf && conf->loc_flags & LOC_F_NEW)
1116                 GOTO(out, result = 0);
1117
1118         /* Search order: 1. per-thread cache. */
1119         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1120                 id = &oic->oic_lid;
1121                 goto iget;
1122         }
1123
1124         id = &info->oti_id;
1125         memset(id, 0, sizeof(struct osd_inode_id));
1126         if (!list_empty(&scrub->os_inconsistent_items)) {
1127                 /* Search order: 2. OI scrub pending list. */
1128                 result = osd_oii_lookup(dev, fid, id);
1129                 if (!result)
1130                         goto iget;
1131         }
1132
1133         /*
1134          * The OI mapping in the OI file can be updated by the OI scrub
1135          * when we locate the inode via FID. So it may be not trustable.
1136          */
1137         trusted = false;
1138
1139         /* Search order: 3. OI files. */
1140         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1141         if (result == -ENOENT) {
1142                 if (!fid_is_norm(fid) ||
1143                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1144                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1145                                       sf->sf_oi_bitmap))
1146                         GOTO(out, result = 0);
1147
1148                 goto trigger;
1149         }
1150
1151         /* -ESTALE is returned if inode of OST object doesn't exist */
1152         if (result == -ESTALE &&
1153             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1154                 GOTO(out, result = 0);
1155         }
1156
1157         if (result)
1158                 GOTO(out, result);
1159
1160 iget:
1161         obj->oo_inode = NULL;
1162         /* for later passes through checks, not true on first pass */
1163         if (!IS_ERR_OR_NULL(inode))
1164                 iput(inode);
1165
1166         inode = osd_iget_check(info, dev, fid, id, trusted);
1167         if (!IS_ERR(inode)) {
1168                 obj->oo_inode = inode;
1169                 result = 0;
1170                 if (remote)
1171                         goto trigger;
1172
1173                 goto check_lma;
1174         }
1175
1176         result = PTR_ERR(inode);
1177         if (result == -ENOENT || result == -ESTALE)
1178                 GOTO(out, result = 0);
1179
1180         if (result != -EREMCHG)
1181                 GOTO(out, result);
1182
1183 trigger:
1184         /*
1185          * We still have chance to get the valid inode: for the
1186          * object which is referenced by remote name entry, the
1187          * object on the local MDT will be linked under the dir
1188          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1189          *
1190          * We do not know whether the object for the given FID
1191          * is referenced by some remote name entry or not, and
1192          * especially for DNE II, a multiple-linked object may
1193          * have many name entries reside on many MDTs.
1194          *
1195          * To simplify the operation, OSD will not distinguish
1196          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1197          * only happened for the RPC from other MDT during the
1198          * OI scrub, or for the client side RPC with FID only,
1199          * such as FID to path, or from old connected client.
1200          */
1201         if (!remote) {
1202                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1203                 if (!rc1) {
1204                         remote = true;
1205                         trusted = true;
1206                         flags |= SS_AUTO_PARTIAL;
1207                         flags &= ~SS_AUTO_FULL;
1208                         goto iget;
1209                 }
1210         }
1211
1212         if (scrub->os_running) {
1213                 if (scrub->os_partial_scan && !scrub->os_in_join)
1214                         goto join;
1215
1216                 if (IS_ERR_OR_NULL(inode) || result) {
1217                         osd_oii_insert(dev, fid, id, result == -ENOENT);
1218                         GOTO(out, result = -EINPROGRESS);
1219                 }
1220
1221                 LASSERT(remote);
1222                 LASSERT(obj->oo_inode == inode);
1223
1224                 osd_oii_insert(dev, fid, id, true);
1225                 goto found;
1226         }
1227
1228         if (dev->od_scrub.os_scrub.os_auto_scrub_interval == AS_NEVER) {
1229                 if (!remote)
1230                         GOTO(out, result = -EREMCHG);
1231
1232                 LASSERT(!result);
1233                 LASSERT(obj->oo_inode == inode);
1234
1235                 osd_add_oi_cache(info, dev, id, fid);
1236                 goto found;
1237         }
1238
1239 join:
1240         rc1 = osd_scrub_start(env, dev, flags);
1241         CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
1242                      "%s: trigger OI scrub by RPC for "DFID"/%u with flags %#x: rc = %d\n",
1243                      osd_name(dev), PFID(fid), id->oii_ino, flags, rc1);
1244         if (rc1 && rc1 != -EALREADY)
1245                 GOTO(out, result = -EREMCHG);
1246
1247         if (IS_ERR_OR_NULL(inode) || result) {
1248                 osd_oii_insert(dev, fid, id, result == -ENOENT);
1249                 GOTO(out, result = -EINPROGRESS);
1250         }
1251
1252         LASSERT(remote);
1253         LASSERT(obj->oo_inode == inode);
1254
1255         osd_oii_insert(dev, fid, id, true);
1256         goto found;
1257
1258 check_lma:
1259         checked = true;
1260         if (unlikely(obj->oo_header))
1261                 goto found;
1262
1263         result = osd_check_lma(env, obj);
1264         if (!result)
1265                 goto found;
1266
1267         LASSERTF(id->oii_ino == inode->i_ino &&
1268                  id->oii_gen == inode->i_generation,
1269                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1270                  PFID(fid), id->oii_ino, id->oii_gen,
1271                  inode->i_ino, inode->i_generation);
1272
1273         saved_ino = inode->i_ino;
1274         saved_gen = inode->i_generation;
1275
1276         if (unlikely(result == -ENODATA)) {
1277                 /*
1278                  * If the OI scrub updated the OI mapping by race, it
1279                  * must be valid. Trust the inode that has no LMA EA.
1280                  */
1281                 if (updated)
1282                         goto found;
1283
1284                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1285                 if (!result) {
1286                         /*
1287                          * The OI mapping is still there, the inode is still
1288                          * valid. It is just becaues the inode has no LMA EA.
1289                          */
1290                         if (saved_ino == id->oii_ino &&
1291                             saved_gen == id->oii_gen)
1292                                 goto found;
1293
1294                         /*
1295                          * It is the OI scrub updated the OI mapping by race.
1296                          * The new OI mapping must be valid.
1297                          */
1298                         trusted = true;
1299                         updated = true;
1300                         goto iget;
1301                 }
1302
1303                 /*
1304                  * "result == -ENOENT" means that the OI mappinghas been
1305                  * removed by race, so the inode belongs to other object.
1306                  *
1307                  * Others error can be returned  directly.
1308                  */
1309                 if (result == -ENOENT) {
1310                         LASSERT(trusted);
1311
1312                         obj->oo_inode = NULL;
1313                         result = 0;
1314                 }
1315         }
1316
1317         if (result != -EREMCHG)
1318                 GOTO(out, result);
1319
1320         LASSERT(!updated);
1321
1322         /*
1323          * if two OST objects map to the same inode, and inode mode is
1324          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1325          * reserved by precreate, and not written yet, in this case, don't
1326          * set inode for the object whose FID mismatch, so that it can create
1327          * inode and not block precreate.
1328          */
1329         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1330             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1331                 obj->oo_inode = NULL;
1332                 GOTO(out, result = 0);
1333         }
1334
1335         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1336         /*
1337          * "result == -ENOENT" means the cached OI mapping has been removed
1338          * from the OI file by race, above inode belongs to other object.
1339          */
1340         if (result == -ENOENT) {
1341                 LASSERT(trusted);
1342
1343                 obj->oo_inode = NULL;
1344                 GOTO(out, result = 0);
1345         }
1346
1347         if (result)
1348                 GOTO(out, result);
1349
1350         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1351                 result = -EREMCHG;
1352                 osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
1353                                           true, 0, NULL);
1354                 goto trigger;
1355         }
1356
1357         /*
1358          * It is the OI scrub updated the OI mapping by race.
1359          * The new OI mapping must be valid.
1360          */
1361         trusted = true;
1362         updated = true;
1363         goto iget;
1364
1365 found:
1366         if (!checked) {
1367                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1368                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1369
1370                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1371                 if (!result) {
1372                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1373                             dev->od_is_ost)
1374                                 obj->oo_pfid_in_lma = 1;
1375                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1376                             !dev->od_is_ost)
1377                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1378                 } else if (result != -ENODATA) {
1379                         GOTO(out, result);
1380                 }
1381         }
1382
1383         obj->oo_compat_dot_created = 1;
1384         obj->oo_compat_dotdot_created = 1;
1385
1386         if (S_ISDIR(inode->i_mode) &&
1387             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1388                 osd_check_lmv(info, dev, inode);
1389
1390         result = osd_attach_jinode(inode);
1391         if (result)
1392                 GOTO(out, result);
1393
1394         if (!ldiskfs_pdo)
1395                 GOTO(out, result = 0);
1396
1397         LASSERT(!obj->oo_hl_head);
1398         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1399
1400         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1401
1402 out:
1403         if (result || !obj->oo_inode) {
1404                 if (!IS_ERR_OR_NULL(inode))
1405                         iput(inode);
1406
1407                 obj->oo_inode = NULL;
1408                 if (trusted)
1409                         fid_zero(&oic->oic_fid);
1410         }
1411
1412         LINVRNT(osd_invariant(obj));
1413         return result;
1414 }
1415
1416 /*
1417  * Concurrency: shouldn't matter.
1418  */
1419 static void osd_object_init0(struct osd_object *obj)
1420 {
1421         LASSERT(obj->oo_inode != NULL);
1422         obj->oo_dt.do_body_ops = &osd_body_ops;
1423         obj->oo_dt.do_lu.lo_header->loh_attr |=
1424                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1425 }
1426
1427 /*
1428  * Concurrency: no concurrent access is possible that early in object
1429  * life-cycle.
1430  */
1431 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1432                            const struct lu_object_conf *conf)
1433 {
1434         struct osd_object *obj = osd_obj(l);
1435         int result;
1436
1437         LINVRNT(osd_invariant(obj));
1438
1439         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1440             cfs_fail_val == 2) {
1441                 struct osd_thread_info *info = osd_oti_get(env);
1442                 struct osd_idmap_cache *oic = &info->oti_cache;
1443                 /* invalidate thread cache */
1444                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1445         }
1446         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1447                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1448                 l->lo_header->loh_attr |= LOHA_EXISTS;
1449                 return 0;
1450         }
1451
1452         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1453         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1454         if (result == 0 && obj->oo_inode != NULL) {
1455                 struct osd_thread_info *oti = osd_oti_get(env);
1456                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1457
1458                 osd_object_init0(obj);
1459                 if (unlikely(obj->oo_header))
1460                         return 0;
1461
1462                 result = osd_get_lma(oti, obj->oo_inode,
1463                                      &oti->oti_obj_dentry, loa);
1464                 if (!result) {
1465                         /*
1466                          * Convert LMAI flags to lustre LMA flags
1467                          * and cache it to oo_lma_flags
1468                          */
1469                         obj->oo_lma_flags =
1470                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1471                 } else if (result == -ENODATA) {
1472                         result = 0;
1473                 }
1474         }
1475         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1476
1477         LINVRNT(osd_invariant(obj));
1478         return result;
1479 }
1480
1481 /*
1482  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1483  * The left part is for value, binary mode.
1484  */
1485 struct osd_xattr_entry {
1486         struct list_head        oxe_list;
1487         size_t                  oxe_len;
1488         size_t                  oxe_namelen;
1489         bool                    oxe_exist;
1490         struct rcu_head         oxe_rcu;
1491         char                    oxe_buf[0];
1492 };
1493
1494 static int osd_oxc_get(struct osd_object *obj, const char *name,
1495                        struct lu_buf *buf)
1496 {
1497         struct osd_xattr_entry *tmp;
1498         struct osd_xattr_entry *oxe = NULL;
1499         size_t namelen = strlen(name);
1500         int rc;
1501
1502         rcu_read_lock();
1503         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1504                 if (namelen == tmp->oxe_namelen &&
1505                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1506                         oxe = tmp;
1507                         break;
1508                 }
1509         }
1510
1511         if (oxe == NULL)
1512                 GOTO(out, rc = -ENOENT);
1513
1514         if (!oxe->oxe_exist)
1515                 GOTO(out, rc = -ENODATA);
1516
1517         /* vallen */
1518         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1519         LASSERT(rc > 0);
1520
1521         if (buf->lb_buf == NULL)
1522                 GOTO(out, rc);
1523
1524         if (buf->lb_len < rc)
1525                 GOTO(out, rc = -ERANGE);
1526
1527         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1528 out:
1529         rcu_read_unlock();
1530
1531         return rc;
1532 }
1533
1534 static void osd_oxc_free(struct rcu_head *head)
1535 {
1536         struct osd_xattr_entry *oxe;
1537
1538         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1539         OBD_FREE(oxe, oxe->oxe_len);
1540 }
1541
1542 static void osd_oxc_add(struct osd_object *obj, const char *name,
1543                         const char *buf, int buflen)
1544 {
1545         struct osd_xattr_entry *oxe;
1546         struct osd_xattr_entry *old = NULL;
1547         struct osd_xattr_entry *tmp;
1548         size_t namelen = strlen(name);
1549         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1550
1551         OBD_ALLOC(oxe, len);
1552         if (oxe == NULL)
1553                 return;
1554
1555         INIT_LIST_HEAD(&oxe->oxe_list);
1556         oxe->oxe_len = len;
1557         oxe->oxe_namelen = namelen;
1558         memcpy(oxe->oxe_buf, name, namelen);
1559         if (buflen > 0) {
1560                 LASSERT(buf != NULL);
1561                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1562                 oxe->oxe_exist = true;
1563         } else {
1564                 oxe->oxe_exist = false;
1565         }
1566
1567         /* this should be rarely called, just remove old and add new */
1568         spin_lock(&obj->oo_guard);
1569         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1570                 if (namelen == tmp->oxe_namelen &&
1571                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1572                         old = tmp;
1573                         break;
1574                 }
1575         }
1576         if (old != NULL) {
1577                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1578                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1579         } else {
1580                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1581         }
1582         spin_unlock(&obj->oo_guard);
1583 }
1584
1585 static void osd_oxc_del(struct osd_object *obj, const char *name)
1586 {
1587         struct osd_xattr_entry *oxe;
1588         size_t namelen = strlen(name);
1589
1590         spin_lock(&obj->oo_guard);
1591         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1592                 if (namelen == oxe->oxe_namelen &&
1593                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1594                         list_del_rcu(&oxe->oxe_list);
1595                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1596                         break;
1597                 }
1598         }
1599         spin_unlock(&obj->oo_guard);
1600 }
1601
1602 static void osd_oxc_fini(struct osd_object *obj)
1603 {
1604         struct osd_xattr_entry *oxe, *next;
1605
1606         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1607                 list_del(&oxe->oxe_list);
1608                 OBD_FREE(oxe, oxe->oxe_len);
1609         }
1610 }
1611
1612 /*
1613  * Concurrency: no concurrent access is possible that late in object
1614  * life-cycle.
1615  */
1616 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1617 {
1618         struct osd_object *obj = osd_obj(l);
1619         struct lu_object_header *h = obj->oo_header;
1620
1621         LINVRNT(osd_invariant(obj));
1622
1623         osd_oxc_fini(obj);
1624         dt_object_fini(&obj->oo_dt);
1625         if (obj->oo_hl_head != NULL)
1626                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1627         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1628         OBD_FREE_PTR(obj);
1629         if (unlikely(h))
1630                 lu_object_header_free(h);
1631 }
1632
1633 /*
1634  * Concurrency: no concurrent access is possible that late in object
1635  * life-cycle.
1636  */
1637 static void osd_index_fini(struct osd_object *o)
1638 {
1639         struct iam_container *bag;
1640
1641         if (o->oo_dir != NULL) {
1642                 bag = &o->oo_dir->od_container;
1643                 if (o->oo_inode != NULL) {
1644                         if (bag->ic_object == o->oo_inode)
1645                                 iam_container_fini(bag);
1646                 }
1647                 OBD_FREE_PTR(o->oo_dir);
1648                 o->oo_dir = NULL;
1649         }
1650 }
1651
1652 enum {
1653         OSD_TXN_OI_DELETE_CREDITS    = 20,
1654         OSD_TXN_INODE_DELETE_CREDITS = 20
1655 };
1656
1657 /*
1658  * Journal
1659  */
1660
1661 #if OSD_THANDLE_STATS
1662 /**
1663  * Set time when the handle is allocated
1664  */
1665 static void osd_th_alloced(struct osd_thandle *oth)
1666 {
1667         oth->oth_alloced = ktime_get();
1668 }
1669
1670 /**
1671  * Set time when the handle started
1672  */
1673 static void osd_th_started(struct osd_thandle *oth)
1674 {
1675         oth->oth_started = ktime_get();
1676 }
1677
1678 /**
1679  * Check whether the we deal with this handle for too long.
1680  */
1681 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1682                                 ktime_t alloced, ktime_t started,
1683                                 ktime_t closed)
1684 {
1685         ktime_t now = ktime_get();
1686
1687         LASSERT(dev != NULL);
1688
1689         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1690                             ktime_us_delta(started, alloced));
1691         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1692                             ktime_us_delta(closed, started));
1693         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1694                             ktime_us_delta(now, closed));
1695
1696         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1697                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1698                                 oth, now, alloced, started, closed);
1699                 libcfs_debug_dumpstack(NULL);
1700         }
1701 }
1702
1703 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1704 {                                                                       \
1705         ktime_t __closed = ktime_get();                                 \
1706         ktime_t __alloced = oth->oth_alloced;                           \
1707         ktime_t __started = oth->oth_started;                           \
1708                                                                         \
1709         expr;                                                           \
1710         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1711 }
1712
1713 #else /* OSD_THANDLE_STATS */
1714
1715 #define osd_th_alloced(h)                  do {} while(0)
1716 #define osd_th_started(h)                  do {} while(0)
1717 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1718
1719 #endif /* OSD_THANDLE_STATS */
1720
1721 /*
1722  * Concurrency: doesn't access mutable data.
1723  */
1724 static int osd_param_is_not_sane(const struct osd_device *dev,
1725                                  const struct thandle *th)
1726 {
1727         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1728
1729         return oh->ot_credits > osd_transaction_size(dev);
1730 }
1731
1732 /*
1733  * Concurrency: shouldn't matter.
1734  */
1735 static void osd_trans_commit_cb(struct super_block *sb,
1736                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1737 {
1738         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1739         struct thandle *th = &oh->ot_super;
1740         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1741         struct osd_device *osd = osd_dev(lud);
1742         struct dt_txn_commit_cb *dcb, *tmp;
1743
1744         LASSERT(oh->ot_handle == NULL);
1745
1746         if (error)
1747                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1748
1749         /* call per-transaction callbacks if any */
1750         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1751                                  dcb_linkage) {
1752                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1753                          "commit callback entry: magic=%x name='%s'\n",
1754                          dcb->dcb_magic, dcb->dcb_name);
1755                 list_del_init(&dcb->dcb_linkage);
1756                 dcb->dcb_func(NULL, th, dcb, error);
1757         }
1758
1759         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1760         if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
1761                 wake_up(&osd->od_commit_cb_done);
1762         th->th_dev = NULL;
1763
1764         OBD_FREE_PTR(oh);
1765 }
1766
1767 static struct thandle *osd_trans_create(const struct lu_env *env,
1768                                         struct dt_device *d)
1769 {
1770         struct osd_thread_info *oti = osd_oti_get(env);
1771         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1772         struct osd_thandle *oh;
1773         struct thandle *th;
1774
1775         ENTRY;
1776
1777         if (d->dd_rdonly) {
1778                 CERROR("%s: someone try to start transaction under "
1779                        "readonly mode, should be disabled.\n",
1780                        osd_name(osd_dt_dev(d)));
1781                 dump_stack();
1782                 RETURN(ERR_PTR(-EROFS));
1783         }
1784
1785         /* on pending IO in this thread should left from prev. request */
1786         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1787
1788         sb_start_write(osd_sb(osd_dt_dev(d)));
1789
1790         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1791         if (!oh) {
1792                 sb_end_write(osd_sb(osd_dt_dev(d)));
1793                 RETURN(ERR_PTR(-ENOMEM));
1794         }
1795
1796         oh->ot_quota_trans = &oti->oti_quota_trans;
1797         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1798         th = &oh->ot_super;
1799         th->th_dev = d;
1800         th->th_result = 0;
1801         oh->ot_credits = 0;
1802         oh->oh_declared_ext = 0;
1803         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1804         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1805         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1806         osd_th_alloced(oh);
1807
1808         memset(oti->oti_declare_ops, 0,
1809                sizeof(oti->oti_declare_ops));
1810         memset(oti->oti_declare_ops_cred, 0,
1811                sizeof(oti->oti_declare_ops_cred));
1812         memset(oti->oti_declare_ops_used, 0,
1813                sizeof(oti->oti_declare_ops_used));
1814
1815         oti->oti_ins_cache_depth++;
1816
1817         RETURN(th);
1818 }
1819
1820 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1821 {
1822         struct osd_thread_info *oti = osd_oti_get(env);
1823         struct osd_thandle *oh;
1824
1825         oh = container_of(th, struct osd_thandle, ot_super);
1826         LASSERT(oh != NULL);
1827
1828         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1829               oti->oti_declare_ops[OSD_OT_CREATE],
1830               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1831               oti->oti_declare_ops_used[OSD_OT_CREATE],
1832               oti->oti_declare_ops[OSD_OT_DESTROY],
1833               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1834               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1835         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1836               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1837               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1838               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1839               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1840               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1841               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1842         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1843               oti->oti_declare_ops[OSD_OT_WRITE],
1844               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1845               oti->oti_declare_ops_used[OSD_OT_WRITE],
1846               oti->oti_declare_ops[OSD_OT_PUNCH],
1847               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1848               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1849               oti->oti_declare_ops[OSD_OT_QUOTA],
1850               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1851               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1852         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1853               oti->oti_declare_ops[OSD_OT_INSERT],
1854               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1855               oti->oti_declare_ops_used[OSD_OT_INSERT],
1856               oti->oti_declare_ops[OSD_OT_DELETE],
1857               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1858               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1859         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1860               oti->oti_declare_ops[OSD_OT_REF_ADD],
1861               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1862               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1863               oti->oti_declare_ops[OSD_OT_REF_DEL],
1864               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1865               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1866 }
1867
1868 /*
1869  * Concurrency: shouldn't matter.
1870  */
1871 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1872                            struct thandle *th)
1873 {
1874         struct osd_thread_info *oti = osd_oti_get(env);
1875         struct osd_device *dev = osd_dt_dev(d);
1876         handle_t *jh;
1877         struct osd_thandle *oh;
1878         int rc;
1879
1880         ENTRY;
1881
1882         LASSERT(current->journal_info == NULL);
1883
1884         oh = container_of(th, struct osd_thandle, ot_super);
1885         LASSERT(oh != NULL);
1886         LASSERT(oh->ot_handle == NULL);
1887
1888         rc = dt_txn_hook_start(env, d, th);
1889         if (rc != 0)
1890                 GOTO(out, rc);
1891
1892         if (unlikely(osd_param_is_not_sane(dev, th))) {
1893                 static unsigned long last_printed;
1894                 static int last_credits;
1895
1896                 /*
1897                  * don't make noise on a tiny testing systems
1898                  * actual credits misuse will be caught anyway
1899                  */
1900                 if (last_credits != oh->ot_credits &&
1901                     time_after(jiffies, last_printed +
1902                                cfs_time_seconds(60)) &&
1903                     osd_transaction_size(dev) > 512) {
1904                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1905                               oh->ot_credits, osd_transaction_size(dev));
1906                         osd_trans_dump_creds(env, th);
1907                         libcfs_debug_dumpstack(NULL);
1908                         last_credits = oh->ot_credits;
1909                         last_printed = jiffies;
1910                 }
1911                 /*
1912                  * XXX Limit the credits to 'max_transaction_buffers', and
1913                  *     let the underlying filesystem to catch the error if
1914                  *     we really need so many credits.
1915                  *
1916                  *     This should be removed when we can calculate the
1917                  *     credits precisely.
1918                  */
1919                 oh->ot_credits = osd_transaction_size(dev);
1920         } else if (ldiskfs_track_declares_assert != 0) {
1921                 /*
1922                  * reserve few credits to prevent an assertion in JBD
1923                  * our debugging mechanism will be able to detected
1924                  * overuse. this can help to debug single-update
1925                  * transactions
1926                  */
1927                 oh->ot_credits += 10;
1928                 if (unlikely(osd_param_is_not_sane(dev, th)))
1929                         oh->ot_credits = osd_transaction_size(dev);
1930         }
1931
1932         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1933                 GOTO(out, rc = -EIO);
1934
1935         /*
1936          * XXX temporary stuff. Some abstraction layer should
1937          * be used.
1938          */
1939         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1940         osd_th_started(oh);
1941         if (!IS_ERR(jh)) {
1942                 oh->ot_handle = jh;
1943                 LASSERT(oti->oti_txns == 0);
1944
1945                 atomic_inc(&dev->od_commit_cb_in_flight);
1946                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1947                               "osd-tx", th);
1948                 oti->oti_txns++;
1949                 rc = 0;
1950         } else {
1951                 rc = PTR_ERR(jh);
1952         }
1953 out:
1954         RETURN(rc);
1955 }
1956
1957 static int osd_seq_exists(const struct lu_env *env,
1958                           struct osd_device *osd, u64 seq)
1959 {
1960         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1961         struct seq_server_site *ss = osd_seq_site(osd);
1962         int rc;
1963
1964         ENTRY;
1965
1966         LASSERT(ss != NULL);
1967         LASSERT(ss->ss_server_fld != NULL);
1968
1969         rc = osd_fld_lookup(env, osd, seq, range);
1970         if (rc != 0) {
1971                 if (rc != -ENOENT)
1972                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1973                                osd_name(osd), seq, rc);
1974                 RETURN(0);
1975         }
1976
1977         RETURN(ss->ss_node_id == range->lsr_index);
1978 }
1979
1980 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1981 {
1982         struct dt_txn_commit_cb *dcb;
1983         struct dt_txn_commit_cb *tmp;
1984
1985         /* call per-transaction stop callbacks if any */
1986         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1987                                  dcb_linkage) {
1988                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1989                          "commit callback entry: magic=%x name='%s'\n",
1990                          dcb->dcb_magic, dcb->dcb_name);
1991                 list_del_init(&dcb->dcb_linkage);
1992                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1993         }
1994 }
1995
1996 /*
1997  * Concurrency: shouldn't matter.
1998  */
1999 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
2000                           struct thandle *th)
2001 {
2002         struct osd_thread_info *oti = osd_oti_get(env);
2003         struct osd_thandle *oh;
2004         struct osd_iobuf *iobuf = &oti->oti_iobuf;
2005         struct osd_device *osd = osd_dt_dev(th->th_dev);
2006         struct qsd_instance *qsd = osd_def_qsd(osd);
2007         struct lquota_trans *qtrans;
2008         LIST_HEAD(truncates);
2009         int rc = 0, remove_agents = 0;
2010
2011         ENTRY;
2012
2013         oh = container_of(th, struct osd_thandle, ot_super);
2014
2015         remove_agents = oh->ot_remove_agents;
2016
2017         qtrans = oh->ot_quota_trans;
2018         oh->ot_quota_trans = NULL;
2019
2020         /* move locks to local list, stop tx, execute truncates */
2021         list_splice(&oh->ot_trunc_locks, &truncates);
2022
2023         if (oh->ot_handle != NULL) {
2024                 int rc2;
2025
2026                 handle_t *hdl = oh->ot_handle;
2027
2028                 /*
2029                  * add commit callback
2030                  * notice we don't do this in osd_trans_start()
2031                  * as underlying transaction can change during truncate
2032                  */
2033                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2034                                              &oh->ot_jcb);
2035
2036                 LASSERT(oti->oti_txns == 1);
2037                 oti->oti_txns--;
2038
2039                 rc = dt_txn_hook_stop(env, th);
2040                 if (rc != 0)
2041                         CERROR("%s: failed in transaction hook: rc = %d\n",
2042                                osd_name(osd), rc);
2043
2044                 osd_trans_stop_cb(oh, rc);
2045                 /* hook functions might modify th_sync */
2046                 hdl->h_sync = th->th_sync;
2047
2048                 oh->ot_handle = NULL;
2049                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2050                 if (rc2 != 0)
2051                         CERROR("%s: failed to stop transaction: rc = %d\n",
2052                                osd_name(osd), rc2);
2053                 if (!rc)
2054                         rc = rc2;
2055
2056                 /* We preserve the origin behavior of ignoring any
2057                  * failures with the underlying punch / truncate
2058                  * operation. We do record for debugging if an error
2059                  * does occur in the lctl dk logs.
2060                  */
2061                 rc2 = osd_process_truncates(env, &truncates);
2062                 if (rc2 != 0)
2063                         CERROR("%s: failed truncate process: rc = %d\n",
2064                                osd_name(osd), rc2);
2065         } else {
2066                 osd_trans_stop_cb(oh, th->th_result);
2067                 OBD_FREE_PTR(oh);
2068         }
2069
2070         osd_trunc_unlock_all(env, &truncates);
2071
2072         /* inform the quota slave device that the transaction is stopping */
2073         qsd_op_end(env, qsd, qtrans);
2074
2075         /*
2076          * as we want IO to journal and data IO be concurrent, we don't block
2077          * awaiting data IO completion in osd_do_bio(), instead we wait here
2078          * once transaction is submitted to the journal. all reqular requests
2079          * don't do direct IO (except read/write), thus this wait_event becomes
2080          * no-op for them.
2081          *
2082          * IMPORTANT: we have to wait till any IO submited by the thread is
2083          * completed otherwise iobuf may be corrupted by different request
2084          */
2085         wait_event(iobuf->dr_wait,
2086                        atomic_read(&iobuf->dr_numreqs) == 0);
2087         osd_fini_iobuf(osd, iobuf);
2088         if (!rc)
2089                 rc = iobuf->dr_error;
2090
2091         if (unlikely(remove_agents != 0))
2092                 osd_process_scheduled_agent_removals(env, osd);
2093
2094         LASSERT(oti->oti_ins_cache_depth > 0);
2095         oti->oti_ins_cache_depth--;
2096         /* reset OI cache for safety */
2097         if (oti->oti_ins_cache_depth == 0)
2098                 oti->oti_ins_cache_used = 0;
2099
2100         sb_end_write(osd_sb(osd));
2101
2102         RETURN(rc);
2103 }
2104
2105 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2106 {
2107         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2108                                               ot_super);
2109
2110         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2111         LASSERT(&dcb->dcb_func != NULL);
2112         if (dcb->dcb_flags & DCB_TRANS_STOP)
2113                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2114         else
2115                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2116
2117         return 0;
2118 }
2119
2120 /*
2121  * Called just before object is freed. Releases all resources except for
2122  * object itself (that is released by osd_object_free()).
2123  *
2124  * Concurrency: no concurrent access is possible that late in object
2125  * life-cycle.
2126  */
2127 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2128 {
2129         struct osd_object *obj = osd_obj(l);
2130         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2131         struct inode *inode = obj->oo_inode;
2132         __u64 projid;
2133         qid_t uid;
2134         qid_t gid;
2135
2136         LINVRNT(osd_invariant(obj));
2137
2138         /*
2139          * If object is unlinked remove fid->ino mapping from object index.
2140          */
2141
2142         osd_index_fini(obj);
2143
2144         if (!inode)
2145                 return;
2146
2147         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2148                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2149
2150         uid = i_uid_read(inode);
2151         gid = i_gid_read(inode);
2152         projid = i_projid_read(inode);
2153
2154         obj->oo_inode = NULL;
2155         iput(inode);
2156
2157         /* do not rebalance quota if the caller needs to release memory
2158          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2159          * transaction and risk to deadlock - LU-12178 */
2160         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2161                 return;
2162
2163         if (!obj->oo_header && qsd) {
2164                 struct osd_thread_info *info = osd_oti_get(env);
2165                 struct lquota_id_info *qi = &info->oti_qi;
2166
2167                 /* Release granted quota to master if necessary */
2168                 qi->lqi_id.qid_uid = uid;
2169                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2170
2171                 qi->lqi_id.qid_uid = gid;
2172                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2173
2174                 qi->lqi_id.qid_uid = projid;
2175                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2176         }
2177 }
2178
2179 /*
2180  * Concurrency: ->loo_object_release() is called under site spin-lock.
2181  */
2182 static void osd_object_release(const struct lu_env *env,
2183                                struct lu_object *l)
2184 {
2185         struct osd_object *o = osd_obj(l);
2186
2187         /*
2188          * nobody should be releasing a non-destroyed object with nlink=0
2189          * the API allows this, but ldiskfs doesn't like and then report
2190          * this inode as deleted
2191          */
2192         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2193                   o->oo_inode->i_nlink == 0));
2194 }
2195
2196 /*
2197  * Concurrency: shouldn't matter.
2198  */
2199 static int osd_object_print(const struct lu_env *env, void *cookie,
2200                             lu_printer_t p, const struct lu_object *l)
2201 {
2202         struct osd_object *o = osd_obj(l);
2203         struct iam_descr *d;
2204
2205         if (o->oo_dir != NULL)
2206                 d = o->oo_dir->od_container.ic_descr;
2207         else
2208                 d = NULL;
2209         return (*p)(env, cookie,
2210                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2211                     o, o->oo_inode,
2212                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2213                     o->oo_inode ? o->oo_inode->i_generation : 0,
2214                     d ? d->id_ops->id_name : "plain");
2215 }
2216
2217 /*
2218  * Concurrency: shouldn't matter.
2219  */
2220 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2221                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2222 {
2223         struct osd_device *osd = osd_dt_dev(d);
2224         struct super_block *sb = osd_sb(osd);
2225         struct kstatfs *ksfs;
2226         __u64 reserved;
2227         int result = 0;
2228
2229         if (unlikely(osd->od_mnt == NULL))
2230                 return -EINPROGRESS;
2231
2232         /* osd_lproc.c call this without env, allocate ksfs for that case */
2233         if (unlikely(env == NULL)) {
2234                 OBD_ALLOC_PTR(ksfs);
2235                 if (ksfs == NULL)
2236                         return -ENOMEM;
2237         } else {
2238                 ksfs = &osd_oti_get(env)->oti_ksfs;
2239         }
2240
2241         result = sb->s_op->statfs(sb->s_root, ksfs);
2242         if (result)
2243                 goto out;
2244
2245         statfs_pack(sfs, ksfs);
2246         if (unlikely(sb->s_flags & SB_RDONLY))
2247                 sfs->os_state |= OS_STATFS_READONLY;
2248
2249         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2250
2251         if (ldiskfs_has_feature_extents(sb))
2252                 sfs->os_maxbytes = sb->s_maxbytes;
2253         else
2254                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2255
2256         /*
2257          * Reserve some space so to avoid fragmenting the filesystem too much.
2258          * Fragmentation not only impacts performance, but can also increase
2259          * metadata overhead significantly, causing grant calculation to be
2260          * wrong.
2261          *
2262          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2263          */
2264         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2265         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2266         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2267                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2268
2269         sfs->os_blocks -= reserved;
2270         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2271         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2272
2273 out:
2274         if (unlikely(env == NULL))
2275                 OBD_FREE_PTR(ksfs);
2276         return result;
2277 }
2278
2279 /**
2280  * Estimate space needed for file creations. We assume the largest filename
2281  * which is 2^64 - 1, hence a filename of 20 chars.
2282  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2283  */
2284 #ifdef __LDISKFS_DIR_REC_LEN
2285 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2286 #else
2287 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2288 #endif
2289
2290 /*
2291  * Concurrency: doesn't access mutable data.
2292  */
2293 static void osd_conf_get(const struct lu_env *env,
2294                          const struct dt_device *dev,
2295                          struct dt_device_param *param)
2296 {
2297         struct osd_device *d = osd_dt_dev(dev);
2298         struct super_block *sb = osd_sb(d);
2299         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2300         const char *name;
2301         int ea_overhead;
2302
2303         /*
2304          * XXX should be taken from not-yet-existing fs abstraction layer.
2305          */
2306         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2307         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2308         param->ddp_symlink_max  = sb->s_blocksize;
2309         param->ddp_mount_type   = LDD_MT_LDISKFS;
2310         if (ldiskfs_has_feature_extents(sb))
2311                 param->ddp_maxbytes = sb->s_maxbytes;
2312         else
2313                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2314         /*
2315          * inode are statically allocated, so per-inode space consumption
2316          * is the space consumed by the directory entry
2317          */
2318         param->ddp_inodespace     = PER_OBJ_USAGE;
2319         /*
2320          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2321          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2322          * maximum length to not under-count the actual number of extents
2323          * needed for writing a file if there are sub-optimal block allocations.
2324          */
2325         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2326         /* worst-case extent insertion metadata overhead */
2327         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2328         param->ddp_mntopts = 0;
2329         if (test_opt(sb, XATTR_USER))
2330                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2331         if (test_opt(sb, POSIX_ACL))
2332                 param->ddp_mntopts |= MNTOPT_ACL;
2333
2334         /*
2335          * LOD might calculate the max stripe count based on max_ea_size,
2336          * so we need take account in the overhead as well,
2337          * xattr_header + magic + xattr_entry_head
2338          */
2339         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2340                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2341
2342 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2343         if (ldiskfs_has_feature_ea_inode(sb))
2344                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2345                                                                 ea_overhead;
2346         else
2347 #endif
2348                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2349
2350         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2351                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2352
2353         /*
2354          * Preferred RPC size for efficient disk IO.  4MB shows good
2355          * all-around performance for ldiskfs, but use bigalloc chunk size
2356          * by default if larger.
2357          */
2358 #if defined(LDISKFS_CLUSTER_SIZE)
2359         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2360                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2361         else
2362 #endif
2363                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2364
2365         param->ddp_t10_cksum_type = 0;
2366         if (bi) {
2367                 unsigned short interval = blk_integrity_interval(bi);
2368                 name = blk_integrity_name(bi);
2369                 /*
2370                  * Expected values:
2371                  * T10-DIF-TYPE1-CRC
2372                  * T10-DIF-TYPE3-CRC
2373                  * T10-DIF-TYPE1-IP
2374                  * T10-DIF-TYPE3-IP
2375                  */
2376                 if (strncmp(name, "T10-DIF-TYPE",
2377                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2378                         /* also skip "1/3-" at end */
2379                         const int type_off = sizeof("T10-DIF-TYPE.");
2380                         char type_number = name[type_off - 2];
2381
2382                         if (interval != 512 && interval != 4096) {
2383                                 CERROR("%s: unsupported T10PI sector size %u\n",
2384                                        d->od_svname, interval);
2385                         } else if (type_number != '1' && type_number != '3') {
2386                                 CERROR("%s: unsupported T10PI type %s\n",
2387                                        d->od_svname, name);
2388                         } else if (strcmp(name + type_off, "CRC") == 0) {
2389                                 d->od_t10_type = type_number == '1' ?
2390                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2391                                 param->ddp_t10_cksum_type = interval == 512 ?
2392                                         OBD_CKSUM_T10CRC512 :
2393                                         OBD_CKSUM_T10CRC4K;
2394                         } else if (strcmp(name + type_off, "IP") == 0) {
2395                                 d->od_t10_type = type_number == '1' ?
2396                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2397                                 param->ddp_t10_cksum_type = interval == 512 ?
2398                                         OBD_CKSUM_T10IP512 :
2399                                         OBD_CKSUM_T10IP4K;
2400                         } else {
2401                                 CERROR("%s: unsupported checksum type of T10PI type '%s'\n",
2402                                        d->od_svname, name);
2403                         }
2404
2405                 } else {
2406                         CERROR("%s: unsupported T10PI type '%s'\n",
2407                                d->od_svname, name);
2408                 }
2409         }
2410
2411         param->ddp_has_lseek_data_hole = true;
2412 }
2413
2414 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2415 {
2416         return osd_sb(osd_dt_dev(d));
2417 }
2418
2419 /*
2420  * Concurrency: shouldn't matter.
2421  */
2422 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2423 {
2424         int rc;
2425         struct super_block *s = osd_sb(osd_dt_dev(d));
2426         ENTRY;
2427
2428         down_read(&s->s_umount);
2429         rc = s->s_op->sync_fs(s, 1);
2430         up_read(&s->s_umount);
2431
2432         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2433                rc);
2434
2435         return rc;
2436 }
2437
2438 /**
2439  * Start commit for OSD device.
2440  *
2441  * An implementation of dt_commit_async method for OSD device.
2442  * Asychronously starts underlayng fs sync and thereby a transaction
2443  * commit.
2444  *
2445  * \param env environment
2446  * \param d dt device
2447  *
2448  * \see dt_device_operations
2449  */
2450 static int osd_commit_async(const struct lu_env *env,
2451                             struct dt_device *d)
2452 {
2453         struct super_block *s = osd_sb(osd_dt_dev(d));
2454         int rc;
2455
2456         ENTRY;
2457
2458         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2459         down_read(&s->s_umount);
2460         rc = s->s_op->sync_fs(s, 0);
2461         up_read(&s->s_umount);
2462
2463         RETURN(rc);
2464 }
2465
2466 /*
2467  * Concurrency: shouldn't matter.
2468  */
2469 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2470 {
2471         struct super_block *sb = osd_sb(osd_dt_dev(d));
2472         struct block_device *dev = sb->s_bdev;
2473         int rc = -EOPNOTSUPP;
2474
2475         ENTRY;
2476
2477         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2478                osd_dt_dev(d)->od_svname, (long)dev, rc);
2479
2480         RETURN(rc);
2481 }
2482
2483 /**
2484  * Note: we do not count into QUOTA here.
2485  * If we mount with --data_journal we may need more.
2486  */
2487 const int osd_dto_credits_noquota[DTO_NR] = {
2488         /**
2489          * Insert.
2490          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2491          * SINGLEDATA_TRANS_BLOCKS(8)
2492          * XXX Note: maybe iam need more, since iam have more level than
2493          *           EXT3 htree.
2494          */
2495         [DTO_INDEX_INSERT]  = 16,
2496         /**
2497          * Delete
2498          * just modify a single entry, probably merge few within a block
2499          */
2500         [DTO_INDEX_DELETE]  = 1,
2501         /**
2502          * Used for OI scrub
2503          */
2504         [DTO_INDEX_UPDATE]  = 16,
2505         /**
2506          * 4(inode, inode bits, groups, GDT)
2507          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2508          */
2509         [DTO_OBJECT_CREATE] = 4,
2510         /**
2511          * 4(inode, inode bits, groups, GDT)
2512          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2513          */
2514         [DTO_OBJECT_DELETE] = 4,
2515         /**
2516          * Attr set credits (inode)
2517          */
2518         [DTO_ATTR_SET_BASE] = 1,
2519         /**
2520          * Xattr set. The same as xattr of EXT3.
2521          * DATA_TRANS_BLOCKS(14)
2522          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2523          * are also counted in. Do not know why?
2524          */
2525         [DTO_XATTR_SET]     = 14,
2526         /**
2527          * credits for inode change during write.
2528          */
2529         [DTO_WRITE_BASE]    = 3,
2530         /**
2531          * credits for single block write.
2532          */
2533         [DTO_WRITE_BLOCK]   = 14,
2534         /**
2535          * Attr set credits for chown.
2536          * This is extra credits for setattr, and it is null without quota
2537          */
2538         [DTO_ATTR_SET_CHOWN] = 0
2539 };
2540
2541 /* reserve or free quota for some operation */
2542 static int osd_reserve_or_free_quota(const struct lu_env *env,
2543                                      struct dt_device *dev,
2544                                      struct lquota_id_info *qi)
2545 {
2546         struct osd_device       *osd = osd_dt_dev(dev);
2547         struct qsd_instance     *qsd = NULL;
2548         int rc;
2549
2550         ENTRY;
2551
2552         if (qi->lqi_is_blk)
2553                 qsd = osd->od_quota_slave_dt;
2554         else
2555                 qsd = osd->od_quota_slave_md;
2556
2557         rc = qsd_reserve_or_free_quota(env, qsd, qi);
2558         RETURN(rc);
2559 }
2560
2561 static const struct dt_device_operations osd_dt_ops = {
2562         .dt_root_get              = osd_root_get,
2563         .dt_statfs                = osd_statfs,
2564         .dt_trans_create          = osd_trans_create,
2565         .dt_trans_start           = osd_trans_start,
2566         .dt_trans_stop            = osd_trans_stop,
2567         .dt_trans_cb_add          = osd_trans_cb_add,
2568         .dt_conf_get              = osd_conf_get,
2569         .dt_mnt_sb_get            = osd_mnt_sb_get,
2570         .dt_sync                  = osd_sync,
2571         .dt_ro                    = osd_ro,
2572         .dt_commit_async          = osd_commit_async,
2573         .dt_reserve_or_free_quota = osd_reserve_or_free_quota,
2574 };
2575
2576 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2577                           unsigned int role)
2578 {
2579         struct osd_object *obj = osd_dt_obj(dt);
2580         struct osd_thread_info *oti = osd_oti_get(env);
2581
2582         LINVRNT(osd_invariant(obj));
2583
2584         LASSERT(obj->oo_owner != env);
2585         down_read_nested(&obj->oo_sem, role);
2586
2587         LASSERT(obj->oo_owner == NULL);
2588         oti->oti_r_locks++;
2589 }
2590
2591 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2592                            unsigned int role)
2593 {
2594         struct osd_object *obj = osd_dt_obj(dt);
2595         struct osd_thread_info *oti = osd_oti_get(env);
2596
2597         LINVRNT(osd_invariant(obj));
2598
2599         LASSERT(obj->oo_owner != env);
2600         down_write_nested(&obj->oo_sem, role);
2601
2602         LASSERT(obj->oo_owner == NULL);
2603         obj->oo_owner = env;
2604         oti->oti_w_locks++;
2605 }
2606
2607 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2608 {
2609         struct osd_object *obj = osd_dt_obj(dt);
2610         struct osd_thread_info *oti = osd_oti_get(env);
2611
2612         LINVRNT(osd_invariant(obj));
2613
2614         LASSERT(oti->oti_r_locks > 0);
2615         oti->oti_r_locks--;
2616         up_read(&obj->oo_sem);
2617 }
2618
2619 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2620 {
2621         struct osd_object *obj = osd_dt_obj(dt);
2622         struct osd_thread_info *oti = osd_oti_get(env);
2623
2624         LINVRNT(osd_invariant(obj));
2625
2626         LASSERT(obj->oo_owner == env);
2627         LASSERT(oti->oti_w_locks > 0);
2628         oti->oti_w_locks--;
2629         obj->oo_owner = NULL;
2630         up_write(&obj->oo_sem);
2631 }
2632
2633 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2634 {
2635         struct osd_object *obj = osd_dt_obj(dt);
2636
2637         LINVRNT(osd_invariant(obj));
2638
2639         return obj->oo_owner == env;
2640 }
2641
2642 static void osd_inode_getattr(const struct lu_env *env,
2643                               struct inode *inode, struct lu_attr *attr)
2644 {
2645         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2646                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2647                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2648                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2649
2650         attr->la_atime = inode->i_atime.tv_sec;
2651         attr->la_mtime = inode->i_mtime.tv_sec;
2652         attr->la_ctime = inode->i_ctime.tv_sec;
2653         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2654         attr->la_mode    = inode->i_mode;
2655         attr->la_size    = i_size_read(inode);
2656         attr->la_blocks  = inode->i_blocks;
2657         attr->la_uid     = i_uid_read(inode);
2658         attr->la_gid     = i_gid_read(inode);
2659         attr->la_projid  = i_projid_read(inode);
2660         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2661         attr->la_nlink   = inode->i_nlink;
2662         attr->la_rdev    = inode->i_rdev;
2663         attr->la_blksize = 1 << inode->i_blkbits;
2664         attr->la_blkbits = inode->i_blkbits;
2665         /*
2666          * Ext4 did not transfer inherit flags from raw inode
2667          * to inode flags, and ext4 internally test raw inode
2668          * @i_flags directly. Instead of patching ext4, we do it here.
2669          */
2670         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2671                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2672 }
2673
2674 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2675                             u64 *count)
2676 {
2677         struct osd_object *obj = osd_dt_obj(dt);
2678         const struct dt_it_ops *iops;
2679         struct dt_it *it;
2680         int rc;
2681
2682         ENTRY;
2683
2684         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2685         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2686
2687         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2688                 *count = obj->oo_dirent_count;
2689                 RETURN(0);
2690         }
2691
2692         /* directory not initialized yet */
2693         if (!dt->do_index_ops) {
2694                 *count = 0;
2695                 RETURN(0);
2696         }
2697
2698         iops = &dt->do_index_ops->dio_it;
2699         it = iops->init(env, dt, LUDA_64BITHASH);
2700         if (IS_ERR(it))
2701                 RETURN(PTR_ERR(it));
2702
2703         rc = iops->load(env, it, 0);
2704         if (rc < 0) {
2705                 if (rc == -ENODATA) {
2706                         rc = 0;
2707                         *count = 0;
2708                 }
2709                 GOTO(out, rc);
2710         }
2711         if (rc > 0)
2712                 rc = iops->next(env, it);
2713
2714         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2715                 if (rc == -ESTALE)
2716                         continue;
2717
2718                 if (iops->key_size(env, it) == 0)
2719                         continue;
2720
2721                 (*count)++;
2722         }
2723         if (rc == 1) {
2724                 obj->oo_dirent_count = *count;
2725                 rc = 0;
2726         }
2727 out:
2728         iops->put(env, it);
2729         iops->fini(env, it);
2730
2731         RETURN(rc);
2732 }
2733
2734 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2735                         struct lu_attr *attr)
2736 {
2737         struct osd_object *obj = osd_dt_obj(dt);
2738         int rc = 0;
2739
2740         if (unlikely(!dt_object_exists(dt)))
2741                 return -ENOENT;
2742         if (unlikely(obj->oo_destroyed))
2743                 return -ENOENT;
2744
2745         LASSERT(!dt_object_remote(dt));
2746         LINVRNT(osd_invariant(obj));
2747
2748         spin_lock(&obj->oo_guard);
2749         osd_inode_getattr(env, obj->oo_inode, attr);
2750         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2751                 attr->la_valid |= LA_FLAGS;
2752                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2753         }
2754         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2755                 attr->la_valid |= LA_FLAGS;
2756                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2757         }
2758         spin_unlock(&obj->oo_guard);
2759
2760         if (S_ISDIR(obj->oo_inode->i_mode) &&
2761             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2762                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2763
2764         return rc;
2765 }
2766
2767 static int osd_declare_attr_qid(const struct lu_env *env,
2768                                 struct osd_object *obj,
2769                                 struct osd_thandle *oh, long long bspace,
2770                                 qid_t old_id, qid_t new_id, bool enforce,
2771                                 unsigned int type)
2772 {
2773         int rc;
2774         struct osd_thread_info *info = osd_oti_get(env);
2775         struct lquota_id_info  *qi = &info->oti_qi;
2776
2777         qi->lqi_type = type;
2778         /* inode accounting */
2779         qi->lqi_is_blk = false;
2780
2781         /* one more inode for the new id ... */
2782         qi->lqi_id.qid_uid = new_id;
2783         qi->lqi_space      = 1;
2784         /* Reserve credits for the new id */
2785         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2786         if (rc == -EDQUOT || rc == -EINPROGRESS)
2787                 rc = 0;
2788         if (rc)
2789                 RETURN(rc);
2790
2791         /* and one less inode for the current id */
2792         qi->lqi_id.qid_uid = old_id;
2793         qi->lqi_space = -1;
2794         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2795         if (rc == -EDQUOT || rc == -EINPROGRESS)
2796                 rc = 0;
2797         if (rc)
2798                 RETURN(rc);
2799
2800         /* block accounting */
2801         qi->lqi_is_blk = true;
2802
2803         /* more blocks for the new id ... */
2804         qi->lqi_id.qid_uid = new_id;
2805         qi->lqi_space      = bspace;
2806         /*
2807          * Credits for the new uid has been reserved, re-use "obj"
2808          * to save credit reservation.
2809          */
2810         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2811         if (rc == -EDQUOT || rc == -EINPROGRESS)
2812                 rc = 0;
2813         if (rc)
2814                 RETURN(rc);
2815
2816         /* and finally less blocks for the current uid */
2817         qi->lqi_id.qid_uid = old_id;
2818         qi->lqi_space      = -bspace;
2819         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2820         if (rc == -EDQUOT || rc == -EINPROGRESS)
2821                 rc = 0;
2822
2823         RETURN(rc);
2824 }
2825
2826 static int osd_declare_attr_set(const struct lu_env *env,
2827                                 struct dt_object *dt,
2828                                 const struct lu_attr *attr,
2829                                 struct thandle *handle)
2830 {
2831         struct osd_thandle *oh;
2832         struct osd_object *obj;
2833         qid_t uid;
2834         qid_t gid;
2835         long long bspace;
2836         int rc = 0;
2837         bool enforce;
2838
2839         ENTRY;
2840
2841         LASSERT(dt != NULL);
2842         LASSERT(handle != NULL);
2843
2844         obj = osd_dt_obj(dt);
2845         LASSERT(osd_invariant(obj));
2846
2847         oh = container_of(handle, struct osd_thandle, ot_super);
2848         LASSERT(oh->ot_handle == NULL);
2849
2850         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2851                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2852
2853         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2854                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2855
2856         if (attr == NULL || obj->oo_inode == NULL)
2857                 RETURN(rc);
2858
2859         bspace   = obj->oo_inode->i_blocks << 9;
2860         bspace   = toqb(bspace);
2861
2862         /*
2863          * Changing ownership is always preformed by super user, it should not
2864          * fail with EDQUOT unless required explicitly.
2865          *
2866          * We still need to call the osd_declare_qid() to calculate the journal
2867          * credits for updating quota accounting files and to trigger quota
2868          * space adjustment once the operation is completed.
2869          */
2870         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2871                 /* USERQUOTA */
2872                 uid = i_uid_read(obj->oo_inode);
2873                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2874                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2875                                           attr->la_uid, enforce, USRQUOTA);
2876                 if (rc)
2877                         RETURN(rc);
2878
2879                 gid = i_gid_read(obj->oo_inode);
2880                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2881                        attr->la_uid, gid, attr->la_gid);
2882                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2883                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2884                                           attr->la_gid, enforce, GRPQUOTA);
2885                 if (rc)
2886                         RETURN(rc);
2887
2888         }
2889 #ifdef HAVE_PROJECT_QUOTA
2890         if (attr->la_valid & LA_PROJID) {
2891                 __u32 projid = i_projid_read(obj->oo_inode);
2892
2893                 enforce = (attr->la_valid & LA_PROJID) &&
2894                                         (attr->la_projid != projid);
2895                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2896                                           (qid_t)projid, (qid_t)attr->la_projid,
2897                                           enforce, PRJQUOTA);
2898                 if (rc)
2899                         RETURN(rc);
2900         }
2901 #endif
2902         RETURN(rc);
2903 }
2904
2905 static int osd_inode_setattr(const struct lu_env *env,
2906                              struct inode *inode, const struct lu_attr *attr)
2907 {
2908         __u64 bits = attr->la_valid;
2909
2910         /* Only allow set size for regular file */
2911         if (!S_ISREG(inode->i_mode))
2912                 bits &= ~(LA_SIZE | LA_BLOCKS);
2913
2914         if (bits == 0)
2915                 return 0;
2916
2917         if (bits & LA_ATIME)
2918                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2919         if (bits & LA_CTIME)
2920                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2921         if (bits & LA_MTIME)
2922                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2923         if (bits & LA_SIZE) {
2924                 spin_lock(&inode->i_lock);
2925                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2926                 i_size_write(inode, attr->la_size);
2927                 spin_unlock(&inode->i_lock);
2928         }
2929
2930         /*
2931          * OSD should not change "i_blocks" which is used by quota.
2932          * "i_blocks" should be changed by ldiskfs only.
2933          */
2934         if (bits & LA_MODE)
2935                 inode->i_mode = (inode->i_mode & S_IFMT) |
2936                                 (attr->la_mode & ~S_IFMT);
2937         if (bits & LA_UID)
2938                 i_uid_write(inode, attr->la_uid);
2939         if (bits & LA_GID)
2940                 i_gid_write(inode, attr->la_gid);
2941         if (bits & LA_PROJID)
2942                 i_projid_write(inode, attr->la_projid);
2943         if (bits & LA_NLINK)
2944                 set_nlink(inode, attr->la_nlink);
2945         if (bits & LA_RDEV)
2946                 inode->i_rdev = attr->la_rdev;
2947
2948         if (bits & LA_FLAGS) {
2949                 /* always keep S_NOCMTIME */
2950                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2951                                  S_NOCMTIME;
2952 #if defined(S_ENCRYPTED)
2953                 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2954                  * aware of encryption status. It is just stored into LMA
2955                  * so that it can be forwared to client side.
2956                  */
2957                 inode->i_flags &= ~S_ENCRYPTED;
2958 #endif
2959                 /*
2960                  * Ext4 did not transfer inherit flags from
2961                  * @inode->i_flags to raw inode i_flags when writing
2962                  * flags, we do it explictly here.
2963                  */
2964                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2965                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2966                 else
2967                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2968         }
2969         return 0;
2970 }
2971
2972 #ifdef HAVE_PROJECT_QUOTA
2973 static int osd_transfer_project(struct inode *inode, __u32 projid,
2974                                 struct thandle *handle)
2975 {
2976         struct super_block *sb = inode->i_sb;
2977         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2978         int err;
2979         kprojid_t kprojid;
2980         struct ldiskfs_iloc iloc;
2981         struct ldiskfs_inode *raw_inode;
2982         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2983
2984         if (!ldiskfs_has_feature_project(sb)) {
2985                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2986                         == LDISKFS_DEF_PROJID);
2987                 if (projid != LDISKFS_DEF_PROJID)
2988                         return -EOPNOTSUPP;
2989                 else
2990                         return 0;
2991         }
2992
2993         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2994                 return -EOPNOTSUPP;
2995
2996         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2997         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2998                 return 0;
2999
3000         err = ldiskfs_get_inode_loc(inode, &iloc);
3001         if (err)
3002                 return err;
3003
3004         raw_inode = ldiskfs_raw_inode(&iloc);
3005         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3006                 struct osd_thandle *oh = container_of(handle,
3007                                                       struct osd_thandle,
3008                                                       ot_super);
3009                 /**
3010                  * try to expand inode size automatically.
3011                  */
3012                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3013                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3014                         err = -EOVERFLOW;
3015                         brelse(iloc.bh);
3016                         return err;
3017                 }
3018         }
3019         brelse(iloc.bh);
3020
3021         dquot_initialize(inode);
3022         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3023         if (transfer_to[PRJQUOTA]) {
3024                 lock_dquot_transfer(inode);
3025                 err = __dquot_transfer(inode, transfer_to);
3026                 unlock_dquot_transfer(inode);
3027                 dqput(transfer_to[PRJQUOTA]);
3028                 if (err)
3029                         return err;
3030         }
3031
3032         return err;
3033 }
3034 #endif
3035
3036 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3037                               struct thandle *handle)
3038 {
3039         int rc;
3040
3041         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3042             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3043                 struct iattr iattr;
3044
3045                 CDEBUG(D_QUOTA,
3046                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3047                        inode->i_ino, i_uid_read(inode), attr->la_uid,
3048                        i_gid_read(inode), attr->la_gid);
3049
3050                 dquot_initialize(inode);
3051                 iattr.ia_valid = 0;
3052                 if (attr->la_valid & LA_UID)
3053                         iattr.ia_valid |= ATTR_UID;
3054                 if (attr->la_valid & LA_GID)
3055                         iattr.ia_valid |= ATTR_GID;
3056                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3057                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3058
3059                 lock_dquot_transfer(inode);
3060                 rc = dquot_transfer(inode, &iattr);
3061                 unlock_dquot_transfer(inode);
3062                 if (rc) {
3063                         CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3064                                osd_ino2name(inode), rc);
3065                         return rc;
3066                 }
3067         }
3068
3069         /* Handle project id transfer here properly */
3070         if (attr->la_valid & LA_PROJID &&
3071             attr->la_projid != i_projid_read(inode)) {
3072                 if (!projid_valid(make_kprojid(&init_user_ns, attr->la_projid)))
3073                         return -EINVAL;
3074 #ifdef HAVE_PROJECT_QUOTA
3075                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3076 #else
3077                 rc = -ENOTSUPP;
3078 #endif
3079                 if (rc) {
3080                         CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3081                                osd_ino2name(inode), rc);
3082                         return rc;
3083                 }
3084         }
3085         return 0;
3086 }
3087
3088 static int osd_attr_set(const struct lu_env *env,
3089                         struct dt_object *dt,
3090                         const struct lu_attr *attr,
3091                         struct thandle *handle)
3092 {
3093         struct osd_object *obj = osd_dt_obj(dt);
3094         struct inode *inode;
3095         int rc;
3096
3097         if (!dt_object_exists(dt))
3098                 return -ENOENT;
3099
3100         LASSERT(handle != NULL);
3101         LASSERT(!dt_object_remote(dt));
3102         LASSERT(osd_invariant(obj));
3103
3104         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3105
3106         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3107             !osd_obj2dev(obj)->od_is_ost) {
3108                 struct osd_thread_info *oti = osd_oti_get(env);
3109                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3110                 struct lu_fid *fid1 = &oti->oti_fid;
3111                 struct osd_inode_id *id = &oti->oti_id;
3112                 struct iam_path_descr *ipd;
3113                 struct iam_container *bag;
3114                 struct osd_thandle *oh;
3115                 int rc;
3116
3117                 fid_cpu_to_be(fid1, fid0);
3118                 memset(id, 1, sizeof(*id));
3119                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3120                                   fid0)->oi_dir.od_container;
3121                 ipd = osd_idx_ipd_get(env, bag);
3122                 if (unlikely(ipd == NULL))
3123                         RETURN(-ENOMEM);
3124
3125                 oh = container_of(handle, struct osd_thandle, ot_super);
3126                 rc = iam_update(oh->ot_handle, bag,
3127                                 (const struct iam_key *)fid1,
3128                                 (const struct iam_rec *)id, ipd);
3129                 osd_ipd_put(env, bag, ipd);
3130                 return(rc > 0 ? 0 : rc);
3131         }
3132
3133         inode = obj->oo_inode;
3134
3135         rc = osd_quota_transfer(inode, attr, handle);
3136         if (rc)
3137                 return rc;
3138
3139         spin_lock(&obj->oo_guard);
3140         rc = osd_inode_setattr(env, inode, attr);
3141         spin_unlock(&obj->oo_guard);
3142         if (rc != 0)
3143                 GOTO(out, rc);
3144
3145         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3146
3147         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3148
3149         if (!(attr->la_valid & LA_FLAGS))
3150                 GOTO(out, rc);
3151
3152         /* Let's check if there are extra flags need to be set into LMA */
3153         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3154                 struct osd_thread_info *info = osd_oti_get(env);
3155                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3156
3157                 LASSERT(!obj->oo_pfid_in_lma);
3158
3159                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3160                                  &info->oti_ost_attrs);
3161                 if (rc)
3162                         GOTO(out, rc);
3163
3164                 lma->lma_incompat |=
3165                         lustre_to_lma_flags(attr->la_flags);
3166                 lustre_lma_swab(lma);
3167
3168                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3169
3170                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3171                                      lma, sizeof(*lma), XATTR_REPLACE);
3172                 if (rc != 0) {
3173                         struct osd_device *osd = osd_obj2dev(obj);
3174
3175                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3176                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3177                               lma->lma_incompat, rc);
3178                 } else {
3179                         obj->oo_lma_flags =
3180                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3181                 }
3182                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3183         }
3184 out:
3185
3186         return rc;
3187 }
3188
3189 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3190                                            struct osd_object *obj,
3191                                            const char *name, const int namelen)
3192 {
3193         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3194 }
3195
3196 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3197                       umode_t mode, struct dt_allocation_hint *hint,
3198                       struct thandle *th, struct lu_attr *attr)
3199 {
3200         int result;
3201         struct osd_device *osd = osd_obj2dev(obj);
3202         struct osd_thandle *oth;
3203         struct dt_object *parent = NULL;
3204         struct inode *inode;
3205         struct iattr iattr = {
3206                 .ia_valid = ATTR_UID | ATTR_GID |
3207                             ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3208                 .ia_ctime.tv_sec = attr->la_ctime,
3209                 .ia_mtime.tv_sec = attr->la_mtime,
3210                 .ia_atime.tv_sec = attr->la_atime,
3211                 .ia_uid = GLOBAL_ROOT_UID,
3212                 .ia_gid = GLOBAL_ROOT_GID,
3213         };
3214         const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
3215
3216         if (attr->la_valid & LA_UID)
3217                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3218         if (attr->la_valid & LA_GID)
3219                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3220
3221         LINVRNT(osd_invariant(obj));
3222         LASSERT(obj->oo_inode == NULL);
3223         LASSERT(obj->oo_hl_head == NULL);
3224
3225         if (S_ISDIR(mode) && ldiskfs_pdo) {
3226                 obj->oo_hl_head =
3227                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3228                 if (obj->oo_hl_head == NULL)
3229                         return -ENOMEM;
3230         }
3231
3232         oth = container_of(th, struct osd_thandle, ot_super);
3233         LASSERT(oth->ot_handle->h_transaction != NULL);
3234
3235         if (hint != NULL && hint->dah_parent != NULL &&
3236             !dt_object_remote(hint->dah_parent))
3237                 parent = hint->dah_parent;
3238
3239         /* if a time component is not valid set it to UTIME_OMIT */
3240         if (!(attr->la_valid & LA_CTIME))
3241                 iattr.ia_ctime = omit;
3242         if (!(attr->la_valid & LA_MTIME))
3243                 iattr.ia_mtime = omit;
3244         if (!(attr->la_valid & LA_ATIME))
3245                 iattr.ia_atime = omit;
3246
3247         inode = ldiskfs_create_inode(oth->ot_handle,
3248                                      parent ? osd_dt_obj(parent)->oo_inode :
3249                                               osd_sb(osd)->s_root->d_inode,
3250                                      mode, &iattr);
3251         if (!IS_ERR(inode)) {
3252                 /* Do not update file c/mtime in ldiskfs. */
3253                 inode->i_flags |= S_NOCMTIME;
3254
3255                 /*
3256                  * For new created object, it must be consistent,
3257                  * and it is unnecessary to scrub against it.
3258                  */
3259                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3260
3261                 obj->oo_inode = inode;
3262                 result = 0;
3263         } else {
3264                 if (obj->oo_hl_head != NULL) {
3265                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3266                         obj->oo_hl_head = NULL;
3267                 }
3268                 result = PTR_ERR(inode);
3269         }
3270         LINVRNT(osd_invariant(obj));
3271         return result;
3272 }
3273
3274 enum {
3275         OSD_NAME_LEN = 255
3276 };
3277
3278 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3279                      struct lu_attr *attr,
3280                      struct dt_allocation_hint *hint,
3281                      struct dt_object_format *dof,
3282                      struct thandle *th)
3283 {
3284         int result;
3285         struct osd_thandle *oth;
3286         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3287
3288         LASSERT(S_ISDIR(attr->la_mode));
3289
3290         oth = container_of(th, struct osd_thandle, ot_super);
3291         LASSERT(oth->ot_handle->h_transaction != NULL);
3292         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3293                 obj->oo_dirent_count = 0;
3294         result = osd_mkfile(info, obj, mode, hint, th, attr);
3295
3296         return result;
3297 }
3298
3299 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3300                         struct lu_attr *attr,
3301                         struct dt_allocation_hint *hint,
3302                         struct dt_object_format *dof,
3303                         struct thandle *th)
3304 {
3305         int result;
3306         struct osd_thandle *oth;
3307         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3308
3309         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3310
3311         LASSERT(S_ISREG(attr->la_mode));
3312
3313         oth = container_of(th, struct osd_thandle, ot_super);
3314         LASSERT(oth->ot_handle->h_transaction != NULL);
3315
3316         result = osd_mkfile(info, obj, mode, hint, th, attr);
3317         if (result == 0) {
3318                 LASSERT(obj->oo_inode != NULL);
3319                 if (feat->dif_flags & DT_IND_VARKEY)
3320                         result = iam_lvar_create(obj->oo_inode,
3321                                                  feat->dif_keysize_max,
3322                                                  feat->dif_ptrsize,
3323                                                  feat->dif_recsize_max,
3324                                                  oth->ot_handle);
3325                 else
3326                         result = iam_lfix_create(obj->oo_inode,
3327                                                  feat->dif_keysize_max,
3328                                                  feat->dif_ptrsize,
3329                                                  feat->dif_recsize_max,
3330                                                  oth->ot_handle);
3331         }
3332         return result;
3333 }
3334
3335 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3336                      struct lu_attr *attr,
3337                      struct dt_allocation_hint *hint,
3338                      struct dt_object_format *dof,
3339                      struct thandle *th)
3340 {
3341         LASSERT(S_ISREG(attr->la_mode));
3342         return osd_mkfile(info, obj, (attr->la_mode &
3343                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3344                           attr);
3345 }
3346
3347 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3348                      struct lu_attr *attr,
3349                      struct dt_allocation_hint *hint,
3350                      struct dt_object_format *dof,
3351                      struct thandle *th)
3352 {
3353         LASSERT(S_ISLNK(attr->la_mode));
3354         return osd_mkfile(info, obj, (attr->la_mode &
3355                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3356                           attr);
3357 }
3358
3359 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3360                      struct lu_attr *attr,
3361                      struct dt_allocation_hint *hint,
3362                      struct dt_object_format *dof,
3363                      struct thandle *th)
3364 {
3365         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3366         int result;
3367
3368         LINVRNT(osd_invariant(obj));
3369         LASSERT(obj->oo_inode == NULL);
3370         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3371                 S_ISFIFO(mode) || S_ISSOCK(mode));
3372
3373         result = osd_mkfile(info, obj, mode, hint, th, attr);
3374         if (result == 0) {
3375                 LASSERT(obj->oo_inode != NULL);
3376                 /*
3377                  * This inode should be marked dirty for i_rdev.  Currently
3378                  * that is done in the osd_attr_init().
3379                  */
3380                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3381                                    attr->la_rdev);
3382         }
3383         LINVRNT(osd_invariant(obj));
3384         return result;
3385 }
3386
3387 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3388                               struct lu_attr *,
3389                               struct dt_allocation_hint *hint,
3390                               struct dt_object_format *dof,
3391                               struct thandle *);
3392
3393 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3394 {
3395         osd_obj_type_f result;
3396
3397         switch (type) {
3398         case DFT_DIR:
3399                 result = osd_mkdir;
3400                 break;
3401         case DFT_REGULAR:
3402                 result = osd_mkreg;
3403                 break;
3404         case DFT_SYM:
3405                 result = osd_mksym;
3406                 break;
3407         case DFT_NODE:
3408                 result = osd_mknod;
3409                 break;
3410         case DFT_INDEX:
3411                 result = osd_mk_index;
3412                 break;
3413
3414         default:
3415                 LBUG();
3416                 break;
3417         }
3418         return result;
3419 }
3420
3421 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3422                         struct dt_object *parent, struct dt_object *child,
3423                         umode_t child_mode)
3424 {
3425         LASSERT(ah);
3426
3427         ah->dah_parent = parent;
3428
3429         if (parent != NULL && !dt_object_remote(parent)) {
3430                 /* will help to find FID->ino at dt_insert("..") */
3431                 struct osd_object *pobj = osd_dt_obj(parent);
3432
3433                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3434         }
3435 }
3436
3437 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3438                           struct lu_attr *attr, struct dt_object_format *dof,
3439                           struct thandle *handle)
3440 {
3441         struct inode *inode = obj->oo_inode;
3442         __u64 valid = attr->la_valid;
3443         int result;
3444
3445         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3446
3447         if (dof->dof_type != DFT_NODE)
3448                 attr->la_valid &= ~LA_RDEV;
3449         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3450                 attr->la_valid &= ~LA_ATIME;
3451         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3452                 attr->la_valid &= ~LA_CTIME;
3453         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3454                 attr->la_valid &= ~LA_MTIME;
3455
3456         result = osd_quota_transfer(inode, attr, handle);
3457         if (result)
3458                 return;
3459
3460         if (attr->la_valid != 0) {
3461                 result = osd_inode_setattr(info->oti_env, inode, attr);
3462                 /*
3463                  * The osd_inode_setattr() should always succeed here.  The
3464                  * only error that could be returned is EDQUOT when we are
3465                  * trying to change the UID or GID of the inode. However, this
3466                  * should not happen since quota enforcement is no longer
3467                  * enabled on ldiskfs (lquota takes care of it).
3468                  */
3469                 LASSERTF(result == 0, "%d\n", result);
3470                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3471         }
3472
3473         attr->la_valid = valid;
3474 }
3475
3476 /**
3477  * Helper function for osd_create()
3478  *
3479  * \retval 0, on success
3480  */
3481 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3482                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3483                         struct dt_object_format *dof, struct thandle *th)
3484 {
3485         int result;
3486         __u32 umask;
3487
3488         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3489
3490         /* we drop umask so that permissions we pass are not affected */
3491         umask = current->fs->umask;
3492         current->fs->umask = 0;
3493
3494         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3495                                                   th);
3496         if (likely(obj->oo_inode != NULL)) {
3497                 LASSERT(obj->oo_inode->i_state & I_NEW);
3498
3499                 /*
3500                  * Unlock the inode before attr initialization to avoid
3501                  * unnecessary dqget operations. LU-6378
3502                  */
3503                 unlock_new_inode(obj->oo_inode);
3504         }
3505
3506         if (likely(result == 0)) {
3507                 osd_attr_init(info, obj, attr, dof, th);
3508                 osd_object_init0(obj);
3509         }
3510
3511         /* restore previous umask value */
3512         current->fs->umask = umask;
3513
3514         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3515
3516         return result;
3517 }
3518
3519 /**
3520  * Helper function for osd_create()
3521  *
3522  * \retval 0, on success
3523  */
3524 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3525                            const struct lu_fid *fid, struct thandle *th)
3526 {
3527         struct osd_thread_info *info = osd_oti_get(env);
3528         struct osd_inode_id    *id   = &info->oti_id;
3529         struct osd_device      *osd  = osd_obj2dev(obj);
3530         struct osd_thandle     *oh;
3531         int rc;
3532
3533         LASSERT(obj->oo_inode != NULL);
3534
3535         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
3536                 return -ENOSPC;
3537
3538         oh = container_of(th, struct osd_thandle, ot_super);
3539         LASSERT(oh->ot_handle);
3540         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3541
3542         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3543         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3544                            OI_CHECK_FLD, NULL);
3545         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3546                 struct lu_fid next_fid = *fid;
3547
3548                 /* insert next object in advance, and map to the same inode */
3549                 next_fid.f_oid++;
3550                 if (next_fid.f_oid != 0) {
3551                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3552                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3553                                       OI_CHECK_FLD, NULL);
3554                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3555                 }
3556         }
3557
3558         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3559
3560         return rc;
3561 }
3562
3563 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3564                    u64 seq, struct lu_seq_range *range)
3565 {
3566         struct seq_server_site *ss = osd_seq_site(osd);
3567
3568         if (fid_seq_is_idif(seq)) {
3569                 fld_range_set_ost(range);
3570                 range->lsr_index = idif_ost_idx(seq);
3571                 return 0;
3572         }
3573
3574         if (!fid_seq_in_fldb(seq)) {
3575                 fld_range_set_mdt(range);
3576                 if (ss != NULL)
3577                         /*
3578                          * FIXME: If ss is NULL, it suppose not get lsr_index
3579                          * at all
3580                          */
3581                         range->lsr_index = ss->ss_node_id;
3582                 return 0;
3583         }
3584
3585         LASSERT(ss != NULL);
3586         fld_range_set_any(range);
3587         /* OSD will only do local fld lookup */
3588         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3589 }
3590
3591 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3592                               struct lu_attr *attr,
3593                               struct dt_allocation_hint *hint,
3594                               struct dt_object_format *dof,
3595                               struct thandle *handle)
3596 {
3597         struct osd_thandle *oh;
3598         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3599         int credits;
3600         int rc;
3601
3602         ENTRY;
3603
3604         LASSERT(handle != NULL);
3605
3606         oh = container_of(handle, struct osd_thandle, ot_super);
3607         LASSERT(oh->ot_handle == NULL);
3608
3609         /*
3610          * EA object consumes more credits than regular object: osd_mk_index
3611          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3612          * leaf_node, could involves the block, block bitmap, groups, GDT
3613          * change for each block, so add 4 * 2 credits in that case.
3614          *
3615          * The default ACL initialization may consume an additional 16 blocks
3616          */
3617         credits = osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3618                   ((dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3619
3620         /**
3621          * While ldiskfs_new_inode() calls ldiskfs_init_acl() we have to add
3622          * credits for possible default ACL creation in new inode
3623          */
3624         if (hint && hint->dah_acl_len)
3625                 credits += osd_calc_bkmap_credits(sb, NULL, 0, -1,
3626                                 (hint->dah_acl_len + sb->s_blocksize - 1) >>
3627                                 sb->s_blocksize_bits);
3628
3629         osd_trans_declare_op(env, oh, OSD_OT_CREATE, credits);
3630
3631         /*
3632          * Reuse idle OI block may cause additional one OI block
3633          * to be changed.
3634          */
3635         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3636                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3637         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3638                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3639                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3640
3641         /* will help to find FID->ino mapping at dt_insert() */
3642         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3643                                    osd_dt_obj(dt));
3644         if (rc != 0)
3645                 RETURN(rc);
3646
3647         if (!attr)
3648                 RETURN(0);
3649
3650         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3651                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3652                                    NULL, OSD_QID_INODE);
3653         if (rc != 0)
3654                 RETURN(rc);
3655
3656         RETURN(rc);
3657 }
3658
3659 /**
3660  * Called to destroy on-disk representation of the object
3661  *
3662  * Concurrency: must be locked
3663  */
3664 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3665                                struct thandle *th)
3666 {
3667         struct osd_object *obj = osd_dt_obj(dt);
3668         struct inode *inode = obj->oo_inode;
3669         struct osd_thandle *oh;
3670         int rc;
3671
3672         ENTRY;
3673
3674         if (inode == NULL)
3675                 RETURN(-ENOENT);
3676
3677         oh = container_of(th, struct osd_thandle, ot_super);
3678         LASSERT(oh->ot_handle == NULL);
3679
3680         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3681                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3682
3683         /* For removing agent entry */
3684         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3685                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3686
3687         /*
3688          * Recycle idle OI leaf may cause additional three OI blocks
3689          * to be changed.
3690          */
3691         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3692                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3693                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3694         /* one less inode */
3695         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3696                                    i_projid_read(inode), -1, oh, obj, NULL,
3697                                    OSD_QID_INODE);
3698         if (rc)
3699                 RETURN(rc);
3700         /* data to be truncated */
3701         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3702                                    i_projid_read(inode), 0, oh, obj, NULL,
3703                                    OSD_QID_BLK);
3704         if (rc)
3705                 RETURN(rc);
3706
3707         /*
3708          * will help to find FID->ino when this object is being
3709          * added to PENDING
3710          */
3711         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3712
3713         RETURN(rc);
3714 }
3715
3716 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3717                        struct thandle *th)
3718 {
3719         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3720         struct osd_object *obj = osd_dt_obj(dt);
3721         struct inode *inode = obj->oo_inode;
3722         struct osd_device *osd = osd_obj2dev(obj);
3723         struct osd_thandle *oh;
3724         int result;
3725
3726         ENTRY;
3727
3728         oh = container_of(th, struct osd_thandle, ot_super);
3729         LASSERT(oh->ot_handle);
3730         LASSERT(inode);
3731         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3732
3733         if (unlikely(fid_is_acct(fid)))
3734                 RETURN(-EPERM);
3735
3736         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3737                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3738                 if (result != 0)
3739                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3740                                osd_name(osd), PFID(fid), result);
3741         }
3742
3743         if (S_ISDIR(inode->i_mode)) {
3744                 if (inode->i_nlink > 2)
3745                         CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3746                                osd_name(osd), PFID(fid), inode->i_ino,
3747                                inode->i_nlink);
3748
3749                 spin_lock(&obj->oo_guard);
3750                 clear_nlink(inode);
3751                 spin_unlock(&obj->oo_guard);
3752                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3753         }
3754
3755         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3756
3757         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3758
3759         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3760                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3761                                        oh->ot_handle, OI_CHECK_FLD);
3762
3763         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3764         /* XXX: add to ext3 orphan list */
3765         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3766
3767         /* not needed in the cache anymore */
3768         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3769         obj->oo_destroyed = 1;
3770
3771         RETURN(0);
3772 }
3773
3774 /**
3775  * Put the fid into lustre_mdt_attrs, and then place the structure
3776  * inode's ea. This fid should not be altered during the life time
3777  * of the inode.
3778  *
3779  * \retval +ve, on success
3780  * \retval -ve, on error
3781  *
3782  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3783  */
3784 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3785                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3786 {
3787         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3788         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3789         int rc;
3790
3791         ENTRY;
3792
3793         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3794                 RETURN(0);
3795
3796         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3797                 rc = -ENOMEM;
3798
3799         lustre_loa_init(loa, fid, compat, incompat);
3800         lustre_loa_swab(loa, false);
3801
3802         /*
3803          * For the OST device with 256 bytes inode size by default,
3804          * the PFID EA will be stored together with LMA EA to avoid
3805          * performance trouble. Otherwise the PFID EA can be stored
3806          * independently. LU-8998
3807          */
3808         if ((compat & LMAC_FID_ON_OST) &&
3809             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3810                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3811                                      sizeof(*loa), XATTR_CREATE);
3812         else
3813                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3814                                      sizeof(*lma), XATTR_CREATE);
3815         /*
3816          * LMA may already exist, but we need to check that all the
3817          * desired compat/incompat flags have been added.
3818          */
3819         if (unlikely(rc == -EEXIST)) {
3820                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3821                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3822                 if (rc < 0)
3823                         RETURN(rc);
3824
3825                 if (rc < sizeof(*lma))
3826                         RETURN(-EINVAL);
3827
3828                 lustre_loa_swab(loa, true);
3829                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3830                     ((compat == 0 && incompat == 0) ||
3831                      (!(~lma->lma_compat & compat) &&
3832                       !(~lma->lma_incompat & incompat))))
3833                         RETURN(0);
3834
3835                 lma->lma_self_fid = *fid;
3836                 lma->lma_compat |= compat;
3837                 lma->lma_incompat |= incompat;
3838                 if (rc == sizeof(*lma)) {
3839                         lustre_lma_swab(lma);
3840                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3841                                              sizeof(*lma), XATTR_REPLACE);
3842                 } else {
3843                         lustre_loa_swab(loa, false);
3844                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3845                                              sizeof(*loa), XATTR_REPLACE);
3846                 }
3847         }
3848
3849         RETURN(rc);
3850 }
3851
3852 /**
3853  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3854  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3855  * To have compatilibility with 1.8 ldiskfs driver we need to have
3856  * magic number at start of fid data.
3857  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3858  * its inmemory API.
3859  */
3860 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3861                                          const struct lu_fid *fid)
3862 {
3863         if (!fid_is_namespace_visible(fid) ||
3864             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3865                 param->edp_magic = 0;
3866                 return;
3867         }
3868
3869         param->edp_magic = LDISKFS_LUFID_MAGIC;
3870         param->edp_len =  sizeof(struct lu_fid) + 1;
3871         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3872 }
3873
3874 /**
3875  * Try to read the fid from inode ea into dt_rec.
3876  *
3877  * \param fid object fid.
3878  *
3879  * \retval 0 on success
3880  */
3881 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3882                           __u32 ino, struct lu_fid *fid,
3883                           struct osd_inode_id *id)
3884 {
3885         struct osd_thread_info *info  = osd_oti_get(env);
3886         struct inode *inode;
3887
3888         ENTRY;
3889
3890         osd_id_gen(id, ino, OSD_OII_NOGEN);
3891         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3892         if (IS_ERR(inode))
3893                 RETURN(PTR_ERR(inode));
3894
3895         iput(inode);
3896         RETURN(0);
3897 }
3898
3899 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3900                                         struct inode *dir,
3901                                         struct inode *parent_dir,
3902                                         const struct lu_fid *dot_fid,
3903                                         const struct lu_fid *dot_dot_fid,
3904                                         struct osd_thandle *oth)
3905 {
3906         struct ldiskfs_dentry_param *dot_ldp;
3907         struct ldiskfs_dentry_param *dot_dot_ldp;
3908         __u32 saved_nlink = dir->i_nlink;
3909         int rc;
3910
3911         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
3912                 return -ENOSPC;
3913
3914         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3915         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3916
3917         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3918         dot_ldp->edp_magic = 0;
3919
3920         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3921                                     dir, dot_ldp, dot_dot_ldp);
3922         /*
3923          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3924          * the subseqent ref_add() will increase the dir->i_nlink
3925          * as 3. That is incorrect for new created directory.
3926          *
3927          * It looks like hack, because we want to make the OSD API
3928          * to be order-independent for new created directory object
3929          * between dt_insert(..) and ref_add() operations.
3930          *
3931          * Here, we only restore the in-RAM dir-inode's nlink attr,
3932          * becuase if the nlink attr is not 2, then there will be
3933          * ref_add() called following the dt_insert(..), such call
3934          * will make both the in-RAM and on-disk dir-inode's nlink
3935          * attr to be set as 2. LU-7447
3936          */
3937         set_nlink(dir, saved_nlink);
3938         return rc;
3939 }
3940
3941 /**
3942  * Create an local agent inode for remote entry
3943  */
3944 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3945                                                   struct osd_device *osd,
3946                                                   struct osd_object *pobj,
3947                                                   const struct lu_fid *fid,
3948                                                   __u32 type,
3949                                                   struct thandle *th)
3950 {
3951         struct osd_thread_info *info = osd_oti_get(env);
3952         struct inode *local;
3953         struct osd_thandle *oh;
3954         struct iattr iattr = {
3955                 .ia_valid = ATTR_UID | ATTR_GID |
3956                             ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3957                 .ia_ctime.tv_nsec = UTIME_OMIT,
3958                 .ia_mtime.tv_nsec = UTIME_OMIT,
3959                 .ia_atime.tv_nsec = UTIME_OMIT,
3960                 .ia_uid = GLOBAL_ROOT_UID,
3961                 .ia_gid = GLOBAL_ROOT_GID,
3962         };
3963         int rc;
3964
3965         ENTRY;
3966
3967         LASSERT(th);
3968         oh = container_of(th, struct osd_thandle, ot_super);
3969         LASSERT(oh->ot_handle->h_transaction != NULL);
3970
3971         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
3972                                      type, &iattr);
3973         if (IS_ERR(local)) {
3974                 CERROR("%s: create local error %d\n", osd_name(osd),
3975                        (int)PTR_ERR(local));
3976                 RETURN(local);
3977         }
3978
3979         /*
3980          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3981          * correct gid on remote file, not agent here
3982          */
3983         local->i_gid = current_fsgid();
3984         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3985
3986         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3987          * That gives e2fsck something to look at and be happy, and allows
3988          * debugging if we need to determine where this symlink came from.
3989          */
3990         if (S_ISLNK(type)) {
3991                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3992                 rc = scnprintf((char *)LDISKFS_I(local)->i_data,
3993                                LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3994
3995                 i_size_write(local, rc);
3996                 LDISKFS_I(local)->i_disksize = rc;
3997         }
3998         unlock_new_inode(local);
3999
4000         /* Agent inode should not have project ID */
4001 #ifdef  HAVE_PROJECT_QUOTA
4002         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
4003             i_projid_read(pobj->oo_inode) != 0) {
4004                 rc = osd_transfer_project(local, 0, th);
4005                 if (rc) {
4006                         CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
4007                                osd_ino2name(local), rc);
4008                         RETURN(ERR_PTR(rc));
4009                 }
4010         }
4011 #endif
4012         /* Set special LMA flag for local agent inode */
4013         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
4014         if (rc != 0) {
4015                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
4016                        osd_name(osd), PFID(fid), rc);
4017                 RETURN(ERR_PTR(rc));
4018         }
4019
4020         if (!S_ISDIR(type))
4021                 RETURN(local);
4022
4023         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
4024                                          lu_object_fid(&pobj->oo_dt.do_lu),
4025                                          fid, oh);
4026         if (rc != 0) {
4027                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
4028                         osd_name(osd), PFID(fid), rc);
4029                 RETURN(ERR_PTR(rc));
4030         }
4031
4032         RETURN(local);
4033 }
4034
4035 /**
4036  * when direntry is deleted, we have to take care of possible agent inode
4037  * referenced by that. unfortunately we can't do this at that point:
4038  * iget() within a running transaction leads to deadlock and we better do
4039  * not call that every delete declaration to save performance. so we put
4040  * a potention agent inode on a list and process that once the transaction
4041  * is over. Notice it's not any worse in terms of real orphans as regular
4042  * object destroy doesn't put inodes on the on-disk orphan list. this should
4043  * be addressed separately
4044  */
4045 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
4046                                             struct osd_thandle *oh,
4047                                             __u32 ino)
4048 {
4049         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
4050         struct osd_obj_orphan *oor;
4051
4052         OBD_ALLOC_PTR(oor);
4053         if (oor == NULL)
4054                 return -ENOMEM;
4055
4056         oor->oor_ino = ino;
4057         oor->oor_env = (struct lu_env *)env;
4058         spin_lock(&osd->od_osfs_lock);
4059         list_add(&oor->oor_list, &osd->od_orphan_list);
4060         spin_unlock(&osd->od_osfs_lock);
4061
4062         oh->ot_remove_agents = 1;
4063
4064         return 0;
4065
4066 }
4067
4068 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
4069                                                 struct osd_device *osd)
4070 {
4071         struct osd_thread_info *info = osd_oti_get(env);
4072         struct osd_obj_orphan *oor, *tmp;
4073         struct osd_inode_id id;
4074         LIST_HEAD(list);
4075         struct inode *inode;
4076         struct lu_fid fid;
4077         handle_t *jh;
4078         __u32 ino;
4079
4080         spin_lock(&osd->od_osfs_lock);
4081         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4082                 if (oor->oor_env == env)
4083                         list_move(&oor->oor_list, &list);
4084         }
4085         spin_unlock(&osd->od_osfs_lock);
4086
4087         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4088
4089                 ino = oor->oor_ino;
4090
4091                 list_del(&oor->oor_list);
4092                 OBD_FREE_PTR(oor);
4093
4094                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4095                 inode = osd_iget_fid(info, osd, &id, &fid);
4096                 if (IS_ERR(inode))
4097                         continue;
4098
4099                 if (!osd_remote_fid(env, osd, &fid)) {
4100                         iput(inode);
4101                         continue;
4102                 }
4103
4104                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4105                 clear_nlink(inode);
4106                 mark_inode_dirty(inode);
4107                 ldiskfs_journal_stop(jh);
4108                 iput(inode);
4109         }
4110
4111         return 0;
4112 }
4113
4114 /**
4115  * OSD layer object create function for OST objects (b=11826).
4116  *
4117  * The FID is inserted into inode xattr here.
4118  *
4119  * \retval   0, on success
4120  * \retval -ve, on error
4121  */
4122 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4123                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4124                       struct dt_object_format *dof, struct thandle *th)
4125 {
4126         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4127         struct osd_object *obj = osd_dt_obj(dt);
4128         struct osd_thread_info *info = osd_oti_get(env);
4129         int result, on_ost = 0;
4130
4131         ENTRY;
4132
4133         if (dt_object_exists(dt))
4134                 RETURN(-EEXIST);
4135
4136         LINVRNT(osd_invariant(obj));
4137         LASSERT(!dt_object_remote(dt));
4138         LASSERT(osd_is_write_locked(env, obj));
4139         LASSERT(th != NULL);
4140
4141         if (unlikely(fid_is_acct(fid)))
4142                 /*
4143                  * Quota files can't be created from the kernel any more,
4144                  * 'tune2fs -O quota' will take care of creating them
4145                  */
4146                 RETURN(-EPERM);
4147
4148         result = __osd_create(info, obj, attr, hint, dof, th);
4149         if (result == 0) {
4150                 if (fid_is_idif(fid) &&
4151                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4152                         struct lu_fid *tfid = &info->oti_fid;
4153                         struct ost_id *oi   = &info->oti_ostid;
4154
4155                         fid_to_ostid(fid, oi);
4156                         ostid_to_fid(tfid, oi, 0);
4157                         on_ost = 1;
4158                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4159                                                 LMAC_FID_ON_OST, 0);
4160                 } else {
4161                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4162                                                fid, OI_CHECK_FLD);
4163                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4164                                                 on_ost ? LMAC_FID_ON_OST : 0,
4165                                                 0);
4166                 }
4167                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4168                         obj->oo_dt.do_body_ops = &osd_body_ops;
4169         }
4170
4171         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
4172                 struct inode *inode = obj->oo_inode;
4173
4174                 result = __osd_oi_insert(env, obj, fid, th);
4175                 if (result && inode) {
4176                         spin_lock(&obj->oo_guard);
4177                         clear_nlink(inode);
4178                         spin_unlock(&obj->oo_guard);
4179                         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4180                         ldiskfs_set_inode_state(inode,
4181                                                 LDISKFS_STATE_LUSTRE_DESTROY);
4182                         iput(inode);
4183                         obj->oo_inode = NULL;
4184                 }
4185         }
4186
4187         /*
4188          * a small optimization - dt_insert() isn't usually applied
4189          * to OST objects, so we don't need to cache OI mapping for
4190          * OST objects
4191          */
4192         if (result == 0 && on_ost == 0) {
4193                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4194
4195                 result = osd_idc_find_and_init(env, osd, obj);
4196                 LASSERT(result == 0);
4197         }
4198
4199         LASSERT(ergo(result == 0,
4200                      dt_object_exists(dt) && !dt_object_remote(dt)));
4201         LINVRNT(osd_invariant(obj));
4202         RETURN(result);
4203 }
4204
4205 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4206                                struct thandle *handle)
4207 {
4208         struct osd_thandle *oh;
4209         int rc;
4210
4211         /* it's possible that object doesn't exist yet */
4212         LASSERT(handle != NULL);
4213
4214         oh = container_of(handle, struct osd_thandle, ot_super);
4215         LASSERT(oh->ot_handle == NULL);
4216
4217         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4218                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4219
4220         rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
4221                                    osd_dt_obj(dt));
4222
4223         return rc;
4224 }
4225
4226 /*
4227  * Concurrency: @dt is write locked.
4228  */
4229 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4230                        struct thandle *th)
4231 {
4232         struct osd_object *obj = osd_dt_obj(dt);
4233         struct inode *inode = obj->oo_inode;
4234         struct osd_thandle *oh;
4235         int rc = 0;
4236
4237         if (!dt_object_exists(dt) || obj->oo_destroyed)
4238                 return -ENOENT;
4239
4240         LINVRNT(osd_invariant(obj));
4241         LASSERT(!dt_object_remote(dt));
4242         LASSERT(osd_is_write_locked(env, obj));
4243         LASSERT(th != NULL);
4244
4245         oh = container_of(th, struct osd_thandle, ot_super);
4246         LASSERT(oh->ot_handle != NULL);
4247
4248         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4249
4250         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4251                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4252         /*
4253          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4254          * (65000) subdirectories by storing "1" in i_nlink if the link count
4255          * would otherwise overflow. Directory tranversal tools understand
4256          * that (st_nlink == 1) indicates that the filesystem dose not track
4257          * hard links count on the directory, and will not abort subdirectory
4258          * scanning early once (st_nlink - 2) subdirs have been found.
4259          *
4260          * This also has to properly handle the case of inodes with nlink == 0
4261          * in case they are being linked into the PENDING directory
4262          */
4263         spin_lock(&obj->oo_guard);
4264         if (unlikely(inode->i_nlink == 0))
4265                 /* inc_nlink from 0 may cause WARN_ON */
4266                 set_nlink(inode, 1);
4267         else {
4268                 osd_ldiskfs_inc_count(oh->ot_handle, inode);
4269                 if (!S_ISDIR(inode->i_mode))
4270                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4271         }
4272         spin_unlock(&obj->oo_guard);
4273
4274         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4275         LINVRNT(osd_invariant(obj));
4276
4277         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4278
4279         return rc;
4280 }
4281
4282 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4283                                struct thandle *handle)
4284 {
4285         struct osd_thandle *oh;
4286
4287         if (!dt_object_exists(dt))
4288                 return -ENOENT;
4289
4290         LASSERT(!dt_object_remote(dt));
4291         LASSERT(handle != NULL);
4292
4293         oh = container_of(handle, struct osd_thandle, ot_super);
4294         LASSERT(oh->ot_handle == NULL);
4295
4296         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4297                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4298
4299         return 0;
4300 }
4301
4302 /*
4303  * Concurrency: @dt is write locked.
4304  */
4305 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4306                        struct thandle *th)
4307 {
4308         struct osd_object *obj = osd_dt_obj(dt);
4309         struct inode *inode = obj->oo_inode;
4310         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4311         struct osd_thandle *oh;
4312
4313         if (!dt_object_exists(dt))
4314                 return -ENOENT;
4315
4316         LINVRNT(osd_invariant(obj));
4317         LASSERT(!dt_object_remote(dt));
4318         LASSERT(osd_is_write_locked(env, obj));
4319         LASSERT(th != NULL);
4320
4321         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4322                 return -EIO;
4323
4324         oh = container_of(th, struct osd_thandle, ot_super);
4325         LASSERT(oh->ot_handle != NULL);
4326
4327         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4328
4329         spin_lock(&obj->oo_guard);
4330         /*
4331          * That can be result of upgrade from old Lustre version and
4332          * applied only to local files.  Just skip this ref_del call.
4333          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4334          */
4335         if (inode->i_nlink == 0) {
4336                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4337                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4338                              ", maybe an upgraded file? (LU-3915)\n",
4339                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4340                 spin_unlock(&obj->oo_guard);
4341                 return 0;
4342         }
4343
4344         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4345                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4346
4347         osd_ldiskfs_dec_count(oh->ot_handle, inode);
4348         spin_unlock(&obj->oo_guard);
4349
4350         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4351         LINVRNT(osd_invariant(obj));
4352
4353         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4354
4355         return 0;
4356 }
4357
4358 /*
4359  * Concurrency: @dt is read locked.
4360  */
4361 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4362                          struct lu_buf *buf, const char *name)
4363 {
4364         struct osd_object      *obj    = osd_dt_obj(dt);
4365         struct inode           *inode  = obj->oo_inode;
4366         struct osd_thread_info *info   = osd_oti_get(env);
4367         struct dentry          *dentry = &info->oti_obj_dentry;
4368         bool cache_xattr = false;
4369         int rc;
4370
4371         LASSERT(buf);
4372
4373         /* version get is not real XATTR but uses xattr API */
4374         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4375                 dt_obj_version_t *ver = buf->lb_buf;
4376
4377                 /*
4378                  * for version we are just using xattr API but change inode
4379                  * field instead
4380                  */
4381                 if (buf->lb_len == 0)
4382                         return sizeof(dt_obj_version_t);
4383
4384                 if (buf->lb_len < sizeof(dt_obj_version_t))
4385                         return -ERANGE;
4386
4387                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4388                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4389
4390                 *ver = LDISKFS_I(inode)->i_fs_version;
4391
4392                 return sizeof(dt_obj_version_t);
4393         }
4394
4395         if (!dt_object_exists(dt))
4396                 return -ENOENT;
4397
4398         LASSERT(!dt_object_remote(dt));
4399         LASSERT(inode->i_op != NULL);
4400 #ifdef HAVE_IOP_XATTR
4401         LASSERT(inode->i_op->getxattr != NULL);
4402 #endif
4403
4404         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4405             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4406                 cache_xattr = true;
4407
4408         if (cache_xattr) {
4409                 rc = osd_oxc_get(obj, name, buf);
4410                 if (rc != -ENOENT)
4411                         return rc;
4412         }
4413
4414         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4415                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4416                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4417                 struct filter_fid *ff;
4418                 struct ost_layout *ol;
4419
4420                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4421                 if (rc)
4422                         return rc;
4423
4424                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4425
4426                 rc = sizeof(*ff);
4427                 if (buf->lb_len == 0 || !buf->lb_buf)
4428                         return rc;
4429
4430                 if (buf->lb_len < rc)
4431                         return -ERANGE;
4432
4433                 ff = buf->lb_buf;
4434                 ol = &ff->ff_layout;
4435                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4436                                                   PFID_STRIPE_IDX_BITS);
4437                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4438                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4439                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4440                 if (lma->lma_compat & LMAC_COMP_INFO) {
4441                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4442                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4443                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4444                 } else {
4445                         ol->ol_comp_start = 0;
4446                         ol->ol_comp_end = 0;
4447                         ol->ol_comp_id = 0;
4448                 }
4449         } else {
4450                 /* Get enc context xattr directly from ldiskfs instead of going
4451                  * through the VFS, as there is no xattr handler for
4452                  * "encryption.".
4453                  */
4454                 if (strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0)
4455                         rc = ldiskfs_xattr_get(inode,
4456                                           LDISKFS_XATTR_INDEX_ENCRYPTION,
4457                                           LDISKFS_XATTR_NAME_ENCRYPTION_CONTEXT,
4458                                           buf->lb_buf, buf->lb_len);
4459                 else
4460                         rc = __osd_xattr_get(inode, dentry, name,
4461                                              buf->lb_buf, buf->lb_len);
4462         }
4463
4464         if (cache_xattr) {
4465                 if (rc == -ENOENT || rc == -ENODATA)
4466                         osd_oxc_add(obj, name, NULL, 0);
4467                 else if (rc > 0 && buf->lb_buf != NULL)
4468                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4469         }
4470
4471         return rc;
4472 }
4473
4474 static int osd_declare_xattr_set(const struct lu_env *env,
4475                                  struct dt_object *dt,
4476                                  const struct lu_buf *buf, const char *name,
4477                                  int fl, struct thandle *handle)
4478 {
4479         struct osd_thandle *oh;
4480         int credits = 0;
4481         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4482
4483         LASSERT(handle != NULL);
4484
4485         oh = container_of(handle, struct osd_thandle, ot_super);
4486         LASSERT(oh->ot_handle == NULL);
4487
4488         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4489                 /*
4490                  * For non-upgrading case, the LMA is set first and
4491                  * usually fit inode. But for upgrade case, the LMA
4492                  * may be in another separated EA block.
4493                  */
4494                 if (dt_object_exists(dt)) {
4495                         if (fl == LU_XATTR_REPLACE)
4496                                 credits = 1;
4497                         else
4498                                 goto upgrade;
4499                 }
4500         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4501                 credits = 1;
4502         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4503                 /* We may need to delete the old PFID EA. */
4504                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4505                 if (fl == LU_XATTR_REPLACE)
4506                         credits += 1;
4507                 else
4508                         goto upgrade;
4509         } else {
4510                 /*
4511                  * If some name entry resides on remote MDT, then will create
4512                  * agent entry under remote parent. On the other hand, if the
4513                  * remote entry will be removed, then related agent entry may
4514                  * need to be removed from the remote parent. So there may be
4515                  * kinds of cases, let's declare enough credits. The credits
4516                  * for create agent entry is enough for remove case.
4517                  */
4518                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4519                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4520                         if (dt_object_exists(dt))
4521                                 credits += 1; /* For updating LMA */
4522                 }
4523
4524 upgrade:
4525                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4526
4527                 if (buf != NULL) {
4528                         ssize_t buflen;
4529
4530                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4531                                 /*
4532                                  * learn xattr size from osd_xattr_get if
4533                                  * attribute has not been read yet
4534                                  */
4535                                 buflen = __osd_xattr_get(
4536                                     osd_dt_obj(dt)->oo_inode,
4537                                     &osd_oti_get(env)->oti_obj_dentry,
4538                                     name, NULL, 0);
4539                                 if (buflen < 0)
4540                                         buflen = 0;
4541                         } else {
4542                                 buflen = buf->lb_len;
4543                         }
4544
4545                         if (buflen > sb->s_blocksize) {
4546                                 credits += osd_calc_bkmap_credits(
4547                                     sb, NULL, 0, -1,
4548                                     (buflen + sb->s_blocksize - 1) >>
4549                                     sb->s_blocksize_bits);
4550                         }
4551                 }
4552                 /*
4553                  * xattr set may involve inode quota change, reserve credits for
4554                  * dquot_initialize()
4555                  */
4556                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4557         }
4558
4559         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4560
4561         return 0;
4562 }
4563
4564 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4565                               const struct lu_buf *buf, int fl,
4566                               struct thandle *handle)
4567 {
4568         struct osd_thread_info *info = osd_oti_get(env);
4569         struct dentry *dentry = &info->oti_obj_dentry;
4570         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4571         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4572         struct inode *inode = obj->oo_inode;
4573         struct filter_fid *ff = buf->lb_buf;
4574         struct ost_layout *ol = &ff->ff_layout;
4575         int flags = XATTR_REPLACE;
4576         int rc;
4577
4578         ENTRY;
4579
4580         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4581                 RETURN(-EINVAL);
4582
4583         rc = osd_get_lma(info, inode, dentry, loa);
4584         if (rc == -ENODATA) {
4585                 /* Usually for upgarding from old device */
4586                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4587                                 LMAC_FID_ON_OST, 0);
4588                 flags = XATTR_CREATE;
4589         } else if (rc) {
4590                 RETURN(rc);
4591         }
4592
4593         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4594                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4595                         RETURN(-EEXIST);
4596
4597                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4598                         /* Separate PFID EA from LMA */
4599                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4600                         lustre_lma_swab(lma);
4601                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4602                                              sizeof(*lma), XATTR_REPLACE);
4603                         if (!rc) {
4604                                 obj->oo_pfid_in_lma = 0;
4605                                 rc = LU_XATTR_CREATE;
4606                         }
4607
4608                         RETURN(rc);
4609                 }
4610         } else {
4611                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4612                         RETURN(fl);
4613
4614                 /*
4615                  * Old client does not send stripe information,
4616                  * then store the PFID EA on disk separatedly.
4617                  */
4618                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4619                              ol->ol_stripe_size == 0))
4620                         RETURN(fl);
4621
4622                 /* Remove old PFID EA entry firstly. */
4623                 dquot_initialize(inode);
4624                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4625                 if (rc == -ENODATA) {
4626                         /* XATTR_NAME_FID is already absent */
4627                         rc = 0;
4628                 } else if (rc) {
4629                         RETURN(rc);
4630                 }
4631         }
4632
4633         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4634         if (likely(ol->ol_stripe_size != 0)) {
4635                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4636                                              PFID_STRIPE_IDX_BITS;
4637                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4638                 lma->lma_compat |= LMAC_STRIPE_INFO;
4639                 if (ol->ol_comp_id != 0) {
4640                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4641                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4642                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4643                         lma->lma_compat |= LMAC_COMP_INFO;
4644                 }
4645         }
4646
4647         lustre_loa_swab(loa, false);
4648
4649         /* Store the PFID EA inside LMA. */
4650         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4651                              flags);
4652         if (!rc)
4653                 obj->oo_pfid_in_lma = 1;
4654
4655         RETURN(rc);
4656 }
4657
4658 /*
4659  * In DNE environment, the object (in spite of regular file or directory)
4660  * and its name entry may reside on different MDTs. Under such case, we will
4661  * create an agent entry on the MDT where the object resides. The agent entry
4662  * references the object locally, that makes the object to be visible to the
4663  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4664  * such as 'tar' can handle the object properly.
4665  *
4666  * We handle the agent entry during set linkEA that is the common interface
4667  * for both regular file and directroy, can handle kinds of cases, such as
4668  * create/link/unlink/rename, and so on.
4669  *
4670  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4671  *
4672  * XXX: There are two known issues:
4673  * 1. For one object, we will create at most one agent entry even if there
4674  *    may be more than one cross-MDTs hard links on the object. So the local
4675  *    e2fsck may claim that the object's nlink is larger than the name entries
4676  *    that reference such inode. And in further, the e2fsck will fix the nlink
4677  *    attribute to match the local references. Then it will cause the object's
4678  *    nlink attribute to be inconsistent with the global references. it is bad
4679  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4680  *    other hand, the global namespace LFSCK can repair the object's attribute
4681  *    according to the linkEA.
4682  * 2. There may be too many hard links on the object as to its linkEA overflow,
4683  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4684  *    case happened, then at this point, we do not know whether there are some
4685  *    cross-MDTs reference. But there are local references, it guarantees that
4686  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4687  */
4688 static int osd_xattr_handle_linkea(const struct lu_env *env,
4689                                    struct osd_device *osd,
4690                                    struct osd_object *obj,
4691                                    const struct lu_buf *buf,
4692                                    struct thandle *handle)
4693 {
4694         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4695         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4696         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4697         struct lu_name tmpname;
4698         struct osd_thandle *oh;
4699         int rc;
4700         bool remote = false;
4701
4702         ENTRY;
4703
4704         oh = container_of(handle, struct osd_thandle, ot_super);
4705         LASSERT(oh->ot_handle != NULL);
4706
4707         rc = linkea_init_with_rec(&ldata);
4708         if (!rc) {
4709                 linkea_first_entry(&ldata);
4710                 while (ldata.ld_lee != NULL && !remote) {
4711                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4712                                             &tmpname, tfid);
4713                         if (osd_remote_fid(env, osd, tfid) > 0)
4714                                 remote = true;
4715                         else
4716                                 linkea_next_entry(&ldata);
4717                 }
4718         } else if (rc == -ENODATA) {
4719                 rc = 0;
4720         } else {
4721                 RETURN(rc);
4722         }
4723
4724         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4725                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4726                 if (rc)
4727                         CERROR("%s: failed to remove agent entry for "DFID
4728                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4729         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4730                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4731                 if (rc)
4732                         CERROR("%s: failed to create agent entry for "DFID
4733                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4734         }
4735
4736         RETURN(rc);
4737 }
4738
4739 /*
4740  * Concurrency: @dt is write locked.
4741  */
4742 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4743                          const struct lu_buf *buf, const char *name, int fl,
4744                          struct thandle *handle)
4745 {
4746         struct osd_object *obj = osd_dt_obj(dt);
4747         struct osd_device *osd = osd_obj2dev(obj);
4748         struct inode *inode = obj->oo_inode;
4749         struct osd_thread_info *info = osd_oti_get(env);
4750         int fs_flags = 0;
4751         int len;
4752         int rc;
4753
4754         ENTRY;
4755
4756         LASSERT(handle);
4757         LASSERT(buf);
4758
4759         /* version set is not real XATTR */
4760         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4761                 dt_obj_version_t *version = buf->lb_buf;
4762
4763                 /*
4764                  * for version we are just using xattr API but change inode
4765                  * field instead
4766                  */
4767                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4768
4769                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4770                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4771
4772                 LDISKFS_I(inode)->i_fs_version = *version;
4773                 /*
4774                  * Version is set after all inode operations are finished,
4775                  * so we should mark it dirty here
4776                  */
4777                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4778
4779                 RETURN(0);
4780         }
4781
4782         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4783                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4784
4785         len = buf->lb_len;
4786         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4787
4788         /*
4789          * For the OST device with 256 bytes inode size by default,
4790          * the PFID EA will be stored together with LMA EA to avoid
4791          * performance trouble. Otherwise the PFID EA can be stored
4792          * independently. LU-8998
4793          */
4794         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4795             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4796                 LASSERT(buf->lb_buf);
4797
4798                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4799                 if (fl <= 0)
4800                         RETURN(fl);
4801         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4802                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4803                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4804
4805                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4806                 if (rc)
4807                         RETURN(rc);
4808
4809                 lma->lma_incompat |= LMAI_STRIPED;
4810                 lustre_lma_swab(lma);
4811                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4812                                      sizeof(*lma), XATTR_REPLACE);
4813                 if (rc != 0)
4814                         RETURN(rc);
4815         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4816                 LASSERT(!osd->od_is_ost);
4817
4818                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4819                 if (rc)
4820                         RETURN(rc);
4821         }
4822
4823         if (fl & LU_XATTR_REPLACE)
4824                 fs_flags |= XATTR_REPLACE;
4825
4826         if (fl & LU_XATTR_CREATE)
4827                 fs_flags |= XATTR_CREATE;
4828
4829         if (strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0) {
4830                 /* Set enc context xattr directly in ldiskfs instead of going
4831                  * through the VFS, as there is no xattr handler for
4832                  * "encryption.".
4833                  */
4834                 struct osd_thandle *oth = container_of(handle,
4835                                                        struct osd_thandle,
4836                                                        ot_super);
4837
4838                 if (!oth->ot_handle)
4839                         /* this should be already part of a transaction */
4840                         RETURN(-EPROTO);
4841
4842                 rc = ldiskfs_xattr_set_handle(oth->ot_handle, inode,
4843                                           LDISKFS_XATTR_INDEX_ENCRYPTION,
4844                                           LDISKFS_XATTR_NAME_ENCRYPTION_CONTEXT,
4845                                           buf->lb_buf, len, fs_flags);
4846         } else {
4847                 rc = __osd_xattr_set(info, inode, name,
4848                                      buf->lb_buf, len, fs_flags);
4849         }
4850         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4851
4852         if (rc == 0 &&
4853             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4854              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4855                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4856
4857         return rc;
4858 }
4859
4860 /*
4861  * Concurrency: @dt is read locked.
4862  */
4863 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4864                           const struct lu_buf *buf)
4865 {
4866         struct osd_object *obj = osd_dt_obj(dt);
4867         struct osd_device *dev = osd_obj2dev(obj);
4868         struct inode *inode = obj->oo_inode;
4869         struct osd_thread_info *info = osd_oti_get(env);
4870         struct dentry *dentry = &info->oti_obj_dentry;
4871         int rc;
4872
4873         if (!dt_object_exists(dt))
4874                 return -ENOENT;
4875
4876         LASSERT(!dt_object_remote(dt));
4877         LASSERT(inode->i_op != NULL);
4878         LASSERT(inode->i_op->listxattr != NULL);
4879
4880         dentry->d_inode = inode;
4881         dentry->d_sb = inode->i_sb;
4882         rc = inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4883
4884         if (rc < 0 || buf->lb_buf == NULL)
4885                 return rc;
4886
4887         /* Hide virtual project ID xattr from list if disabled */
4888         if (!dev->od_enable_projid_xattr) {
4889                 char *end = (char *)buf->lb_buf + rc;
4890                 char *p = buf->lb_buf;
4891
4892                 while (p < end) {
4893                         char *next = p + strlen(p) + 1;
4894
4895                         if (strcmp(p, XATTR_NAME_PROJID) == 0) {
4896                                 if (end - next > 0)
4897                                         memmove(p, next, end - next);
4898                                 rc -= next - p;
4899                                 break;
4900                         }
4901
4902                         p = next;
4903                 }
4904         }
4905
4906         return rc;
4907 }
4908
4909 static int osd_declare_xattr_del(const struct lu_env *env,
4910                                  struct dt_object *dt, const char *name,
4911                                  struct thandle *handle)
4912 {
4913         struct osd_thandle *oh;
4914         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4915
4916         LASSERT(!dt_object_remote(dt));
4917         LASSERT(handle != NULL);
4918
4919         oh = container_of(handle, struct osd_thandle, ot_super);
4920         LASSERT(oh->ot_handle == NULL);
4921
4922         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4923                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4924         /*
4925          * xattr del may involve inode quota change, reserve credits for
4926          * dquot_initialize()
4927          */
4928         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4929
4930         return 0;
4931 }
4932
4933 /*
4934  * Concurrency: @dt is write locked.
4935  */
4936 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4937                          const char *name, struct thandle *handle)
4938 {
4939         struct osd_object *obj = osd_dt_obj(dt);
4940         struct inode *inode = obj->oo_inode;
4941         struct osd_thread_info *info = osd_oti_get(env);
4942         struct dentry *dentry = &info->oti_obj_dentry;
4943         int rc;
4944
4945         if (!dt_object_exists(dt))
4946                 return -ENOENT;
4947
4948         LASSERT(!dt_object_remote(dt));
4949         LASSERT(inode->i_op != NULL);
4950         LASSERT(handle != NULL);
4951 #ifdef HAVE_IOP_XATTR
4952         LASSERT(inode->i_op->removexattr != NULL);
4953 #endif
4954
4955         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4956
4957         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4958                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4959
4960                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4961                                  &info->oti_ost_attrs);
4962                 if (!rc) {
4963                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4964
4965                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4966                         lustre_lma_swab(lma);
4967                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4968                                              sizeof(*lma), XATTR_REPLACE);
4969                         if (!rc)
4970                                 obj->oo_pfid_in_lma = 0;
4971                 }
4972         } else {
4973                 dquot_initialize(inode);
4974                 dentry->d_inode = inode;
4975                 dentry->d_sb = inode->i_sb;
4976                 rc = ll_vfs_removexattr(dentry, inode, name);
4977         }
4978
4979         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4980
4981         if (rc == 0 &&
4982             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4983              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4984                 osd_oxc_del(obj, name);
4985
4986         return rc;
4987 }
4988
4989 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4990                            __u64 start, __u64 end)
4991 {
4992         struct osd_object *obj = osd_dt_obj(dt);
4993         struct osd_device *dev = osd_obj2dev(obj);
4994         struct inode *inode = obj->oo_inode;
4995         struct file *file;
4996         int rc;
4997
4998         ENTRY;
4999         file = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
5000                                  inode->i_fop);
5001         if (IS_ERR(file))
5002                 RETURN(PTR_ERR(file));
5003
5004         file->f_mode |= FMODE_64BITHASH;
5005         rc = vfs_fsync_range(file, start, end, 0);
5006         ihold(inode);
5007         fput(file);
5008
5009         RETURN(rc);
5010 }
5011
5012 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
5013 {
5014         return 0;
5015 }
5016
5017 static bool osd_check_stale(struct dt_object *dt)
5018 {
5019         return false;
5020 }
5021 /*
5022  * Index operations.
5023  */
5024
5025 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
5026                                const struct dt_index_features *feat)
5027 {
5028         struct iam_descr *descr;
5029
5030         if (osd_object_is_root(o))
5031                 return feat == &dt_directory_features;
5032
5033         LASSERT(o->oo_dir != NULL);
5034
5035         descr = o->oo_dir->od_container.ic_descr;
5036         if (feat == &dt_directory_features) {
5037                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
5038                         return 1;
5039                 else
5040                         return 0;
5041         } else {
5042                 return feat->dif_keysize_min <= descr->id_key_size &&
5043                        descr->id_key_size <= feat->dif_keysize_max &&
5044                        feat->dif_recsize_min <= descr->id_rec_size &&
5045                        descr->id_rec_size <= feat->dif_recsize_max &&
5046                        !(feat->dif_flags & (DT_IND_VARKEY |
5047                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
5048                        ergo(feat->dif_flags & DT_IND_UPDATE,
5049                             1 /* XXX check that object (and fs) is writable */);
5050         }
5051 }
5052
5053 static int osd_iam_container_init(const struct lu_env *env,
5054                                   struct osd_object *obj,
5055                                   struct osd_directory *dir)
5056 {
5057         struct iam_container *bag = &dir->od_container;
5058         int result;
5059
5060         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
5061         if (result != 0)
5062                 return result;
5063
5064         result = iam_container_setup(bag);
5065         if (result == 0)
5066                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
5067         else
5068                 iam_container_fini(bag);
5069
5070         return result;
5071 }
5072
5073
5074 /*
5075  * Concurrency: no external locking is necessary.
5076  */
5077 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
5078                          const struct dt_index_features *feat)
5079 {
5080         int result;
5081         int skip_iam = 0;
5082         struct osd_object *obj = osd_dt_obj(dt);
5083
5084         LINVRNT(osd_invariant(obj));
5085
5086         if (osd_object_is_root(obj)) {
5087                 dt->do_index_ops = &osd_index_ea_ops;
5088                 result = 0;
5089         } else if (feat == &dt_directory_features) {
5090                 dt->do_index_ops = &osd_index_ea_ops;
5091                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
5092                         result = 0;
5093                 else
5094                         result = -ENOTDIR;
5095                 skip_iam = 1;
5096         } else if (unlikely(feat == &dt_otable_features)) {
5097                 dt->do_index_ops = &osd_otable_ops;
5098                 return 0;
5099         } else if (unlikely(feat == &dt_acct_features)) {
5100                 dt->do_index_ops = &osd_acct_index_ops;
5101                 result = 0;
5102                 skip_iam = 1;
5103         } else if (!osd_has_index(obj)) {
5104                 struct osd_directory *dir;
5105                 struct osd_device *osd = osd_obj2dev(obj);
5106                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
5107
5108                 OBD_ALLOC_PTR(dir);
5109                 if (dir) {
5110
5111                         spin_lock(&obj->oo_guard);
5112                         if (obj->oo_dir == NULL)
5113                                 obj->oo_dir = dir;
5114                         else
5115                                 /*
5116                                  * Concurrent thread allocated container data.
5117                                  */
5118                                 OBD_FREE_PTR(dir);
5119                         spin_unlock(&obj->oo_guard);
5120                         /*
5121                          * Now, that we have container data, serialize its
5122                          * initialization.
5123                          */
5124                         down_write(&obj->oo_ext_idx_sem);
5125                         /*
5126                          * recheck under lock.
5127                          */
5128
5129                         if (osd_has_index(obj)) {
5130                                 result = 0;
5131                                 goto unlock;
5132                         }
5133
5134                         result = osd_iam_container_init(env, obj, obj->oo_dir);
5135                         if (result || feat == &dt_lfsck_namespace_features ||
5136                             feat == &dt_lfsck_layout_orphan_features ||
5137                             feat == &dt_lfsck_layout_dangling_features)
5138                                 goto unlock;
5139
5140                         result = osd_index_register(osd, fid,
5141                                                     feat->dif_keysize_max,
5142                                                     feat->dif_recsize_max);
5143                         if (result < 0)
5144                                 CWARN("%s: failed to register index "
5145                                       DFID": rc = %d\n",
5146                                       osd_name(osd), PFID(fid), result);
5147                         else if (result > 0)
5148                                 result = 0;
5149                         else
5150                                 CDEBUG(D_LFSCK, "%s: index object "DFID
5151                                        " (%d/%d) registered\n",
5152                                        osd_name(osd), PFID(fid),
5153                                        (int)feat->dif_keysize_max,
5154                                        (int)feat->dif_recsize_max);
5155
5156 unlock:
5157                         up_write(&obj->oo_ext_idx_sem);
5158                 } else {
5159                         result = -ENOMEM;
5160                 }
5161         } else {
5162                 result = 0;
5163         }
5164
5165         if (result == 0 && skip_iam == 0) {
5166                 if (!osd_iam_index_probe(env, obj, feat))
5167                         result = -ENOTDIR;
5168         }
5169         LINVRNT(osd_invariant(obj));
5170
5171         return result;
5172 }
5173
5174 static int osd_otable_it_attr_get(const struct lu_env *env,
5175                                  struct dt_object *dt,
5176                                  struct lu_attr *attr)
5177 {
5178         attr->la_valid = 0;
5179         return 0;
5180 }
5181
5182 static const struct dt_object_operations osd_obj_ops = {
5183         .do_read_lock           = osd_read_lock,
5184         .do_write_lock          = osd_write_lock,
5185         .do_read_unlock         = osd_read_unlock,
5186         .do_write_unlock        = osd_write_unlock,
5187         .do_write_locked        = osd_write_locked,
5188         .do_attr_get            = osd_attr_get,
5189         .do_declare_attr_set    = osd_declare_attr_set,
5190         .do_attr_set            = osd_attr_set,
5191         .do_ah_init             = osd_ah_init,
5192         .do_declare_create      = osd_declare_create,
5193         .do_create              = osd_create,
5194         .do_declare_destroy     = osd_declare_destroy,
5195         .do_destroy             = osd_destroy,
5196         .do_index_try           = osd_index_try,
5197         .do_declare_ref_add     = osd_declare_ref_add,
5198         .do_ref_add             = osd_ref_add,
5199         .do_declare_ref_del     = osd_declare_ref_del,
5200         .do_ref_del             = osd_ref_del,
5201         .do_xattr_get           = osd_xattr_get,
5202         .do_declare_xattr_set   = osd_declare_xattr_set,
5203         .do_xattr_set           = osd_xattr_set,
5204         .do_declare_xattr_del   = osd_declare_xattr_del,
5205         .do_xattr_del           = osd_xattr_del,
5206         .do_xattr_list          = osd_xattr_list,
5207         .do_object_sync         = osd_object_sync,
5208         .do_invalidate          = osd_invalidate,
5209         .do_check_stale         = osd_check_stale,
5210 };
5211
5212 static const struct dt_object_operations osd_obj_otable_it_ops = {
5213         .do_attr_get    = osd_otable_it_attr_get,
5214         .do_index_try   = osd_index_try,
5215 };
5216
5217 static int osd_index_declare_iam_delete(const struct lu_env *env,
5218                                         struct dt_object *dt,
5219                                         const struct dt_key *key,
5220                                         struct thandle *handle)
5221 {
5222         struct osd_thandle *oh;
5223
5224         oh = container_of(handle, struct osd_thandle, ot_super);
5225         LASSERT(oh->ot_handle == NULL);
5226
5227         /* Recycle  may cause additional three blocks to be changed. */
5228         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5229                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5230
5231         return 0;
5232 }
5233
5234 /**
5235  *      delete a (key, value) pair from index \a dt specified by \a key
5236  *
5237  *      \param  dt      osd index object
5238  *      \param  key     key for index
5239  *      \param  rec     record reference
5240  *      \param  handle  transaction handler
5241  *
5242  *      \retval  0  success
5243  *      \retval -ve   failure
5244  */
5245 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5246                                 const struct dt_key *key,
5247                                 struct thandle *handle)
5248 {
5249         struct osd_thread_info *oti = osd_oti_get(env);
5250         struct osd_object *obj = osd_dt_obj(dt);
5251         struct osd_thandle *oh;
5252         struct iam_path_descr *ipd;
5253         struct iam_container *bag = &obj->oo_dir->od_container;
5254         int rc;
5255
5256         ENTRY;
5257
5258         if (!dt_object_exists(dt))
5259                 RETURN(-ENOENT);
5260
5261         LINVRNT(osd_invariant(obj));
5262         LASSERT(!dt_object_remote(dt));
5263         LASSERT(bag->ic_object == obj->oo_inode);
5264         LASSERT(handle != NULL);
5265
5266         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5267
5268         ipd = osd_idx_ipd_get(env, bag);
5269         if (unlikely(ipd == NULL))
5270                 RETURN(-ENOMEM);
5271
5272         oh = container_of(handle, struct osd_thandle, ot_super);
5273         LASSERT(oh->ot_handle != NULL);
5274         LASSERT(oh->ot_handle->h_transaction != NULL);
5275
5276         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5277                 /* swab quota uid/gid provided by caller */
5278                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5279                 key = (const struct dt_key *)&oti->oti_quota_id;
5280         }
5281
5282         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5283         osd_ipd_put(env, bag, ipd);
5284         LINVRNT(osd_invariant(obj));
5285         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5286         RETURN(rc);
5287 }
5288
5289 static int osd_index_declare_ea_delete(const struct lu_env *env,
5290                                        struct dt_object *dt,
5291                                        const struct dt_key *key,
5292                                        struct thandle *handle)
5293 {
5294         struct osd_thandle *oh;
5295         struct inode *inode;
5296         int rc, credits;
5297
5298         ENTRY;
5299
5300         LASSERT(!dt_object_remote(dt));
5301         LASSERT(handle != NULL);
5302
5303         oh = container_of(handle, struct osd_thandle, ot_super);
5304         LASSERT(oh->ot_handle == NULL);
5305
5306         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5307         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5308
5309         inode = osd_dt_obj(dt)->oo_inode;
5310         if (inode == NULL)
5311                 RETURN(-ENOENT);
5312
5313         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5314                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5315                                    NULL, OSD_QID_BLK);
5316         RETURN(rc);
5317 }
5318
5319 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5320                                           struct dt_rec *fid)
5321 {
5322         struct osd_fid_pack *rec;
5323         int rc = -ENODATA;
5324
5325         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5326                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5327                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5328                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5329                         rc = -EINVAL;
5330         }
5331         return rc;
5332 }
5333
5334 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5335                           const struct lu_fid *fid)
5336 {
5337         struct seq_server_site *ss = osd_seq_site(osd);
5338
5339         ENTRY;
5340
5341         /* FID seqs not in FLDB, must be local seq */
5342         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5343                 RETURN(0);
5344
5345         /*
5346          * If FLD is not being initialized yet, it only happens during the
5347          * initialization, likely during mgs initialization, and we assume
5348          * this is local FID.
5349          */
5350         if (ss == NULL || ss->ss_server_fld == NULL)
5351                 RETURN(0);
5352
5353         /* Only check the local FLDB here */
5354         if (osd_seq_exists(env, osd, fid_seq(fid)))
5355                 RETURN(0);
5356
5357         RETURN(1);
5358 }
5359
5360 static void osd_take_care_of_agent(const struct lu_env *env,
5361                                    struct osd_device *osd,
5362                                    struct osd_thandle *oh,
5363                                    struct ldiskfs_dir_entry_2 *de)
5364 {
5365         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5366         struct osd_idmap_cache *idc;
5367         int rc, schedule = 0;
5368
5369         LASSERT(de != NULL);
5370
5371         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5372         if (likely(rc == 0)) {
5373                 idc = osd_idc_find_or_init(env, osd, fid);
5374                 if (IS_ERR(idc) || idc->oic_remote)
5375                         schedule = 1;
5376         } else if (rc == -ENODATA) {
5377                 /*
5378                  * can't get FID, postpone to the end of the
5379                  * transaction when iget() is safe
5380                  */
5381                 schedule = 1;
5382         } else {
5383                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5384         }
5385         if (schedule)
5386                 osd_schedule_agent_inode_removal(env, oh,
5387                                                  le32_to_cpu(de->inode));
5388 }
5389
5390 /**
5391  * Utility function to get real name from object name
5392  *
5393  * \param[in] obj      pointer to the object to be handled
5394  * \param[in] name     object name
5395  * \param[in] len      object name len
5396  * \param[out]ln       pointer to the struct lu_name to hold the real name
5397  *
5398  * If file is not encrypted, real name is just the object name.
5399  * If file is encrypted, object name needs to be decoded. In
5400  * this case a new buffer is allocated, and ln->ln_name needs to be freed by
5401  * the caller.
5402  *
5403  * \retval   0, on success
5404  * \retval -ve, on error
5405  */
5406 static int obj_name2lu_name(struct osd_object *obj, const char *name,
5407                             int len, struct lu_name *ln)
5408 {
5409         if (!(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
5410                 ln->ln_name = name;
5411                 ln->ln_namelen = len;
5412         } else {
5413                 char *buf = kmalloc(len, GFP_NOFS);
5414
5415                 if (!buf)
5416                         return -ENOMEM;
5417
5418                 len = critical_decode(name, len, buf);
5419                 ln->ln_name = buf;
5420                 ln->ln_namelen = len;
5421         }
5422
5423         return 0;
5424 }
5425
5426 /**
5427  * Index delete function for interoperability mode (b11826).
5428  * It will remove the directory entry added by osd_index_ea_insert().
5429  * This entry is needed to maintain name->fid mapping.
5430  *
5431  * \param key,  key i.e. file entry to be deleted
5432  *
5433  * \retval   0, on success
5434  * \retval -ve, on error
5435  */
5436 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5437                                const struct dt_key *key, struct thandle *handle)
5438 {
5439         struct osd_object *obj = osd_dt_obj(dt);
5440         struct inode *dir = obj->oo_inode;
5441         struct dentry *dentry;
5442         struct osd_thandle *oh;
5443         struct ldiskfs_dir_entry_2 *de = NULL;
5444         struct buffer_head *bh;
5445         struct htree_lock *hlock = NULL;
5446         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5447         struct lu_name ln;
5448         int rc;
5449
5450         ENTRY;
5451
5452         if (!dt_object_exists(dt))
5453                 RETURN(-ENOENT);
5454
5455         LINVRNT(osd_invariant(obj));
5456         LASSERT(!dt_object_remote(dt));
5457         LASSERT(handle != NULL);
5458
5459         rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
5460         if (rc)
5461                 RETURN(rc);
5462
5463         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5464
5465         oh = container_of(handle, struct osd_thandle, ot_super);
5466         LASSERT(oh->ot_handle != NULL);
5467         LASSERT(oh->ot_handle->h_transaction != NULL);
5468
5469         dquot_initialize(dir);
5470         dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
5471
5472         if (obj->oo_hl_head != NULL) {
5473                 hlock = osd_oti_get(env)->oti_hlock;
5474                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5475                                    dir, LDISKFS_HLOCK_DEL);
5476         } else {
5477                 down_write(&obj->oo_ext_idx_sem);
5478         }
5479
5480         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5481         if (!IS_ERR(bh)) {
5482                 /*
5483                  * If this is not the ".." entry, it might be a remote DNE
5484                  * entry and  we need to check if the FID is for a remote
5485                  * MDT.  If the FID is  not in the directory entry (e.g.
5486                  * upgraded 1.8 filesystem without dirdata enabled) then
5487                  * we need to get the FID from the LMA. For a remote directory
5488                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5489                  * case.
5490                  *
5491                  * Delete the entry before the agent inode in order to
5492                  * simplify error handling.  At worst an error after deleting
5493                  * the entry first might leak the agent inode afterward. The
5494                  * reverse would need filesystem abort in case of error deleting
5495                  * the entry after the agent had been removed, or leave a
5496                  * dangling entry pointing at a random inode.
5497                  */
5498                 if (strcmp((char *)key, dotdot) != 0)
5499                         osd_take_care_of_agent(env, osd, oh, de);
5500                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5501                 brelse(bh);
5502         } else {
5503                 rc = PTR_ERR(bh);
5504         }
5505
5506         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5507             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5508                 /* NB, dirent count may not be accurate, because it's counted
5509                  * without lock.
5510                  */
5511                 if (obj->oo_dirent_count)
5512                         obj->oo_dirent_count--;
5513                 else
5514                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5515         }
5516         if (hlock != NULL)
5517                 ldiskfs_htree_unlock(hlock);
5518         else
5519                 up_write(&obj->oo_ext_idx_sem);
5520         GOTO(out, rc);
5521 out:
5522         LASSERT(osd_invariant(obj));
5523         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5524         if (ln.ln_name != (char *)key)
5525                 kfree(ln.ln_name);
5526         RETURN(rc);
5527 }
5528
5529 /**
5530  *      Lookup index for \a key and copy record to \a rec.
5531  *
5532  *      \param  dt      osd index object
5533  *      \param  key     key for index
5534  *      \param  rec     record reference
5535  *
5536  *      \retval  +ve  success : exact mach
5537  *      \retval  0    return record with key not greater than \a key
5538  *      \retval -ve   failure
5539  */
5540 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5541                                 struct dt_rec *rec, const struct dt_key *key)
5542 {
5543         struct osd_object *obj = osd_dt_obj(dt);
5544         struct iam_path_descr *ipd;
5545         struct iam_container *bag = &obj->oo_dir->od_container;
5546         struct osd_thread_info *oti = osd_oti_get(env);
5547         struct iam_iterator *it = &oti->oti_idx_it;
5548         struct iam_rec *iam_rec;
5549         int rc;
5550
5551         ENTRY;
5552
5553         if (!dt_object_exists(dt))
5554                 RETURN(-ENOENT);
5555
5556         LASSERT(osd_invariant(obj));
5557         LASSERT(!dt_object_remote(dt));
5558         LASSERT(bag->ic_object == obj->oo_inode);
5559
5560         ipd = osd_idx_ipd_get(env, bag);
5561         if (IS_ERR(ipd))
5562                 RETURN(-ENOMEM);
5563
5564         /* got ipd now we can start iterator. */
5565         iam_it_init(it, bag, 0, ipd);
5566
5567         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5568                 /* swab quota uid/gid provided by caller */
5569                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5570                 key = (const struct dt_key *)&oti->oti_quota_id;
5571         }
5572
5573         rc = iam_it_get(it, (struct iam_key *)key);
5574         if (rc >= 0) {
5575                 if (S_ISDIR(obj->oo_inode->i_mode))
5576                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5577                 else
5578                         iam_rec = (struct iam_rec *)rec;
5579
5580                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5581
5582                 if (S_ISDIR(obj->oo_inode->i_mode))
5583                         osd_fid_unpack((struct lu_fid *)rec,
5584                                        (struct osd_fid_pack *)iam_rec);
5585                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5586                         osd_quota_unpack(obj, rec);
5587         }
5588
5589         iam_it_put(it);
5590         iam_it_fini(it);
5591         osd_ipd_put(env, bag, ipd);
5592
5593         LINVRNT(osd_invariant(obj));
5594
5595         RETURN(rc);
5596 }
5597
5598 static int osd_index_declare_iam_insert(const struct lu_env *env,
5599                                         struct dt_object *dt,
5600                                         const struct dt_rec *rec,
5601                                         const struct dt_key *key,
5602                                         struct thandle *handle)
5603 {
5604         struct osd_thandle *oh;
5605
5606         LASSERT(handle != NULL);
5607
5608         oh = container_of(handle, struct osd_thandle, ot_super);
5609         LASSERT(oh->ot_handle == NULL);
5610
5611         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5612                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5613
5614         return 0;
5615 }
5616
5617 /**
5618  *      Inserts (key, value) pair in \a dt index object.
5619  *
5620  *      \param  dt      osd index object
5621  *      \param  key     key for index
5622  *      \param  rec     record reference
5623  *      \param  th      transaction handler
5624  *
5625  *      \retval  0  success
5626  *      \retval -ve failure
5627  */
5628 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5629                                 const struct dt_rec *rec,
5630                                 const struct dt_key *key, struct thandle *th)
5631 {
5632         struct osd_object *obj = osd_dt_obj(dt);
5633         struct iam_path_descr *ipd;
5634         struct osd_thandle *oh;
5635         struct iam_container *bag;
5636         struct osd_thread_info *oti = osd_oti_get(env);
5637         struct iam_rec *iam_rec;
5638         int rc;
5639
5640         ENTRY;
5641
5642         if (!dt_object_exists(dt))
5643                 RETURN(-ENOENT);
5644
5645         LINVRNT(osd_invariant(obj));
5646         LASSERT(!dt_object_remote(dt));
5647
5648         bag = &obj->oo_dir->od_container;
5649         LASSERT(bag->ic_object == obj->oo_inode);
5650         LASSERT(th != NULL);
5651
5652         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5653
5654         ipd = osd_idx_ipd_get(env, bag);
5655         if (unlikely(ipd == NULL))
5656                 RETURN(-ENOMEM);
5657
5658         oh = container_of(th, struct osd_thandle, ot_super);
5659         LASSERT(oh->ot_handle != NULL);
5660         LASSERT(oh->ot_handle->h_transaction != NULL);
5661         if (S_ISDIR(obj->oo_inode->i_mode)) {
5662                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5663                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5664                              &oti->oti_fid);
5665         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5666                 /* pack quota uid/gid */
5667                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5668                 key = (const struct dt_key *)&oti->oti_quota_id;
5669                 /* pack quota record */
5670                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5671                 iam_rec = (struct iam_rec *)rec;
5672         } else {
5673                 iam_rec = (struct iam_rec *)rec;
5674         }
5675
5676         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5677                         iam_rec, ipd);
5678         osd_ipd_put(env, bag, ipd);
5679         LINVRNT(osd_invariant(obj));
5680         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5681         RETURN(rc);
5682 }
5683
5684 /**
5685  * Calls ldiskfs_add_entry() to add directory entry
5686  * into the directory. This is required for
5687  * interoperability mode (b11826)
5688  *
5689  * \retval   0, on success
5690  * \retval -ve, on error
5691  */
5692 static int __osd_ea_add_rec(struct osd_thread_info *info,
5693                             struct osd_object *pobj, struct inode  *cinode,
5694                             const char *name, const struct lu_fid *fid,
5695                             struct htree_lock *hlock, struct thandle *th)
5696 {
5697         struct ldiskfs_dentry_param *ldp;
5698         struct dentry *child;
5699         struct osd_thandle *oth;
5700         struct lu_name ln;
5701         int rc;
5702
5703         oth = container_of(th, struct osd_thandle, ot_super);
5704         LASSERT(oth->ot_handle != NULL);
5705         LASSERT(oth->ot_handle->h_transaction != NULL);
5706         LASSERT(pobj->oo_inode);
5707
5708         rc = obj_name2lu_name(pobj, name, strlen(name), &ln);
5709         if (rc)
5710                 RETURN(rc);
5711
5712         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5713         if (unlikely(osd_object_is_root(pobj)))
5714                 ldp->edp_magic = 0;
5715         else
5716                 osd_get_ldiskfs_dirent_param(ldp, fid);
5717         child = osd_child_dentry_get(info->oti_env, pobj,
5718                                      ln.ln_name, ln.ln_namelen);
5719         child->d_fsdata = (void *)ldp;
5720         dquot_initialize(pobj->oo_inode);
5721         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5722                                    child, cinode, hlock);
5723         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5724                 struct ldiskfs_dir_entry_2      *de;
5725                 struct buffer_head              *bh;
5726                 int                              rc1;
5727
5728                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5729                                             NULL, hlock);
5730                 if (!IS_ERR(bh)) {
5731                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5732                                                                bh);
5733                         if (rc1 == 0) {
5734                                 if (S_ISDIR(cinode->i_mode))
5735                                         de->file_type = LDISKFS_DIRENT_LUFID |
5736                                                         LDISKFS_FT_REG_FILE;
5737                                 else
5738                                         de->file_type = LDISKFS_DIRENT_LUFID |
5739                                                         LDISKFS_FT_DIR;
5740                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5741                                                               NULL, bh);
5742                         }
5743                         brelse(bh);
5744                 }
5745         }
5746
5747         if (ln.ln_name != name)
5748                 kfree(ln.ln_name);
5749         RETURN(rc);
5750 }
5751
5752 /**
5753  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5754  * into the directory.Also sets flags into osd object to
5755  * indicate dot and dotdot are created. This is required for
5756  * interoperability mode (b11826)
5757  *
5758  * \param dir   directory for dot and dotdot fixup.
5759  * \param obj   child object for linking
5760  *
5761  * \retval   0, on success
5762  * \retval -ve, on error
5763  */
5764 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5765                               struct osd_object *dir,
5766                               struct inode *parent_dir, const char *name,
5767                               const struct lu_fid *dot_fid,
5768                               const struct lu_fid *dot_dot_fid,
5769                               struct thandle *th)
5770 {
5771         struct inode *inode = dir->oo_inode;
5772         struct osd_thandle *oth;
5773         int result = 0;
5774
5775         oth = container_of(th, struct osd_thandle, ot_super);
5776         LASSERT(oth->ot_handle->h_transaction != NULL);
5777         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5778
5779         if (strcmp(name, dot) == 0) {
5780                 if (dir->oo_compat_dot_created) {
5781                         result = -EEXIST;
5782                 } else {
5783                         LASSERT(inode->i_ino == parent_dir->i_ino);
5784                         dir->oo_compat_dot_created = 1;
5785                         result = 0;
5786                 }
5787         } else if (strcmp(name, dotdot) == 0) {
5788                 if (!dir->oo_compat_dot_created)
5789                         return -EINVAL;
5790                 /* in case of rename, dotdot is already created */
5791                 if (dir->oo_compat_dotdot_created) {
5792                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5793                                                 dot_dot_fid, NULL, th);
5794                 }
5795
5796                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5797                         struct lu_fid tfid = *dot_dot_fid;
5798
5799                         tfid.f_oid--;
5800                         result = osd_add_dot_dotdot_internal(info,
5801                                         dir->oo_inode, parent_dir, dot_fid,
5802                                         &tfid, oth);
5803                 } else {
5804                         result = osd_add_dot_dotdot_internal(info,
5805                                         dir->oo_inode, parent_dir, dot_fid,
5806                                         dot_dot_fid, oth);
5807                 }
5808
5809                 if (result == 0)
5810                         dir->oo_compat_dotdot_created = 1;
5811         }
5812
5813         return result;
5814 }
5815
5816
5817 /**
5818  * It will call the appropriate osd_add* function and return the
5819  * value, return by respective functions.
5820  */
5821 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5822                           struct inode *cinode, const char *name,
5823                           const struct lu_fid *fid, struct thandle *th)
5824 {
5825         struct osd_thread_info *info = osd_oti_get(env);
5826         struct htree_lock *hlock;
5827         int rc;
5828
5829         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5830
5831         if (name[0] == '.' && (name[1] == '\0' ||
5832                                (name[1] == '.' && name[2] == '\0'))) {
5833                 if (hlock != NULL) {
5834                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5835                                            pobj->oo_inode, 0);
5836                 } else {
5837                         down_write(&pobj->oo_ext_idx_sem);
5838                 }
5839
5840                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5841                                         lu_object_fid(&pobj->oo_dt.do_lu),
5842                                         fid, th);
5843         } else {
5844                 if (hlock != NULL) {
5845                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5846                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5847                 } else {
5848                         down_write(&pobj->oo_ext_idx_sem);
5849                 }
5850
5851                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5852                         struct lu_fid *tfid = &info->oti_fid;
5853
5854                         *tfid = *fid;
5855                         tfid->f_ver = ~0;
5856                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5857                                               tfid, hlock, th);
5858                 } else {
5859                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5860                                               hlock, th);
5861                 }
5862         }
5863         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5864             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5865                 pobj->oo_dirent_count++;
5866
5867         if (hlock != NULL)
5868                 ldiskfs_htree_unlock(hlock);
5869         else
5870                 up_write(&pobj->oo_ext_idx_sem);
5871
5872         return rc;
5873 }
5874
5875 static int
5876 osd_ldiskfs_consistency_check(struct osd_thread_info *oti,
5877                               struct osd_device *dev,
5878                               const struct lu_fid *fid,
5879                               struct osd_inode_id *id)
5880 {
5881         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5882         struct inode *inode = NULL;
5883         int once = 0;
5884         bool insert;
5885         int rc;
5886
5887         ENTRY;
5888         if (!scrub_needs_check(scrub, fid, id->oii_ino))
5889                 RETURN(0);
5890 again:
5891         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5892         if (rc == -ENOENT) {
5893                 __u32 gen = id->oii_gen;
5894
5895                 insert = true;
5896                 if (inode != NULL)
5897                         goto trigger;
5898
5899                 inode = osd_iget(oti, dev, id);
5900                 /* The inode has been removed (by race maybe). */
5901                 if (IS_ERR(inode)) {
5902                         rc = PTR_ERR(inode);
5903
5904                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5905                 }
5906
5907                 /* The OI mapping is lost. */
5908                 if (gen != OSD_OII_NOGEN)
5909                         goto trigger;
5910
5911                 /*
5912                  * The inode may has been reused by others, we do not know,
5913                  * leave it to be handled by subsequent osd_fid_lookup().
5914                  */
5915                 GOTO(out, rc = 0);
5916         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5917                 GOTO(out, rc);
5918         }
5919
5920         insert = false;
5921
5922 trigger:
5923         if (scrub->os_running) {
5924                 if (inode == NULL) {
5925                         inode = osd_iget(oti, dev, id);
5926                         /* The inode has been removed (by race maybe). */
5927                         if (IS_ERR(inode)) {
5928                                 rc = PTR_ERR(inode);
5929
5930                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5931                         }
5932                 }
5933
5934                 rc = osd_oii_insert(dev, fid, id, insert);
5935                 /*
5936                  * There is race condition between osd_oi_lookup and OI scrub.
5937                  * The OI scrub finished just after osd_oi_lookup() failure.
5938                  * Under such case, it is unnecessary to trigger OI scrub again,
5939                  * but try to call osd_oi_lookup() again.
5940                  */
5941                 if (unlikely(rc == -EAGAIN))
5942                         goto again;
5943
5944                 if (!S_ISDIR(inode->i_mode))
5945                         rc = 0;
5946                 else
5947                         rc = osd_check_lmv(oti, dev, inode);
5948
5949                 GOTO(out, rc);
5950         }
5951
5952         if (dev->od_scrub.os_scrub.os_auto_scrub_interval != AS_NEVER &&
5953             ++once == 1) {
5954                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5955                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5956                 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
5957                              "%s: trigger partial OI scrub for RPC inconsistency, checking FID "DFID"/%u: rc = %d\n",
5958                              osd_dev2name(dev), PFID(fid), id->oii_ino, rc);
5959                 if (rc == 0 || rc == -EALREADY)
5960                         goto again;
5961         }
5962
5963         GOTO(out, rc);
5964
5965 out:
5966         iput(inode);
5967
5968         RETURN(rc);
5969 }
5970
5971 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5972                                struct osd_device *dev,
5973                                struct lu_fid *fid, __u32 ino)
5974 {
5975         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5976         struct osd_idmap_cache *oic = &oti->oti_cache;
5977         struct inode *inode;
5978         int rc;
5979
5980         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5981         inode = osd_iget(oti, dev, &oic->oic_lid);
5982         if (IS_ERR(inode)) {
5983                 fid_zero(&oic->oic_fid);
5984                 return PTR_ERR(inode);
5985         }
5986
5987         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5988         iput(inode);
5989         if (rc != 0)
5990                 fid_zero(&oic->oic_fid);
5991         else
5992                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5993         return rc;
5994 }
5995
5996 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5997                       struct osd_inode_id *id, const struct lu_fid *fid)
5998 {
5999         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
6000                id->oii_ino, id->oii_gen, info);
6001         info->oti_cache.oic_lid = *id;
6002         info->oti_cache.oic_fid = *fid;
6003         info->oti_cache.oic_dev = osd;
6004 }
6005
6006 /**
6007  * Get parent FID from the linkEA.
6008  *
6009  * For a directory which parent resides on remote MDT, to satisfy the
6010  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
6011  * the other hand, to make the lookup(..) on the directory can return
6012  * the real parent FID, we append the real parent FID after its ".."
6013  * name entry in the /REMOTE_PARENT_DIR.
6014  *
6015  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
6016  * backup. So after the restore, we cannot get the right parent FID from
6017  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
6018  * we have stored the real parent FID in the directory object's linkEA,
6019  * we can parse the linkEA for the real parent FID.
6020  *
6021  * \param[in] env       pointer to the thread context
6022  * \param[in] obj       pointer to the object to be handled
6023  * \param[out]fid       pointer to the buffer to hold the parent FID
6024  *
6025  * \retval              0 for getting the real parent FID successfully
6026  * \retval              negative error number on failure
6027  */
6028 static int osd_get_pfid_from_linkea(const struct lu_env *env,
6029                                     struct osd_object *obj,
6030                                     struct lu_fid *fid)
6031 {
6032         struct osd_thread_info *oti = osd_oti_get(env);
6033         struct lu_buf *buf = &oti->oti_big_buf;
6034         struct dentry *dentry = &oti->oti_obj_dentry;
6035         struct inode *inode = obj->oo_inode;
6036         struct linkea_data ldata = { NULL };
6037         int rc;
6038
6039         ENTRY;
6040
6041         fid_zero(fid);
6042         if (!S_ISDIR(inode->i_mode))
6043                 RETURN(-EIO);
6044
6045 again:
6046         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6047                              buf->lb_buf, buf->lb_len);
6048         if (rc == -ERANGE) {
6049                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6050                                      NULL, 0);
6051                 if (rc > 0) {
6052                         lu_buf_realloc(buf, rc);
6053                         if (buf->lb_buf == NULL)
6054                                 RETURN(-ENOMEM);
6055
6056                         goto again;
6057                 }
6058         }
6059
6060         if (unlikely(rc == 0))
6061                 RETURN(-ENODATA);
6062
6063         if (rc < 0)
6064                 RETURN(rc);
6065
6066         if (unlikely(buf->lb_buf == NULL)) {
6067                 lu_buf_realloc(buf, rc);
6068                 if (buf->lb_buf == NULL)
6069                         RETURN(-ENOMEM);
6070
6071                 goto again;
6072         }
6073
6074         ldata.ld_buf = buf;
6075         rc = linkea_init_with_rec(&ldata);
6076         if (!rc) {
6077                 linkea_first_entry(&ldata);
6078                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
6079         }
6080
6081         RETURN(rc);
6082 }
6083
6084 static int osd_verify_ent_by_linkea(const struct lu_env *env,
6085                                     struct inode *inode,
6086                                     const struct lu_fid *pfid,
6087                                     const char *name, const int namelen)
6088 {
6089         struct osd_thread_info *oti = osd_oti_get(env);
6090         struct lu_buf *buf = &oti->oti_big_buf;
6091         struct dentry *dentry = &oti->oti_obj_dentry;
6092         struct linkea_data ldata = { NULL };
6093         struct lu_name cname = { .ln_name = name,
6094                                  .ln_namelen = namelen };
6095         int rc;
6096
6097         ENTRY;
6098
6099 again:
6100         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6101                              buf->lb_buf, buf->lb_len);
6102         if (rc == -ERANGE)
6103                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
6104
6105         if (rc < 0)
6106                 RETURN(rc);
6107
6108         if (unlikely(rc == 0))
6109                 RETURN(-ENODATA);
6110
6111         if (buf->lb_len < rc) {
6112                 lu_buf_realloc(buf, rc);
6113                 if (buf->lb_buf == NULL)
6114                         RETURN(-ENOMEM);
6115
6116                 goto again;
6117         }
6118
6119         ldata.ld_buf = buf;
6120         rc = linkea_init_with_rec(&ldata);
6121         if (!rc)
6122                 rc = linkea_links_find(&ldata, &cname, pfid);
6123
6124         RETURN(rc);
6125 }
6126
6127 /**
6128  * Calls ->lookup() to find dentry. From dentry get inode and
6129  * read inode's ea to get fid. This is required for  interoperability
6130  * mode (b11826)
6131  *
6132  * \retval   0, on success
6133  * \retval -ve, on error
6134  */
6135 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
6136                              struct dt_rec *rec, const struct dt_key *key)
6137 {
6138         struct inode *dir = obj->oo_inode;
6139         struct dentry *dentry;
6140         struct ldiskfs_dir_entry_2 *de;
6141         struct buffer_head *bh;
6142         struct lu_fid *fid = (struct lu_fid *)rec;
6143         struct htree_lock *hlock = NULL;
6144         struct lu_name ln;
6145         int ino;
6146         int rc;
6147
6148         ENTRY;
6149
6150         LASSERT(dir->i_op != NULL);
6151         LASSERT(dir->i_op->lookup != NULL);
6152
6153         rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
6154         if (rc)
6155                 RETURN(rc);
6156
6157         dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
6158
6159         if (obj->oo_hl_head != NULL) {
6160                 hlock = osd_oti_get(env)->oti_hlock;
6161                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6162                                    dir, LDISKFS_HLOCK_LOOKUP);
6163         } else {
6164                 down_read(&obj->oo_ext_idx_sem);
6165         }
6166
6167         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6168         if (!IS_ERR(bh)) {
6169                 struct osd_thread_info *oti = osd_oti_get(env);
6170                 struct osd_inode_id *id = &oti->oti_id;
6171                 struct osd_device *dev = osd_obj2dev(obj);
6172
6173                 ino = le32_to_cpu(de->inode);
6174                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
6175                         brelse(bh);
6176                         rc = osd_fail_fid_lookup(oti, dev, fid, ino);
6177                         GOTO(out, rc);
6178                 }
6179
6180                 rc = osd_get_fid_from_dentry(de, rec);
6181
6182                 /* done with de, release bh */
6183                 brelse(bh);
6184                 if (rc != 0) {
6185                         if (unlikely(is_remote_parent_ino(dev, ino))) {
6186                                 const char *name = (const char *)key;
6187
6188                                 /*
6189                                  * If the parent is on remote MDT, and there
6190                                  * is no FID-in-dirent, then we have to get
6191                                  * the parent FID from the linkEA.
6192                                  */
6193                                 if (likely(strlen(name) == 2 &&
6194                                            name[0] == '.' && name[1] == '.'))
6195                                         rc = osd_get_pfid_from_linkea(env, obj,
6196                                                                       fid);
6197                         } else {
6198                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6199                         }
6200                 } else {
6201                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6202                 }
6203
6204                 if (rc != 0 || osd_remote_fid(env, dev, fid))
6205                         GOTO(out, rc);
6206
6207                 rc = osd_ldiskfs_consistency_check(oti, dev, fid, id);
6208                 if (rc != -ENOENT) {
6209                         /* Other error should not affect lookup result. */
6210                         rc = 0;
6211
6212                         /* Normal file mapping should be added into OI cache
6213                          * after FID in LMA check, but for local files like
6214                          * hsm_actions, their FIDs are not stored in OI files,
6215                          * see osd_initial_OI_scrub(), and here is the only
6216                          * place to load mapping into OI cache.
6217                          */
6218                         if (!fid_is_namespace_visible(fid))
6219                                 osd_add_oi_cache(osd_oti_get(env),
6220                                                  osd_obj2dev(obj), id, fid);
6221                 }
6222         } else {
6223                 rc = PTR_ERR(bh);
6224         }
6225
6226         GOTO(out, rc);
6227
6228 out:
6229         if (hlock != NULL)
6230                 ldiskfs_htree_unlock(hlock);
6231         else
6232                 up_read(&obj->oo_ext_idx_sem);
6233         if (ln.ln_name != (char *)key)
6234                 kfree(ln.ln_name);
6235         RETURN(rc);
6236 }
6237
6238 static int osd_index_declare_ea_insert(const struct lu_env *env,
6239                                        struct dt_object *dt,
6240                                        const struct dt_rec *rec,
6241                                        const struct dt_key *key,
6242                                        struct thandle *handle)
6243 {
6244         struct osd_thandle *oh;
6245         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6246         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6247         const struct lu_fid *fid = rec1->rec_fid;
6248         int credits, rc = 0;
6249         struct osd_idmap_cache *idc;
6250
6251         ENTRY;
6252
6253         LASSERT(!dt_object_remote(dt));
6254         LASSERT(handle != NULL);
6255         LASSERT(fid != NULL);
6256         LASSERT(rec1->rec_type != 0);
6257
6258         oh = container_of(handle, struct osd_thandle, ot_super);
6259         LASSERT(oh->ot_handle == NULL);
6260
6261         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6262
6263         /*
6264          * we can't call iget() while a transactions is running
6265          * (this can lead to a deadlock), but we need to know
6266          * inum and object type. so we find this information at
6267          * declaration and cache in per-thread info
6268          */
6269         idc = osd_idc_find_or_init(env, osd, fid);
6270         if (IS_ERR(idc))
6271                 RETURN(PTR_ERR(idc));
6272         if (idc->oic_remote) {
6273                 /*
6274                  * a reference to remote inode is represented by an
6275                  * agent inode which we have to create
6276                  */
6277                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6278                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6279         }
6280
6281         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6282
6283         if (osd_dt_obj(dt)->oo_inode != NULL) {
6284                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6285
6286                 /*
6287                  * We ignore block quota on meta pool (MDTs), so needn't
6288                  * calculate how many blocks will be consumed by this index
6289                  * insert
6290                  */
6291                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6292                                            i_gid_read(inode),
6293                                            i_projid_read(inode), 0,
6294                                            oh, osd_dt_obj(dt), NULL,
6295                                            OSD_QID_BLK);
6296                 if (rc)
6297                         RETURN(rc);
6298
6299 #ifdef HAVE_PROJECT_QUOTA
6300                 /*
6301                  * Reserve credits for local agent inode to transfer
6302                  * to 0, quota enforcement is ignored in this case.
6303                  */
6304                 if (idc->oic_remote &&
6305                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6306                     i_projid_read(inode) != 0)
6307                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6308                                                   0, i_projid_read(inode),
6309                                                   0, false, PRJQUOTA);
6310 #endif
6311         }
6312
6313         RETURN(rc);
6314 }
6315
6316 /**
6317  * Index add function for interoperability mode (b11826).
6318  * It will add the directory entry.This entry is needed to
6319  * maintain name->fid mapping.
6320  *
6321  * \param key it is key i.e. file entry to be inserted
6322  * \param rec it is value of given key i.e. fid
6323  *
6324  * \retval   0, on success
6325  * \retval -ve, on error
6326  */
6327 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6328                                const struct dt_rec *rec,
6329                                const struct dt_key *key, struct thandle *th)
6330 {
6331         struct osd_object *obj = osd_dt_obj(dt);
6332         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6333         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6334         const struct lu_fid *fid = rec1->rec_fid;
6335         const char *name = (const char *)key;
6336         struct osd_thread_info *oti = osd_oti_get(env);
6337         struct inode *child_inode = NULL;
6338         struct osd_idmap_cache *idc;
6339         int rc;
6340
6341         ENTRY;
6342
6343         if (!dt_object_exists(dt))
6344                 RETURN(-ENOENT);
6345
6346         LASSERT(osd_invariant(obj));
6347         LASSERT(!dt_object_remote(dt));
6348         LASSERT(th != NULL);
6349
6350         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6351
6352         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6353
6354         idc = osd_idc_find(env, osd, fid);
6355         if (unlikely(idc == NULL)) {
6356                 idc = osd_idc_find_or_init(env, osd, fid);
6357                 if (IS_ERR(idc)) {
6358                         /*
6359                          * this dt_insert() wasn't declared properly, so
6360                          * FID is missing in OI cache. we better do not
6361                          * lookup FID in FLDB/OI and don't risk to deadlock,
6362                          * but in some special cases (lfsck testing, etc)
6363                          * it's much simpler than fixing a caller.
6364                          *
6365                          * normally this error should be placed after the first
6366                          * find, but migrate may attach source stripes to
6367                          * target, which doesn't create stripes.
6368                          */
6369                         CERROR("%s: "DFID" wasn't declared for insert\n",
6370                                osd_name(osd), PFID(fid));
6371                         dump_stack();
6372                         RETURN(PTR_ERR(idc));
6373                 }
6374         }
6375
6376         if (idc->oic_remote) {
6377                 /* Insert remote entry */
6378                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6379                         child_inode =
6380                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6381                 } else {
6382                         child_inode = osd_create_local_agent_inode(env, osd,
6383                                         obj, fid, rec1->rec_type & S_IFMT, th);
6384                         if (IS_ERR(child_inode))
6385                                 RETURN(PTR_ERR(child_inode));
6386                 }
6387         } else {
6388                 /* Insert local entry */
6389                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6390                         /* for a reason OI cache wasn't filled properly */
6391                         CERROR("%s: OIC for "DFID" isn't filled\n",
6392                                osd_name(osd), PFID(fid));
6393                         RETURN(-EINVAL);
6394                 }
6395                 child_inode = oti->oti_inode;
6396                 if (unlikely(child_inode == NULL)) {
6397                         struct ldiskfs_inode_info *lii;
6398
6399                         OBD_ALLOC_PTR(lii);
6400                         if (lii == NULL)
6401                                 RETURN(-ENOMEM);
6402                         child_inode = oti->oti_inode = &lii->vfs_inode;
6403                 }
6404                 child_inode->i_sb = osd_sb(osd);
6405                 child_inode->i_ino = idc->oic_lid.oii_ino;
6406                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6407         }
6408
6409         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6410
6411         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6412                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6413
6414         if (child_inode && child_inode != oti->oti_inode)
6415                 iput(child_inode);
6416         LASSERT(osd_invariant(obj));
6417         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6418
6419         RETURN(rc);
6420 }
6421
6422 /**
6423  *  Initialize osd Iterator for given osd index object.
6424  *
6425  *  \param  dt      osd index object
6426  */
6427
6428 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6429                                      struct dt_object *dt,
6430                                      __u32 unused)
6431 {
6432         struct osd_it_iam *it;
6433         struct osd_object *obj = osd_dt_obj(dt);
6434         struct lu_object *lo = &dt->do_lu;
6435         struct iam_path_descr *ipd;
6436         struct iam_container *bag = &obj->oo_dir->od_container;
6437
6438         if (!dt_object_exists(dt))
6439                 return ERR_PTR(-ENOENT);
6440
6441         OBD_ALLOC_PTR(it);
6442         if (it == NULL)
6443                 return ERR_PTR(-ENOMEM);
6444
6445         ipd = osd_it_ipd_get(env, bag);
6446         if (likely(ipd != NULL)) {
6447                 it->oi_obj = obj;
6448                 it->oi_ipd = ipd;
6449                 lu_object_get(lo);
6450                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6451                 return (struct dt_it *)it;
6452         } else {
6453                 OBD_FREE_PTR(it);
6454                 return ERR_PTR(-ENOMEM);
6455         }
6456 }
6457
6458 /**
6459  * free given Iterator.
6460  */
6461 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6462 {
6463         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6464         struct osd_object *obj = it->oi_obj;
6465
6466         iam_it_fini(&it->oi_it);
6467         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6468         osd_object_put(env, obj);
6469         OBD_FREE_PTR(it);
6470 }
6471
6472 /**
6473  *  Move Iterator to record specified by \a key
6474  *
6475  *  \param  di      osd iterator
6476  *  \param  key     key for index
6477  *
6478  *  \retval +ve  di points to record with least key not larger than key
6479  *  \retval  0   di points to exact matched key
6480  *  \retval -ve  failure
6481  */
6482
6483 static int osd_it_iam_get(const struct lu_env *env,
6484                           struct dt_it *di, const struct dt_key *key)
6485 {
6486         struct osd_thread_info *oti = osd_oti_get(env);
6487         struct osd_it_iam *it = (struct osd_it_iam *)di;
6488
6489         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6490                 /* swab quota uid/gid */
6491                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6492                 key = (struct dt_key *)&oti->oti_quota_id;
6493         }
6494
6495         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6496 }
6497
6498 /**
6499  *  Release Iterator
6500  *
6501  *  \param  di      osd iterator
6502  */
6503 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6504 {
6505         struct osd_it_iam *it = (struct osd_it_iam *)di;
6506
6507         iam_it_put(&it->oi_it);
6508 }
6509
6510 /**
6511  *  Move iterator by one record
6512  *
6513  *  \param  di      osd iterator
6514  *
6515  *  \retval +1   end of container reached
6516  *  \retval  0   success
6517  *  \retval -ve  failure
6518  */
6519
6520 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6521 {
6522         struct osd_it_iam *it = (struct osd_it_iam *)di;
6523
6524         return iam_it_next(&it->oi_it);
6525 }
6526
6527 /**
6528  * Return pointer to the key under iterator.
6529  */
6530
6531 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6532                                      const struct dt_it *di)
6533 {
6534         struct osd_thread_info *oti = osd_oti_get(env);
6535         struct osd_it_iam *it = (struct osd_it_iam *)di;
6536         struct osd_object *obj = it->oi_obj;
6537         struct dt_key *key;
6538
6539         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6540
6541         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6542                 /* swab quota uid/gid */
6543                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6544                 key = (struct dt_key *)&oti->oti_quota_id;
6545         }
6546
6547         return key;
6548 }
6549
6550 /**
6551  * Return size of key under iterator (in bytes)
6552  */
6553
6554 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6555 {
6556         struct osd_it_iam *it = (struct osd_it_iam *)di;
6557
6558         return iam_it_key_size(&it->oi_it);
6559 }
6560
6561 static inline void
6562 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6563 {
6564         /* check if file type is required */
6565         if (ent->lde_attrs & LUDA_TYPE) {
6566                 struct luda_type *lt;
6567                 int align = sizeof(*lt) - 1;
6568
6569                 len = (len + align) & ~align;
6570                 lt = (struct luda_type *)(ent->lde_name + len);
6571                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6572         }
6573
6574         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6575 }
6576
6577 /**
6578  * build lu direct from backend fs dirent.
6579  */
6580
6581 static inline void
6582 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6583                    char *name, __u16 namelen, __u16 type, __u32 attr)
6584 {
6585         ent->lde_attrs = attr | LUDA_FID;
6586         fid_cpu_to_le(&ent->lde_fid, fid);
6587
6588         ent->lde_hash = cpu_to_le64(offset);
6589         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6590
6591         strncpy(ent->lde_name, name, namelen);
6592         ent->lde_name[namelen] = '\0';
6593         ent->lde_namelen = cpu_to_le16(namelen);
6594
6595         /* append lustre attributes */
6596         osd_it_append_attrs(ent, namelen, type);
6597 }
6598
6599 /**
6600  * Return pointer to the record under iterator.
6601  */
6602 static int osd_it_iam_rec(const struct lu_env *env,
6603                           const struct dt_it *di,
6604                           struct dt_rec *dtrec, __u32 attr)
6605 {
6606         struct osd_it_iam *it = (struct osd_it_iam *)di;
6607         struct osd_thread_info *info = osd_oti_get(env);
6608
6609         ENTRY;
6610
6611         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6612                 const struct osd_fid_pack *rec;
6613                 struct lu_fid *fid = &info->oti_fid;
6614                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6615                 char *name;
6616                 int namelen;
6617                 __u64 hash;
6618                 int rc;
6619
6620                 name = (char *)iam_it_key_get(&it->oi_it);
6621                 if (IS_ERR(name))
6622                         RETURN(PTR_ERR(name));
6623
6624                 namelen = iam_it_key_size(&it->oi_it);
6625
6626                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6627                 if (IS_ERR(rec))
6628                         RETURN(PTR_ERR(rec));
6629
6630                 rc = osd_fid_unpack(fid, rec);
6631                 if (rc)
6632                         RETURN(rc);
6633
6634                 hash = iam_it_store(&it->oi_it);
6635
6636                 /* IAM does not store object type in IAM index (dir) */
6637                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6638                                    0, LUDA_FID);
6639         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6640                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6641                            (struct iam_rec *)dtrec);
6642                 osd_quota_unpack(it->oi_obj, dtrec);
6643         } else {
6644                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6645                            (struct iam_rec *)dtrec);
6646         }
6647
6648         RETURN(0);
6649 }
6650
6651 /**
6652  * Returns cookie for current Iterator position.
6653  */
6654 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6655 {
6656         struct osd_it_iam *it = (struct osd_it_iam *)di;
6657
6658         return iam_it_store(&it->oi_it);
6659 }
6660
6661 /**
6662  * Restore iterator from cookie.
6663  *
6664  * \param  di      osd iterator
6665  * \param  hash    Iterator location cookie
6666  *
6667  * \retval +ve  di points to record with least key not larger than key.
6668  * \retval  0   di points to exact matched key
6669  * \retval -ve  failure
6670  */
6671
6672 static int osd_it_iam_load(const struct lu_env *env,
6673                            const struct dt_it *di, __u64 hash)
6674 {
6675         struct osd_it_iam *it = (struct osd_it_iam *)di;
6676
6677         return iam_it_load(&it->oi_it, hash);
6678 }
6679
6680 static const struct dt_index_operations osd_index_iam_ops = {
6681         .dio_lookup         = osd_index_iam_lookup,
6682         .dio_declare_insert = osd_index_declare_iam_insert,
6683         .dio_insert         = osd_index_iam_insert,
6684         .dio_declare_delete = osd_index_declare_iam_delete,
6685         .dio_delete         = osd_index_iam_delete,
6686         .dio_it     = {
6687                 .init     = osd_it_iam_init,
6688                 .fini     = osd_it_iam_fini,
6689                 .get      = osd_it_iam_get,
6690                 .put      = osd_it_iam_put,
6691                 .next     = osd_it_iam_next,
6692                 .key      = osd_it_iam_key,
6693                 .key_size = osd_it_iam_key_size,
6694                 .rec      = osd_it_iam_rec,
6695                 .store    = osd_it_iam_store,
6696                 .load     = osd_it_iam_load
6697         }
6698 };
6699
6700 struct osd_it_ea *osd_it_dir_init(const struct lu_env *env,
6701                                   struct osd_device *dev,
6702                                   struct inode *inode, u32 attr)
6703 {
6704         struct osd_thread_info *info = osd_oti_get(env);
6705         struct osd_it_ea *oie;
6706         struct file *file;
6707
6708         ENTRY;
6709         file = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
6710                                  inode->i_fop);
6711         if (IS_ERR(file))
6712                 RETURN(ERR_CAST(file));
6713
6714         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6715         if (attr & LUDA_64BITHASH)
6716                 file->f_mode |= FMODE_64BITHASH;
6717         else
6718                 file->f_mode |= FMODE_32BITHASH;
6719         ihold(inode);
6720
6721         OBD_SLAB_ALLOC_PTR(oie, osd_itea_cachep);
6722         if (!oie)
6723                 goto out_fput;
6724
6725         oie->oie_rd_dirent       = 0;
6726         oie->oie_it_dirent       = 0;
6727         oie->oie_dirent          = NULL;
6728         if (unlikely(!info->oti_it_ea_buf_used)) {
6729                 oie->oie_buf = info->oti_it_ea_buf;
6730                 info->oti_it_ea_buf_used = 1;
6731         } else {
6732                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6733                 if (!oie->oie_buf)
6734                         goto out_free;
6735         }
6736         oie->oie_obj = NULL;
6737         oie->oie_file = file;
6738
6739         RETURN(oie);
6740
6741 out_free:
6742         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6743 out_fput:
6744         fput(file);
6745
6746         return ERR_PTR(-ENOMEM);
6747 }
6748
6749 /**
6750  * Creates or initializes iterator context.
6751  *
6752  * \retval struct osd_it_ea, iterator structure on success
6753  *
6754  */
6755 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6756                                     struct dt_object *dt,
6757                                     __u32 attr)
6758 {
6759         struct osd_object *obj = osd_dt_obj(dt);
6760         struct osd_device *dev = osd_obj2dev(obj);
6761         struct lu_object *lo = &dt->do_lu;
6762         struct osd_it_ea *oie;
6763
6764         ENTRY;
6765
6766         if (!dt_object_exists(dt) || obj->oo_destroyed)
6767                 RETURN(ERR_PTR(-ENOENT));
6768
6769         oie = osd_it_dir_init(env, dev, obj->oo_inode, attr);
6770         if (IS_ERR(oie))
6771                 RETURN(ERR_CAST(oie));
6772
6773         oie->oie_obj = obj;
6774         lu_object_get(lo);
6775         RETURN((struct dt_it *)oie);
6776 }
6777
6778 void osd_it_dir_fini(const struct lu_env *env, struct osd_it_ea *oie,
6779                      struct inode *inode)
6780 {
6781         struct osd_thread_info *info = osd_oti_get(env);
6782
6783         ENTRY;
6784         fput(oie->oie_file);
6785         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6786                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6787         else
6788                 info->oti_it_ea_buf_used = 0;
6789         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6790         EXIT;
6791 }
6792
6793 /**
6794  * Destroy or finishes iterator context.
6795  *
6796  * \param di iterator structure to be destroyed
6797  */
6798 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6799 {
6800         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6801         struct osd_object *obj = oie->oie_obj;
6802         struct inode *inode = obj->oo_inode;
6803
6804         ENTRY;
6805         osd_it_dir_fini(env, (struct osd_it_ea *)di, inode);
6806         osd_object_put(env, obj);
6807         EXIT;
6808 }
6809
6810 /**
6811  * It position the iterator at given key, so that next lookup continues from
6812  * that key Or it is similar to dio_it->load() but based on a key,
6813  * rather than file position.
6814  *
6815  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6816  * to the beginning.
6817  *
6818  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6819  */
6820 static int osd_it_ea_get(const struct lu_env *env,
6821                          struct dt_it *di, const struct dt_key *key)
6822 {
6823         struct osd_it_ea *it = (struct osd_it_ea *)di;
6824
6825         ENTRY;
6826         LASSERT(((const char *)key)[0] == '\0');
6827         it->oie_file->f_pos = 0;
6828         it->oie_rd_dirent = 0;
6829         it->oie_it_dirent = 0;
6830         it->oie_dirent = NULL;
6831
6832         RETURN(+1);
6833 }
6834
6835 /**
6836  * Does nothing
6837  */
6838 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6839 {
6840 }
6841
6842 struct osd_filldir_cbs {
6843         struct dir_context ctx;
6844         struct osd_it_ea  *it;
6845 };
6846 /**
6847  * It is called internally by ->iterate*(). It fills the
6848  * iterator's in-memory data structure with required
6849  * information i.e. name, namelen, rec_size etc.
6850  *
6851  * \param buf in which information to be filled in.
6852  * \param name name of the file in given dir
6853  *
6854  * \retval 0 on success
6855  * \retval 1 on buffer full
6856  */
6857 #ifdef HAVE_FILLDIR_USE_CTX
6858 static int osd_ldiskfs_filldir(struct dir_context *ctx,
6859 #else
6860 static int osd_ldiskfs_filldir(void *ctx,
6861 #endif
6862                                const char *name, int namelen,
6863                                loff_t offset, __u64 ino, unsigned int d_type)
6864 {
6865         struct osd_it_ea *it = ((struct osd_filldir_cbs *)ctx)->it;
6866         struct osd_object *obj = it->oie_obj;
6867         struct osd_it_ea_dirent *ent = it->oie_dirent;
6868         struct lu_fid *fid = &ent->oied_fid;
6869         char *buf = it->oie_buf;
6870         struct osd_fid_pack *rec;
6871         ENTRY;
6872
6873         /* this should never happen */
6874         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6875                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6876                 RETURN(-EIO);
6877         }
6878
6879         /* Check for enough space. Note oied_name is not NUL terminated. */
6880         if (&ent->oied_name[namelen] > buf + OSD_IT_EA_BUFSIZE)
6881                 RETURN(1);
6882
6883         /* "." is just the object itself. */
6884         if (namelen == 1 && name[0] == '.') {
6885                 if (obj != NULL)
6886                         *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6887         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6888                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6889                 if (osd_fid_unpack(fid, rec) != 0)
6890                         fid_zero(fid);
6891         } else {
6892                 fid_zero(fid);
6893         }
6894         d_type &= ~LDISKFS_DIRENT_LUFID;
6895
6896         /* NOT export local root. */
6897         if (obj != NULL &&
6898             unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6899                 ino = obj->oo_inode->i_ino;
6900                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6901         }
6902
6903         if (obj == NULL || !(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
6904                 ent->oied_namelen = namelen;
6905                 memcpy(ent->oied_name, name, namelen);
6906         } else {
6907                 int encoded_namelen = critical_chars(name, namelen);
6908
6909                 /* Check again for enough space. */
6910                 if (&ent->oied_name[encoded_namelen] > buf + OSD_IT_EA_BUFSIZE)
6911                         RETURN(1);
6912
6913                 ent->oied_namelen = encoded_namelen;
6914
6915                 if (encoded_namelen == namelen)
6916                         memcpy(ent->oied_name, name, namelen);
6917                 else
6918                         critical_encode(name, namelen, ent->oied_name);
6919         }
6920
6921         ent->oied_ino     = ino;
6922         ent->oied_off     = offset;
6923         ent->oied_type    = d_type;
6924
6925         it->oie_rd_dirent++;
6926         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + ent->oied_namelen);
6927         RETURN(0);
6928 }
6929
6930 /**
6931  * Calls ->iterate*() to load a directory entry at a time
6932  * and stored it in iterator's in-memory data structure.
6933  *
6934  * \param di iterator's in memory structure
6935  *
6936  * \retval   0 on success
6937  * \retval -ve on error
6938  * \retval +1 reach the end of entry
6939  */
6940 int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di)
6941 {
6942         struct osd_it_ea *it = (struct osd_it_ea *)di;
6943         struct osd_object *obj = it->oie_obj;
6944         struct htree_lock *hlock = NULL;
6945         struct file *filp = it->oie_file;
6946         int rc = 0;
6947         struct osd_filldir_cbs buf = {
6948                 .ctx.actor = osd_ldiskfs_filldir,
6949                 .it = it
6950         };
6951
6952         ENTRY;
6953         it->oie_dirent = it->oie_buf;
6954         it->oie_rd_dirent = 0;
6955
6956         if (obj) {
6957                 if (obj->oo_hl_head != NULL) {
6958                         hlock = osd_oti_get(env)->oti_hlock;
6959                         ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6960                                            obj->oo_inode,
6961                                            LDISKFS_HLOCK_READDIR);
6962                 } else {
6963                         down_read(&obj->oo_ext_idx_sem);
6964                 }
6965         }
6966
6967         rc = iterate_dir(filp, &buf.ctx);
6968         if (rc)
6969                 GOTO(unlock, rc);
6970
6971         if (it->oie_rd_dirent == 0) {
6972                 /*
6973                  * If it does not get any dirent, it means it has been reached
6974                  * to the end of the dir
6975                  */
6976                 it->oie_file->f_pos = ldiskfs_get_htree_eof(it->oie_file);
6977                 if (rc == 0)
6978                         rc = 1;
6979         } else {
6980                 it->oie_dirent = it->oie_buf;
6981                 it->oie_it_dirent = 1;
6982         }
6983 unlock:
6984         if (obj) {
6985                 if (hlock != NULL)
6986                         ldiskfs_htree_unlock(hlock);
6987                 else
6988                         up_read(&obj->oo_ext_idx_sem);
6989         }
6990
6991         RETURN(rc);
6992 }
6993
6994 /**
6995  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6996  * to load a directory entry at a time and stored it in
6997  * iterator's in-memory data structure.
6998  *
6999  * \param di iterator's in memory structure
7000  *
7001  * \retval +ve iterator reached to end
7002  * \retval   0 iterator not reached to end
7003  * \retval -ve on error
7004  */
7005 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
7006 {
7007         struct osd_it_ea *it = (struct osd_it_ea *)di;
7008         int rc;
7009
7010         ENTRY;
7011
7012         if (it->oie_it_dirent < it->oie_rd_dirent) {
7013                 it->oie_dirent =
7014                         (void *)it->oie_dirent +
7015                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
7016                                        it->oie_dirent->oied_namelen);
7017                 it->oie_it_dirent++;
7018                 rc = 0;
7019         } else {
7020                 if (it->oie_file->f_pos == ldiskfs_get_htree_eof(it->oie_file))
7021                         rc = 1;
7022                 else
7023                         rc = osd_ldiskfs_it_fill(env, di);
7024         }
7025
7026         RETURN(rc);
7027 }
7028
7029 /**
7030  * Returns the key at current position from iterator's in memory structure.
7031  *
7032  * \param di iterator's in memory structure
7033  *
7034  * \retval key i.e. struct dt_key on success
7035  */
7036 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
7037                                     const struct dt_it *di)
7038 {
7039         struct osd_it_ea *it = (struct osd_it_ea *)di;
7040
7041         return (struct dt_key *)it->oie_dirent->oied_name;
7042 }
7043
7044 /**
7045  * Returns key's size at current position from iterator's in memory structure.
7046  *
7047  * \param di iterator's in memory structure
7048  *
7049  * \retval key_size i.e. struct dt_key on success
7050  */
7051 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
7052 {
7053         struct osd_it_ea *it = (struct osd_it_ea *)di;
7054
7055         return it->oie_dirent->oied_namelen;
7056 }
7057
7058 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
7059 #undef LDISKFS_DIR_REC_LEN
7060 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
7061 #endif
7062
7063 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
7064 {
7065         if (LDISKFS_DIR_REC_LEN(de) >=
7066             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
7067                 return true;
7068
7069         return false;
7070 }
7071
7072 static inline bool
7073 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
7074                      unsigned int blocksize, bool dotdot)
7075 {
7076         if (dotdot)
7077                 return osd_dotdot_has_space(de);
7078
7079         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
7080             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
7081                 return true;
7082
7083         return false;
7084 }
7085
7086 static int
7087 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
7088                     handle_t *jh, struct dentry *dentry,
7089                     const struct lu_fid *fid, struct buffer_head *bh,
7090                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
7091                     bool dotdot)
7092 {
7093         struct inode *dir = dentry->d_parent->d_inode;
7094         struct inode *inode = dentry->d_inode;
7095         struct osd_fid_pack *rec;
7096         struct ldiskfs_dentry_param *ldp;
7097         int namelen = dentry->d_name.len;
7098         int rc;
7099         struct osd_thread_info *info = osd_oti_get(env);
7100
7101         ENTRY;
7102
7103         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
7104                 RETURN(0);
7105
7106         /* There is enough space to hold the FID-in-dirent. */
7107         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
7108                 rc = ldiskfs_journal_get_write_access(jh, bh);
7109                 if (rc != 0)
7110                         RETURN(rc);
7111
7112                 de->name[namelen] = 0;
7113                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
7114                 rec->fp_len = sizeof(struct lu_fid) + 1;
7115                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
7116                 de->file_type |= LDISKFS_DIRENT_LUFID;
7117                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
7118
7119                 RETURN(rc);
7120         }
7121
7122         LASSERT(!dotdot);
7123
7124         rc = ldiskfs_delete_entry(jh, dir, de, bh);
7125         if (rc != 0)
7126                 RETURN(rc);
7127
7128         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
7129         osd_get_ldiskfs_dirent_param(ldp, fid);
7130         dentry->d_fsdata = (void *)ldp;
7131         dquot_initialize(dir);
7132         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
7133         /*
7134          * It is too bad, we cannot reinsert the name entry back.
7135          * That means we lose it!
7136          */
7137         if (rc != 0)
7138                 CDEBUG(D_LFSCK,
7139                        "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
7140                        osd_ino2name(inode), dir->i_ino, dir->i_generation,
7141                        namelen, dentry->d_name.name, PFID(fid), rc);
7142
7143         RETURN(rc);
7144 }
7145
7146 static int
7147 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
7148                         struct osd_it_ea *it, struct lu_fid *fid,
7149                         struct osd_inode_id *id, __u32 *attr)
7150 {
7151         struct osd_thread_info *info = osd_oti_get(env);
7152         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
7153         struct osd_device *dev = osd_obj2dev(obj);
7154         struct super_block *sb = osd_sb(dev);
7155         const char *devname = osd_name(dev);
7156         struct osd_it_ea_dirent *ent = it->oie_dirent;
7157         struct inode *dir = obj->oo_inode;
7158         struct htree_lock *hlock = NULL;
7159         struct buffer_head *bh = NULL;
7160         handle_t *jh = NULL;
7161         struct ldiskfs_dir_entry_2 *de;
7162         struct dentry *dentry;
7163         struct inode *inode;
7164         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
7165         int credits;
7166         int rc;
7167         bool dotdot = false;
7168         bool dirty = false;
7169         struct lu_name ln;
7170
7171         ENTRY;
7172
7173         if (ent->oied_name[0] == '.') {
7174                 if (ent->oied_namelen == 1)
7175                         RETURN(0);
7176
7177                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
7178                         dotdot = true;
7179         }
7180
7181         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
7182         inode = osd_iget(info, dev, id);
7183         if (IS_ERR(inode)) {
7184                 rc = PTR_ERR(inode);
7185                 if (rc == -ENOENT || rc == -ESTALE) {
7186                         /*
7187                          * Maybe dangling name entry, or
7188                          * corrupted directory entry.
7189                          */
7190                         *attr |= LUDA_UNKNOWN;
7191                         rc = 0;
7192                 } else {
7193                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
7194                                "check_repair, dir = %lu/%u, name = %.*s, "
7195                                "ino = %llu, rc = %d\n",
7196                                devname, dir->i_ino, dir->i_generation,
7197                                ent->oied_namelen, ent->oied_name,
7198                                ent->oied_ino, rc);
7199                 }
7200
7201                 RETURN(rc);
7202         }
7203
7204         rc = obj_name2lu_name(obj, ent->oied_name, ent->oied_namelen, &ln);
7205         if (rc)
7206                 RETURN(rc);
7207
7208         dentry = osd_child_dentry_by_inode(env, dir, ln.ln_name, ln.ln_namelen);
7209         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
7210         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
7211                 lma = NULL;
7212         else if (rc != 0)
7213                 GOTO(out, rc);
7214
7215         /*
7216          * We need to ensure that the name entry is still valid.
7217          * Because it may be removed or renamed by other already.
7218          *
7219          * The unlink or rename operation will start journal before PDO lock,
7220          * so to avoid deadlock, here we need to start journal handle before
7221          * related PDO lock also. But because we do not know whether there
7222          * will be something to be repaired before PDO lock, we just start
7223          * journal without conditions.
7224          *
7225          * We may need to remove the name entry firstly, then insert back.
7226          * One credit is for user quota file update.
7227          * One credit is for group quota file update.
7228          * Two credits are for dirty inode.
7229          */
7230         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
7231                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
7232
7233         if (dev->od_dirent_journal != 0) {
7234
7235 again:
7236                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
7237                 if (IS_ERR(jh)) {
7238                         rc = PTR_ERR(jh);
7239                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
7240                                "check_repair, dir = %lu/%u, credits = %d, "
7241                                "name = %.*s, ino = %llu: rc = %d\n",
7242                                devname, dir->i_ino, dir->i_generation, credits,
7243                                ent->oied_namelen, ent->oied_name,
7244                                ent->oied_ino, rc);
7245
7246                         GOTO(out_inode, rc);
7247                 }
7248
7249                 if (obj->oo_hl_head != NULL) {
7250                         hlock = osd_oti_get(env)->oti_hlock;
7251                         /*
7252                          * "0" means exclusive lock for the whole directory.
7253                          * We need to prevent others access such name entry
7254                          * during the delete + insert. Neither HLOCK_ADD nor
7255                          * HLOCK_DEL cannot guarantee the atomicity.
7256                          */
7257                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7258                 } else {
7259                         down_write(&obj->oo_ext_idx_sem);
7260                 }
7261         } else {
7262                 if (obj->oo_hl_head != NULL) {
7263                         hlock = osd_oti_get(env)->oti_hlock;
7264                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7265                                            LDISKFS_HLOCK_LOOKUP);
7266                 } else {
7267                         down_read(&obj->oo_ext_idx_sem);
7268                 }
7269         }
7270
7271         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7272         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7273                 *attr |= LUDA_IGNORE;
7274
7275                 GOTO(out, rc = 0);
7276         }
7277
7278         /*
7279          * For dotdot entry, if there is not enough space to hold the
7280          * FID-in-dirent, just keep them there. It only happens when the
7281          * device upgraded from 1.8 or restored from MDT file-level backup.
7282          * For the whole directory, only dotdot entry have no FID-in-dirent
7283          * and needs to get FID from LMA when readdir, it will not affect the
7284          * performance much.
7285          */
7286         if (dotdot && !osd_dotdot_has_space(de)) {
7287                 *attr |= LUDA_UNKNOWN;
7288
7289                 GOTO(out, rc = 0);
7290         }
7291
7292         if (lma != NULL) {
7293                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7294                         GOTO(out, rc = 0);
7295
7296                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7297                         struct lu_fid *tfid = &lma->lma_self_fid;
7298
7299                         if (likely(dotdot &&
7300                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7301                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7302                                 /*
7303                                  * It must be REMOTE_PARENT_DIR and as the
7304                                  * 'dotdot' entry of remote directory
7305                                  */
7306                                 *attr |= LUDA_IGNORE;
7307                         } else {
7308                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7309                                        "parent directory, but got %.*s under "
7310                                        "dir = %lu/%u with the FID "DFID"\n",
7311                                        devname, ent->oied_namelen,
7312                                        ent->oied_name, dir->i_ino,
7313                                        dir->i_generation, PFID(tfid));
7314
7315                                 *attr |= LUDA_UNKNOWN;
7316                         }
7317
7318                         GOTO(out, rc = 0);
7319                 }
7320         }
7321
7322         if (!fid_is_zero(fid)) {
7323                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7324                                               ent->oied_namelen);
7325                 if (rc == -ENOENT ||
7326                     (rc == -ENODATA &&
7327                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7328                         /*
7329                          * linkEA does not recognize the dirent entry,
7330                          * it may because the dirent entry corruption
7331                          * and points to other's inode.
7332                          */
7333                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7334                                "recognize the dirent, dir = %lu/%u, "
7335                                " name = %.*s, ino = %llu, "
7336                                DFID": rc = %d\n", devname, dir->i_ino,
7337                                dir->i_generation, ent->oied_namelen,
7338                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7339                         *attr |= LUDA_UNKNOWN;
7340
7341                         GOTO(out, rc = 0);
7342                 }
7343
7344                 if (rc && rc != -ENODATA) {
7345                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7346                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7347                                DFID": rc = %d\n", devname, dir->i_ino,
7348                                dir->i_generation, ent->oied_namelen,
7349                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7350                         *attr |= LUDA_UNKNOWN;
7351
7352                         GOTO(out, rc = 0);
7353                 }
7354         }
7355
7356         if (lma != NULL) {
7357                 /*
7358                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7359                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7360                  */
7361                 if (*attr & LUDA_VERIFY_DRYRUN) {
7362                         *fid = lma->lma_self_fid;
7363                         *attr |= LUDA_REPAIR;
7364
7365                         GOTO(out, rc = 0);
7366                 }
7367
7368                 if (jh == NULL) {
7369                         brelse(bh);
7370                         dev->od_dirent_journal = 1;
7371                         if (hlock != NULL) {
7372                                 ldiskfs_htree_unlock(hlock);
7373                                 hlock = NULL;
7374                         } else {
7375                                 up_read(&obj->oo_ext_idx_sem);
7376                         }
7377
7378                         goto again;
7379                 }
7380
7381                 *fid = lma->lma_self_fid;
7382                 dirty = true;
7383                 /* Update or append the FID-in-dirent. */
7384                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7385                                          bh, de, hlock, dotdot);
7386                 if (rc == 0)
7387                         *attr |= LUDA_REPAIR;
7388                 else
7389                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7390                                "the dirent, dir = %lu/%u, name = %.*s, "
7391                                "ino = %llu, "DFID": rc = %d\n",
7392                                devname, dir->i_ino, dir->i_generation,
7393                                ent->oied_namelen, ent->oied_name,
7394                                ent->oied_ino, PFID(fid), rc);
7395         } else {
7396                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7397                 if (*attr & LUDA_VERIFY_DRYRUN) {
7398                         if (fid_is_sane(fid)) {
7399                                 *attr |= LUDA_REPAIR;
7400                         } else if (dev->od_index == 0) {
7401                                 lu_igif_build(fid, inode->i_ino,
7402                                               inode->i_generation);
7403                                 *attr |= LUDA_UPGRADE;
7404                         }
7405
7406                         GOTO(out, rc = 0);
7407                 }
7408
7409                 if (jh == NULL) {
7410                         brelse(bh);
7411                         dev->od_dirent_journal = 1;
7412                         if (hlock != NULL) {
7413                                 ldiskfs_htree_unlock(hlock);
7414                                 hlock = NULL;
7415                         } else {
7416                                 up_read(&obj->oo_ext_idx_sem);
7417                         }
7418
7419                         goto again;
7420                 }
7421
7422                 dirty = true;
7423                 if (unlikely(fid_is_sane(fid))) {
7424                         /*
7425                          * FID-in-dirent exists, but FID-in-LMA is lost.
7426                          * Trust the FID-in-dirent, and add FID-in-LMA.
7427                          */
7428                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7429                         if (rc == 0)
7430                                 *attr |= LUDA_REPAIR;
7431                         else
7432                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7433                                        "update dirent, dir = %lu/%u, "
7434                                        "name = %.*s, ino = %llu, "
7435                                        DFID": rc = %d\n",
7436                                        devname, dir->i_ino, dir->i_generation,
7437                                        ent->oied_namelen, ent->oied_name,
7438                                        ent->oied_ino, PFID(fid), rc);
7439                 } else if (dev->od_index == 0) {
7440                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7441                         /*
7442                          * It is probably IGIF object. Only aappend the
7443                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7444                          */
7445                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7446                                                  bh, de, hlock, dotdot);
7447                         if (rc == 0)
7448                                 *attr |= LUDA_UPGRADE;
7449                         else
7450                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7451                                        "after the dirent, dir = %lu/%u, "
7452                                        "name = %.*s, ino = %llu, "
7453                                        DFID": rc = %d\n",
7454                                        devname, dir->i_ino, dir->i_generation,
7455                                        ent->oied_namelen, ent->oied_name,
7456                                        ent->oied_ino, PFID(fid), rc);
7457                 }
7458         }
7459
7460         GOTO(out, rc);
7461
7462 out:
7463         if (!IS_ERR(bh))
7464                 brelse(bh);
7465         if (hlock != NULL) {
7466                 ldiskfs_htree_unlock(hlock);
7467         } else {
7468                 if (dev->od_dirent_journal != 0)
7469                         up_write(&obj->oo_ext_idx_sem);
7470                 else
7471                         up_read(&obj->oo_ext_idx_sem);
7472         }
7473
7474         if (jh != NULL)
7475                 ldiskfs_journal_stop(jh);
7476
7477 out_inode:
7478         iput(inode);
7479         if (rc >= 0 && !dirty)
7480                 dev->od_dirent_journal = 0;
7481         if (ln.ln_name != ent->oied_name)
7482                 kfree(ln.ln_name);
7483
7484         return rc;
7485 }
7486
7487 /**
7488  * Returns the value at current position from iterator's in memory structure.
7489  *
7490  * \param di struct osd_it_ea, iterator's in memory structure
7491  * \param attr attr requested for dirent.
7492  * \param lde lustre dirent
7493  *
7494  * \retval   0 no error and \param lde has correct lustre dirent.
7495  * \retval -ve on error
7496  */
7497 static inline int osd_it_ea_rec(const struct lu_env *env,
7498                                 const struct dt_it *di,
7499                                 struct dt_rec *dtrec, __u32 attr)
7500 {
7501         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7502         struct osd_object      *obj   = it->oie_obj;
7503         struct osd_device      *dev   = osd_obj2dev(obj);
7504         struct osd_thread_info *oti   = osd_oti_get(env);
7505         struct osd_inode_id    *id    = &oti->oti_id;
7506         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7507         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7508         __u32 ino = it->oie_dirent->oied_ino;
7509         int rc = 0;
7510
7511         ENTRY;
7512
7513         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7514
7515         if (attr & LUDA_VERIFY) {
7516                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7517                         attr |= LUDA_IGNORE;
7518                         /*
7519                          * If the parent is on remote MDT, and there
7520                          * is no FID-in-dirent, then we have to get
7521                          * the parent FID from the linkEA.
7522                          */
7523                         if (!fid_is_sane(fid) &&
7524                             it->oie_dirent->oied_namelen == 2 &&
7525                             it->oie_dirent->oied_name[0] == '.' &&
7526                             it->oie_dirent->oied_name[1] == '.')
7527                                 osd_get_pfid_from_linkea(env, obj, fid);
7528                 } else {
7529                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7530                                                      &attr);
7531                 }
7532
7533                 if (!fid_is_sane(fid))
7534                         attr |= LUDA_UNKNOWN;
7535         } else {
7536                 attr &= ~LU_DIRENT_ATTRS_MASK;
7537                 if (!fid_is_sane(fid)) {
7538                         bool is_dotdot = false;
7539
7540                         if (it->oie_dirent->oied_namelen == 2 &&
7541                             it->oie_dirent->oied_name[0] == '.' &&
7542                             it->oie_dirent->oied_name[1] == '.')
7543                                 is_dotdot = true;
7544                         /*
7545                          * If the parent is on remote MDT, and there
7546                          * is no FID-in-dirent, then we have to get
7547                          * the parent FID from the linkEA.
7548                          */
7549                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7550                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7551                         } else {
7552                                 if (is_dotdot == false &&
7553                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7554                                         RETURN(-ENOENT);
7555
7556                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7557                         }
7558                 }
7559         }
7560
7561         /* Pack the entry anyway, at least the offset is right. */
7562         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7563                            it->oie_dirent->oied_name,
7564                            it->oie_dirent->oied_namelen,
7565                            it->oie_dirent->oied_type, attr);
7566
7567         RETURN(rc > 0 ? 0 : rc);
7568 }
7569
7570 /**
7571  * Returns the record size size at current position.
7572  *
7573  * This function will return record(lu_dirent) size in bytes.
7574  *
7575  * \param[in] env       execution environment
7576  * \param[in] di        iterator's in memory structure
7577  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7578  *                      calculate the lu_dirent size.
7579  *
7580  * \retval      record size(in bytes & in memory) of the current lu_dirent
7581  *              entry.
7582  */
7583 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7584                               __u32 attr)
7585 {
7586         struct osd_it_ea *it = (struct osd_it_ea *)di;
7587
7588         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7589 }
7590
7591 /**
7592  * Returns a cookie for current position of the iterator head, so that
7593  * user can use this cookie to load/start the iterator next time.
7594  *
7595  * \param di iterator's in memory structure
7596  *
7597  * \retval cookie for current position, on success
7598  */
7599 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7600 {
7601         struct osd_it_ea *it = (struct osd_it_ea *)di;
7602
7603         return it->oie_dirent->oied_off;
7604 }
7605
7606 /**
7607  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7608  * to load a directory entry at a time and stored it i inn,
7609  * in iterator's in-memory data structure.
7610  *
7611  * \param di struct osd_it_ea, iterator's in memory structure
7612  *
7613  * \retval +ve on success
7614  * \retval -ve on error
7615  */
7616 static int osd_it_ea_load(const struct lu_env *env,
7617                           const struct dt_it *di, __u64 hash)
7618 {
7619         struct osd_it_ea *it = (struct osd_it_ea *)di;
7620         int rc;
7621
7622         ENTRY;
7623         it->oie_file->f_pos = hash;
7624
7625         rc =  osd_ldiskfs_it_fill(env, di);
7626         if (rc > 0)
7627                 rc = -ENODATA;
7628
7629         if (rc == 0)
7630                 rc = 1;
7631
7632         RETURN(rc);
7633 }
7634
7635 /**
7636  * Index lookup function for interoperability mode (b11826).
7637  *
7638  * \param key,  key i.e. file name to be searched
7639  *
7640  * \retval +ve, on success
7641  * \retval -ve, on error
7642  */
7643 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7644                                struct dt_rec *rec, const struct dt_key *key)
7645 {
7646         struct osd_object *obj = osd_dt_obj(dt);
7647         int rc = 0;
7648
7649         ENTRY;
7650
7651         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7652         LINVRNT(osd_invariant(obj));
7653
7654         rc = osd_ea_lookup_rec(env, obj, rec, key);
7655         if (rc == 0)
7656                 rc = 1;
7657         RETURN(rc);
7658 }
7659
7660 /**
7661  * Index and Iterator operations for interoperability
7662  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7663  */
7664 static const struct dt_index_operations osd_index_ea_ops = {
7665         .dio_lookup         = osd_index_ea_lookup,
7666         .dio_declare_insert = osd_index_declare_ea_insert,
7667         .dio_insert         = osd_index_ea_insert,
7668         .dio_declare_delete = osd_index_declare_ea_delete,
7669         .dio_delete         = osd_index_ea_delete,
7670         .dio_it     = {
7671                 .init     = osd_it_ea_init,
7672                 .fini     = osd_it_ea_fini,
7673                 .get      = osd_it_ea_get,
7674                 .put      = osd_it_ea_put,
7675                 .next     = osd_it_ea_next,
7676                 .key      = osd_it_ea_key,
7677                 .key_size = osd_it_ea_key_size,
7678                 .rec      = osd_it_ea_rec,
7679                 .rec_size = osd_it_ea_rec_size,
7680                 .store    = osd_it_ea_store,
7681                 .load     = osd_it_ea_load
7682         }
7683 };
7684
7685 static void *osd_key_init(const struct lu_context *ctx,
7686                           struct lu_context_key *key)
7687 {
7688         struct osd_thread_info *info;
7689
7690         OBD_ALLOC_PTR(info);
7691         if (info == NULL)
7692                 return ERR_PTR(-ENOMEM);
7693
7694         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7695         if (info->oti_it_ea_buf == NULL)
7696                 goto out_free_info;
7697
7698         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7699
7700         info->oti_hlock = ldiskfs_htree_lock_alloc();
7701         if (info->oti_hlock == NULL)
7702                 goto out_free_ea;
7703
7704         return info;
7705
7706 out_free_ea:
7707         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7708 out_free_info:
7709         OBD_FREE_PTR(info);
7710         return ERR_PTR(-ENOMEM);
7711 }
7712
7713 static void osd_key_fini(const struct lu_context *ctx,
7714                          struct lu_context_key *key, void *data)
7715 {
7716         struct osd_thread_info *info = data;
7717         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7718         struct osd_idmap_cache *idc = info->oti_ins_cache;
7719
7720         if (info->oti_dio_pages) {
7721                 int i;
7722                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7723                         struct page *page = info->oti_dio_pages[i];
7724                         if (page) {
7725                                 LASSERT(PagePrivate2(page));
7726                                 LASSERT(PageLocked(page));
7727                                 ClearPagePrivate2(page);
7728                                 unlock_page(page);
7729                                 __free_page(page);
7730                         }
7731                 }
7732                 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7733                                          PTLRPC_MAX_BRW_PAGES);
7734         }
7735
7736         if (info->oti_inode != NULL)
7737                 OBD_FREE_PTR(lli);
7738         if (info->oti_hlock != NULL)
7739                 ldiskfs_htree_lock_free(info->oti_hlock);
7740         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7741         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7742         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7743         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7744         lu_buf_free(&info->oti_big_buf);
7745         if (idc != NULL) {
7746                 LASSERT(info->oti_ins_cache_size > 0);
7747                 OBD_FREE_PTR_ARRAY_LARGE(idc, info->oti_ins_cache_size);
7748                 info->oti_ins_cache = NULL;
7749                 info->oti_ins_cache_size = 0;
7750         }
7751         OBD_FREE_PTR(info);
7752 }
7753
7754 static void osd_key_exit(const struct lu_context *ctx,
7755                          struct lu_context_key *key, void *data)
7756 {
7757         struct osd_thread_info *info = data;
7758
7759         LASSERT(info->oti_r_locks == 0);
7760         LASSERT(info->oti_w_locks == 0);
7761         LASSERT(info->oti_txns    == 0);
7762         LASSERTF(info->oti_dio_pages_used == 0, "%d\n",
7763                  info->oti_dio_pages_used);
7764 }
7765
7766 /* type constructor/destructor: osd_type_init, osd_type_fini */
7767 LU_TYPE_INIT_FINI(osd, &osd_key);
7768
7769 struct lu_context_key osd_key = {
7770         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7771         .lct_init = osd_key_init,
7772         .lct_fini = osd_key_fini,
7773         .lct_exit = osd_key_exit
7774 };
7775
7776
7777 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7778                            const char *name, struct lu_device *next)
7779 {
7780         struct osd_device *osd = osd_dev(d);
7781
7782         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7783             sizeof(osd->od_svname))
7784                 return -E2BIG;
7785         return osd_procfs_init(osd, name);
7786 }
7787
7788 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7789 {
7790         struct seq_server_site *ss = osd_seq_site(osd);
7791         int rc = 0;
7792
7793         ENTRY;
7794
7795         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7796                 RETURN(0);
7797
7798         if (unlikely(ss == NULL))
7799                 RETURN(-ENODEV);
7800
7801         OBD_ALLOC_PTR(osd->od_cl_seq);
7802         if (osd->od_cl_seq == NULL)
7803                 RETURN(-ENOMEM);
7804
7805         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7806                         osd->od_svname, ss->ss_server_seq);
7807
7808         if (ss->ss_node_id == 0) {
7809                 /*
7810                  * If the OSD on the sequence controller(MDT0), then allocate
7811                  * sequence here, otherwise allocate sequence after connected
7812                  * to MDT0 (see mdt_register_lwp_callback()).
7813                  */
7814                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7815                                    &osd->od_cl_seq->lcs_space, env);
7816         }
7817
7818         RETURN(rc);
7819 }
7820
7821 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7822 {
7823         if (osd->od_cl_seq == NULL)
7824                 return;
7825
7826         seq_client_fini(osd->od_cl_seq);
7827         OBD_FREE_PTR(osd->od_cl_seq);
7828         osd->od_cl_seq = NULL;
7829 }
7830
7831 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7832 {
7833         ENTRY;
7834
7835         /* shutdown quota slave instance associated with the device */
7836         if (o->od_quota_slave_md != NULL) {
7837                 struct qsd_instance *qsd = o->od_quota_slave_md;
7838
7839                 o->od_quota_slave_md = NULL;
7840                 qsd_fini(env, qsd);
7841         }
7842
7843         if (o->od_quota_slave_dt != NULL) {
7844                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7845
7846                 o->od_quota_slave_dt = NULL;
7847                 qsd_fini(env, qsd);
7848         }
7849
7850         osd_fid_fini(env, o);
7851         osd_scrub_cleanup(env, o);
7852
7853         RETURN(0);
7854 }
7855
7856 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7857 {
7858         ENTRY;
7859
7860         if (o->od_mnt != NULL) {
7861                 shrink_dcache_sb(osd_sb(o));
7862                 osd_sync(env, &o->od_dt_dev);
7863                 wait_event(o->od_commit_cb_done,
7864                           !atomic_read(&o->od_commit_cb_in_flight));
7865
7866                 mntput(o->od_mnt);
7867                 o->od_mnt = NULL;
7868         }
7869
7870         EXIT;
7871 }
7872
7873 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7874 # ifndef LDISKFS_HAS_INCOMPAT_FEATURE
7875 /* Newer kernels provide the ldiskfs_set_feature_largedir() wrapper already,
7876  * which calls ldiskfs_update_dynamic_rev() to update ancient filesystems.
7877  * All ldiskfs filesystems are already v2, so it is a no-op and unnecessary.
7878  * This avoids maintaining patches to export this otherwise-useless function.
7879  */
7880 void ldiskfs_update_dynamic_rev(struct super_block *sb)
7881 {
7882         /* do nothing */
7883 }
7884 # endif
7885 #endif
7886
7887 static int osd_mount(const struct lu_env *env,
7888                      struct osd_device *o, struct lustre_cfg *cfg)
7889 {
7890         const char *name = lustre_cfg_string(cfg, 0);
7891         const char *dev = lustre_cfg_string(cfg, 1);
7892         const char *opts;
7893         unsigned long page, s_flags = 0, lmd_flags = 0;
7894         struct page *__page;
7895         struct file_system_type *type;
7896         char *options = NULL;
7897         const char *str;
7898         struct osd_thread_info *info = osd_oti_get(env);
7899         struct lu_fid *fid = &info->oti_fid;
7900         struct inode *inode;
7901         int rc = 0, force_over_1024tb = 0;
7902
7903         ENTRY;
7904
7905         if (o->od_mnt != NULL)
7906                 RETURN(0);
7907
7908         if (strlen(dev) >= sizeof(o->od_mntdev))
7909                 RETURN(-E2BIG);
7910         strcpy(o->od_mntdev, dev);
7911
7912         str = lustre_cfg_buf(cfg, 2);
7913         sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
7914
7915         opts = lustre_cfg_string(cfg, 3);
7916 #ifdef __BIG_ENDIAN
7917         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7918                 CERROR("%s: device %s extents feature is not guaranteed to "
7919                        "work on big-endian systems. Use \"bigendian_extents\" "
7920                        "mount option to override.\n", name, dev);
7921                 RETURN(-EINVAL);
7922         }
7923 #endif
7924 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7925         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7926                 CWARN("force_over_128tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7927         }
7928 #endif
7929 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7930         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7931                 CWARN("force_over_256tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7932         }
7933 #endif
7934 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7935         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7936                 CWARN("force_over_512tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7937         }
7938 #endif
7939
7940         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7941                 force_over_1024tb = 1;
7942
7943         __page = alloc_page(GFP_KERNEL);
7944         if (__page == NULL)
7945                 GOTO(out, rc = -ENOMEM);
7946         page = (unsigned long)page_address(__page);
7947         options = (char *)page;
7948         *options = '\0';
7949         if (opts != NULL) {
7950                 /* strip out the options for back compatiblity */
7951                 static const char * const sout[] = {
7952                         "mballoc",
7953                         "iopen",
7954                         "noiopen",
7955                         "iopen_nopriv",
7956                         "extents",
7957                         "noextents",
7958                         /* strip out option we processed in osd */
7959                         "bigendian_extents",
7960                         "force_over_128tb",
7961                         "force_over_256tb",
7962                         "force_over_512tb",
7963                         "force_over_1024tb",
7964                         "resetoi",
7965                         NULL
7966                 };
7967                 strncat(options, opts, PAGE_SIZE);
7968                 for (rc = 0, str = options; sout[rc]; ) {
7969                         char *op = strstr(str, sout[rc]);
7970
7971                         if (op == NULL) {
7972                                 rc++;
7973                                 str = options;
7974                                 continue;
7975                         }
7976                         if (op == options || *(op - 1) == ',') {
7977                                 str = op + strlen(sout[rc]);
7978                                 if (*str == ',' || *str == '\0') {
7979                                         *str == ',' ? str++ : str;
7980                                         memmove(op, str, strlen(str) + 1);
7981                                 }
7982                         }
7983                         for (str = op; *str != ',' && *str != '\0'; str++)
7984                                 ;
7985                 }
7986         } else {
7987                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7988         }
7989
7990         /* Glom up mount options */
7991         if (*options != '\0')
7992                 strncat(options, ",", PAGE_SIZE);
7993         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7994
7995         type = get_fs_type("ldiskfs");
7996         if (!type) {
7997                 CERROR("%s: cannot find ldiskfs module\n", name);
7998                 GOTO(out, rc = -ENODEV);
7999         }
8000
8001         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
8002         module_put(type->owner);
8003
8004         if (IS_ERR(o->od_mnt)) {
8005                 rc = PTR_ERR(o->od_mnt);
8006                 o->od_mnt = NULL;
8007                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
8008                 GOTO(out, rc);
8009         }
8010
8011         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
8012                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
8013                                  force_over_1024tb == 0) {
8014                 CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
8015                        name, dev);
8016                 GOTO(out_mnt, rc = -EINVAL);
8017         }
8018
8019         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
8020                 LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
8021                               name);
8022
8023                 GOTO(out_mnt, rc = -EOPNOTSUPP);
8024         }
8025
8026         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
8027                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
8028                 GOTO(out_mnt, rc = -EINVAL);
8029         }
8030
8031         if (ldiskfs_has_feature_fast_commit(o->od_mnt->mnt_sb)) {
8032                 CERROR("%s: device %s is mounted with fast_commit that breaks recovery\n",
8033                        name, dev);
8034                 GOTO(out_mnt, rc = -EOPNOTSUPP);
8035         }
8036
8037 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
8038 #ifdef LDISKFS_MOUNT_DIRDATA
8039         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
8040                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
8041         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
8042                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
8043                       "enabling the dirdata feature. If you do not want to "
8044                       "downgrade to Lustre-1.x again, you can enable it via "
8045                       "'tune2fs -O dirdata device'\n", name, dev);
8046 #endif
8047         /* enable large_dir on MDTs to avoid REMOTE_PARENT_DIR overflow,
8048          * and on very large OSTs to avoid object directory overflow */
8049         if (unlikely(!ldiskfs_has_feature_largedir(o->od_mnt->mnt_sb) &&
8050                      !strstr(name, "MGS"))) {
8051                 ldiskfs_set_feature_largedir(o->od_mnt->mnt_sb);
8052                 LCONSOLE_INFO("%s: enabled 'large_dir' feature on device %s\n",
8053                               name, dev);
8054         }
8055 #endif
8056         inode = osd_sb(o)->s_root->d_inode;
8057         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
8058         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
8059         if (rc != 0) {
8060                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
8061                 GOTO(out_mnt, rc);
8062         }
8063
8064         if (lmd_flags & LMD_FLG_NOSCRUB)
8065                 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_NEVER;
8066
8067         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
8068                 /* do not use pagecache with flash-backed storage */
8069                 o->od_writethrough_cache = 0;
8070                 o->od_read_cache = 0;
8071         }
8072
8073         GOTO(out, rc = 0);
8074
8075 out_mnt:
8076         mntput(o->od_mnt);
8077         o->od_mnt = NULL;
8078
8079 out:
8080         if (__page)
8081                 __free_page(__page);
8082
8083         return rc;
8084 }
8085
8086 static struct lu_device *osd_device_fini(const struct lu_env *env,
8087                                          struct lu_device *d)
8088 {
8089         struct osd_device *o = osd_dev(d);
8090
8091         ENTRY;
8092
8093         osd_index_backup(env, o, false);
8094         osd_shutdown(env, o);
8095         osd_procfs_fini(o);
8096         if (o->od_oi_table != NULL)
8097                 osd_oi_fini(osd_oti_get(env), o);
8098         if (o->od_extent_bytes_percpu)
8099                 free_percpu(o->od_extent_bytes_percpu);
8100         osd_obj_map_fini(o);
8101         osd_umount(env, o);
8102
8103         RETURN(NULL);
8104 }
8105
8106 static int osd_device_init0(const struct lu_env *env,
8107                             struct osd_device *o,
8108                             struct lustre_cfg *cfg)
8109 {
8110         struct lu_device *l = osd2lu_dev(o);
8111         struct osd_thread_info *info;
8112         int cplen = 0;
8113         char *opts = NULL;
8114         bool restored = false;
8115         int rc;
8116
8117         /* if the module was re-loaded, env can loose its keys */
8118         rc = lu_env_refill((struct lu_env *)env);
8119         if (rc)
8120                 GOTO(out, rc);
8121         info = osd_oti_get(env);
8122         LASSERT(info);
8123
8124         l->ld_ops = &osd_lu_ops;
8125         o->od_dt_dev.dd_ops = &osd_dt_ops;
8126
8127         spin_lock_init(&o->od_osfs_lock);
8128         mutex_init(&o->od_otable_mutex);
8129         INIT_LIST_HEAD(&o->od_orphan_list);
8130         INIT_LIST_HEAD(&o->od_index_backup_list);
8131         INIT_LIST_HEAD(&o->od_index_restore_list);
8132         spin_lock_init(&o->od_lock);
8133         o->od_index_backup_policy = LIBP_NONE;
8134         o->od_t10_type = 0;
8135         init_waitqueue_head(&o->od_commit_cb_done);
8136
8137         o->od_read_cache = 1;
8138         o->od_writethrough_cache = 1;
8139         o->od_enable_projid_xattr = 0;
8140         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
8141         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
8142         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
8143         o->od_scrub.os_scrub.os_auto_scrub_interval = AS_DEFAULT;
8144         /* default fallocate to unwritten extents: LU-14326/LU-14333 */
8145         o->od_fallocate_zero_blocks = 0;
8146
8147         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
8148                         sizeof(o->od_svname));
8149         if (cplen >= sizeof(o->od_svname)) {
8150                 rc = -E2BIG;
8151                 GOTO(out, rc);
8152         }
8153
8154         o->od_index_backup_stop = 0;
8155         o->od_index = -1; /* -1 means index is invalid */
8156         rc = server_name2index(o->od_svname, &o->od_index, NULL);
8157         if (rc == LDD_F_SV_TYPE_OST)
8158                 o->od_is_ost = 1;
8159
8160         o->od_full_scrub_ratio = OFSR_DEFAULT;
8161         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
8162         rc = osd_mount(env, o, cfg);
8163         if (rc != 0)
8164                 GOTO(out, rc);
8165
8166         /* Can only check block device after mount */
8167         o->od_nonrotational =
8168                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
8169
8170         rc = osd_obj_map_init(env, o);
8171         if (rc != 0)
8172                 GOTO(out_mnt, rc);
8173
8174         rc = lu_site_init(&o->od_site, l);
8175         if (rc != 0)
8176                 GOTO(out_compat, rc);
8177         o->od_site.ls_bottom_dev = l;
8178
8179         rc = lu_site_init_finish(&o->od_site);
8180         if (rc != 0)
8181                 GOTO(out_site, rc);
8182
8183         opts = lustre_cfg_string(cfg, 3);
8184         if (opts && strstr(opts, "resetoi"))
8185                 restored = true;
8186
8187         INIT_LIST_HEAD(&o->od_ios_list);
8188
8189         rc = lprocfs_init_brw_stats(&o->od_brw_stats);
8190         if (rc)
8191                 GOTO(out_brw_stats, rc);
8192
8193         /* setup scrub, including OI files initialization */
8194         o->od_in_init = 1;
8195         rc = osd_scrub_setup(env, o, restored);
8196         o->od_in_init = 0;
8197         if (rc < 0)
8198                 GOTO(out_brw_stats, rc);
8199
8200         rc = osd_procfs_init(o, o->od_svname);
8201         if (rc != 0) {
8202                 CERROR("%s: can't initialize procfs: rc = %d\n",
8203                        o->od_svname, rc);
8204                 GOTO(out_scrub, rc);
8205         }
8206
8207         LASSERT(l->ld_site->ls_linkage.next != NULL);
8208         LASSERT(l->ld_site->ls_linkage.prev != NULL);
8209
8210         /* initialize quota slave instance */
8211         /* currently it's no need to prepare qsd_instance_md for OST */
8212         if (!o->od_is_ost) {
8213                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
8214                                                 &o->od_dt_dev, o->od_proc_entry,
8215                                                 true, true);
8216                 if (IS_ERR(o->od_quota_slave_md)) {
8217                         rc = PTR_ERR(o->od_quota_slave_md);
8218                         o->od_quota_slave_md = NULL;
8219                         GOTO(out_procfs, rc);
8220                 }
8221         }
8222
8223         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
8224                                         o->od_proc_entry, false, true);
8225
8226         if (IS_ERR(o->od_quota_slave_dt)) {
8227                 if (o->od_quota_slave_md != NULL) {
8228                         qsd_fini(env, o->od_quota_slave_md);
8229                         o->od_quota_slave_md = NULL;
8230                 }
8231
8232                 rc = PTR_ERR(o->od_quota_slave_dt);
8233                 o->od_quota_slave_dt = NULL;
8234                 GOTO(out_procfs, rc);
8235         }
8236
8237         o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
8238         if (!o->od_extent_bytes_percpu) {
8239                 rc = -ENOMEM;
8240                 GOTO(out_procfs, rc);
8241         }
8242
8243         RETURN(0);
8244
8245 out_procfs:
8246         osd_procfs_fini(o);
8247 out_scrub:
8248         osd_scrub_cleanup(env, o);
8249 out_brw_stats:
8250         lprocfs_fini_brw_stats(&o->od_brw_stats);
8251 out_site:
8252         lu_site_fini(&o->od_site);
8253 out_compat:
8254         osd_obj_map_fini(o);
8255 out_mnt:
8256         osd_umount(env, o);
8257 out:
8258         return rc;
8259 }
8260
8261 static struct lu_device *osd_device_alloc(const struct lu_env *env,
8262                                           struct lu_device_type *t,
8263                                           struct lustre_cfg *cfg)
8264 {
8265         struct osd_device *o;
8266         int rc;
8267
8268         OBD_ALLOC_PTR(o);
8269         if (o == NULL)
8270                 return ERR_PTR(-ENOMEM);
8271
8272         rc = dt_device_init(&o->od_dt_dev, t);
8273         if (rc == 0) {
8274                 /*
8275                  * Because the ctx might be revived in dt_device_init,
8276                  * refill the env here
8277                  */
8278                 lu_env_refill((struct lu_env *)env);
8279                 rc = osd_device_init0(env, o, cfg);
8280                 if (rc)
8281                         dt_device_fini(&o->od_dt_dev);
8282         }
8283
8284         if (unlikely(rc != 0))
8285                 OBD_FREE_PTR(o);
8286
8287         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
8288 }
8289
8290 static struct lu_device *osd_device_free(const struct lu_env *env,
8291                                          struct lu_device *d)
8292 {
8293         struct osd_device *o = osd_dev(d);
8294
8295         ENTRY;
8296
8297         /* XXX: make osd top device in order to release reference */
8298         d->ld_site->ls_top_dev = d;
8299         lu_site_purge(env, d->ld_site, -1);
8300         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8301                       D_ERROR, lu_cdebug_printer);
8302         lu_site_fini(&o->od_site);
8303         dt_device_fini(&o->od_dt_dev);
8304         OBD_FREE_PTR(o);
8305         RETURN(NULL);
8306 }
8307
8308 static int osd_process_config(const struct lu_env *env,
8309                               struct lu_device *d, struct lustre_cfg *cfg)
8310 {
8311         struct osd_device *o = osd_dev(d);
8312         ssize_t count;
8313         int rc;
8314
8315         ENTRY;
8316
8317         switch (cfg->lcfg_command) {
8318         case LCFG_SETUP:
8319                 rc = osd_mount(env, o, cfg);
8320                 break;
8321         case LCFG_CLEANUP:
8322                 /*
8323                  * For the case LCFG_PRE_CLEANUP is not called in advance,
8324                  * that may happend if hit failure during mount process.
8325                  */
8326                 osd_index_backup(env, o, false);
8327                 lu_dev_del_linkage(d->ld_site, d);
8328                 rc = osd_shutdown(env, o);
8329                 break;
8330         case LCFG_PARAM:
8331                 LASSERT(&o->od_dt_dev);
8332                 count  = class_modify_config(cfg, PARAM_OSD,
8333                                              &o->od_dt_dev.dd_kobj);
8334                 if (count < 0)
8335                         count = class_modify_config(cfg, PARAM_OST,
8336                                                     &o->od_dt_dev.dd_kobj);
8337                 rc = count > 0 ? 0 : count;
8338                 break;
8339         case LCFG_PRE_CLEANUP:
8340                 osd_scrub_stop(o);
8341                 osd_index_backup(env, o,
8342                                  o->od_index_backup_policy != LIBP_NONE);
8343                 rc = 0;
8344                 break;
8345         default:
8346                 rc = -ENOSYS;
8347         }
8348
8349         RETURN(rc);
8350 }
8351
8352 static int osd_recovery_complete(const struct lu_env *env,
8353                                  struct lu_device *d)
8354 {
8355         struct osd_device *osd = osd_dev(d);
8356         int rc = 0;
8357
8358         ENTRY;
8359
8360         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8361                 RETURN(0);
8362
8363         /*
8364          * start qsd instance on recovery completion, this notifies the quota
8365          * slave code that we are about to process new requests now
8366          */
8367         rc = qsd_start(env, osd->od_quota_slave_dt);
8368         if (rc == 0 && osd->od_quota_slave_md != NULL)
8369                 rc = qsd_start(env, osd->od_quota_slave_md);
8370
8371         RETURN(rc);
8372 }
8373
8374 /*
8375  * we use exports to track all osd users
8376  */
8377 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8378                            struct obd_device *obd, struct obd_uuid *cluuid,
8379                            struct obd_connect_data *data, void *localdata)
8380 {
8381         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8382         struct lustre_handle conn;
8383         int rc;
8384
8385         ENTRY;
8386
8387         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8388
8389         rc = class_connect(&conn, obd, cluuid);
8390         if (rc)
8391                 RETURN(rc);
8392
8393         *exp = class_conn2export(&conn);
8394
8395         spin_lock(&osd->od_osfs_lock);
8396         osd->od_connects++;
8397         spin_unlock(&osd->od_osfs_lock);
8398
8399         RETURN(0);
8400 }
8401
8402 /*
8403  * once last export (we don't count self-export) disappeared
8404  * osd can be released
8405  */
8406 static int osd_obd_disconnect(struct obd_export *exp)
8407 {
8408         struct obd_device *obd = exp->exp_obd;
8409         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8410         int rc, release = 0;
8411
8412         ENTRY;
8413
8414         /* Only disconnect the underlying layers on the final disconnect. */
8415         spin_lock(&osd->od_osfs_lock);
8416         osd->od_connects--;
8417         if (osd->od_connects == 0)
8418                 release = 1;
8419         spin_unlock(&osd->od_osfs_lock);
8420
8421         rc = class_disconnect(exp); /* bz 9811 */
8422
8423         if (rc == 0 && release)
8424                 class_manual_cleanup(obd);
8425         RETURN(rc);
8426 }
8427
8428 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8429                        struct lu_device *dev)
8430 {
8431         struct osd_device *osd = osd_dev(dev);
8432         struct lr_server_data *lsd =
8433                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8434         int result = 0;
8435
8436         ENTRY;
8437
8438         if (osd->od_quota_slave_md != NULL) {
8439                 /* set up quota slave objects for inode */
8440                 result = qsd_prepare(env, osd->od_quota_slave_md);
8441                 if (result != 0)
8442                         RETURN(result);
8443         }
8444
8445         if (osd->od_quota_slave_dt != NULL) {
8446                 /* set up quota slave objects for block */
8447                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8448                 if (result != 0)
8449                         RETURN(result);
8450         }
8451
8452
8453         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8454 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8455                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8456                         osd->od_index_in_idif = 1;
8457                 } else {
8458                         osd->od_index_in_idif = 0;
8459                         result = osd_register_proc_index_in_idif(osd);
8460                         if (result != 0)
8461                                 RETURN(result);
8462                 }
8463 #else
8464                 osd->od_index_in_idif = 1;
8465 #endif
8466         }
8467
8468         result = osd_fid_init(env, osd);
8469
8470         RETURN(result);
8471 }
8472
8473 /**
8474  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8475  *
8476  * Allocate FID.
8477  *
8478  * see include/lu_object.h for the details.
8479  */
8480 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8481                          struct lu_fid *fid, struct lu_object *parent,
8482                          const struct lu_name *name)
8483 {
8484         struct osd_device *osd = osd_dev(d);
8485
8486         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8487 }
8488
8489 static const struct lu_object_operations osd_lu_obj_ops = {
8490         .loo_object_init      = osd_object_init,
8491         .loo_object_delete    = osd_object_delete,
8492         .loo_object_release   = osd_object_release,
8493         .loo_object_free      = osd_object_free,
8494         .loo_object_print     = osd_object_print,
8495         .loo_object_invariant = osd_object_invariant
8496 };
8497
8498 const struct lu_device_operations osd_lu_ops = {
8499         .ldo_object_alloc      = osd_object_alloc,
8500         .ldo_process_config    = osd_process_config,
8501         .ldo_recovery_complete = osd_recovery_complete,
8502         .ldo_prepare           = osd_prepare,
8503         .ldo_fid_alloc         = osd_fid_alloc,
8504 };
8505
8506 static const struct lu_device_type_operations osd_device_type_ops = {
8507         .ldto_init = osd_type_init,
8508         .ldto_fini = osd_type_fini,
8509
8510         .ldto_start = osd_type_start,
8511         .ldto_stop  = osd_type_stop,
8512
8513         .ldto_device_alloc = osd_device_alloc,
8514         .ldto_device_free  = osd_device_free,
8515
8516         .ldto_device_init = osd_device_init,
8517         .ldto_device_fini = osd_device_fini
8518 };
8519
8520 static struct lu_device_type osd_device_type = {
8521         .ldt_tags     = LU_DEVICE_DT,
8522         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8523         .ldt_ops      = &osd_device_type_ops,
8524         .ldt_ctx_tags = LCT_LOCAL,
8525 };
8526
8527 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8528 {
8529         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8530         struct super_block *sb = osd_sb(osd);
8531
8532         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8533 }
8534
8535 /*
8536  * lprocfs legacy support.
8537  */
8538 static const struct obd_ops osd_obd_device_ops = {
8539         .o_owner = THIS_MODULE,
8540         .o_connect      = osd_obd_connect,
8541         .o_disconnect   = osd_obd_disconnect,
8542         .o_health_check = osd_health_check,
8543 };
8544
8545 static ssize_t track_declares_assert_show(struct kobject *kobj,
8546                                    struct attribute *attr,
8547                                    char *buf)
8548 {
8549         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8550 }
8551
8552 static ssize_t track_declares_assert_store(struct kobject *kobj,
8553                                            struct attribute *attr,
8554                                            const char *buffer, size_t count)
8555 {
8556         bool track_declares_assert;
8557         int rc;
8558
8559         rc = kstrtobool(buffer, &track_declares_assert);
8560         if (rc)
8561                 return rc;
8562
8563         ldiskfs_track_declares_assert = track_declares_assert;
8564
8565         return count;
8566 }
8567 LUSTRE_RW_ATTR(track_declares_assert);
8568
8569 static int __init osd_init(void)
8570 {
8571         struct kobject *kobj;
8572         int rc;
8573
8574         BUILD_BUG_ON(BH_DXLock >=
8575                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8576 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8577         /* please, try to keep osd_thread_info smaller than a page */
8578         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8579 #endif
8580
8581         osd_oi_mod_init();
8582
8583         rc = lu_kmem_init(ldiskfs_caches);
8584         if (rc)
8585                 return rc;
8586
8587         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8588                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8589         if (rc) {
8590                 lu_kmem_fini(ldiskfs_caches);
8591                 return rc;
8592         }
8593
8594         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8595         if (kobj) {
8596                 rc = sysfs_create_file(kobj,
8597                                        &lustre_attr_track_declares_assert.attr);
8598                 kobject_put(kobj);
8599                 if (rc) {
8600                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8601                         rc = 0;
8602                 }
8603         }
8604
8605         return rc;
8606 }
8607
8608 static void __exit osd_exit(void)
8609 {
8610         struct kobject *kobj;
8611
8612         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8613         if (kobj) {
8614                 sysfs_remove_file(kobj,
8615                                   &lustre_attr_track_declares_assert.attr);
8616                 kobject_put(kobj);
8617         }
8618         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8619         lu_kmem_fini(ldiskfs_caches);
8620 }
8621
8622 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8623 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8624 MODULE_VERSION(LUSTRE_VERSION_STRING);
8625 MODULE_LICENSE("GPL");
8626
8627 module_init(osd_init);
8628 module_exit(osd_exit);