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