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