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, 2015, 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/module.h>
  43 #include <linux/user_namespace.h>
  44 #ifdef HAVE_UIDGID_HEADER
  45 # include <linux/uidgid.h>
  46 #endif
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 #include <ldiskfs/ldiskfs.h>
  58 #include <ldiskfs/xattr.h>
  59 #include <ldiskfs/ldiskfs_extents.h>
  60 #undef ENTRY
  61 /*
  62  * struct OBD_{ALLOC,FREE}*()
  63  * OBD_FAIL_CHECK
  64  */
  65 #include <obd_support.h>
  66 /* struct ptlrpc_thread */
  67 #include <lustre_net.h>
  68 #include <lustre_fid.h>
  69 /* process_config */
  70 #include <lustre_param.h>
  71
  72 #include "osd_internal.h"
  73 #include "osd_dynlocks.h"
  74
  75 /* llo_* api support */
  76 #include <md_object.h>
  77 #include <lustre_quota.h>
  78
  79 #include <lustre_linkea.h>
  80
  81 int ldiskfs_pdo = 1;
  82 module_param(ldiskfs_pdo, int, 0644);
  83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  84
  85 int ldiskfs_track_declares_assert;
  86 module_param(ldiskfs_track_declares_assert, int, 0644);
  87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  88
  89 /* Slab to allocate dynlocks */
  90 struct kmem_cache *dynlock_cachep;
  91
  92 /* Slab to allocate osd_it_ea */
  93 struct kmem_cache *osd_itea_cachep;
  94
  95 static struct lu_kmem_descr ldiskfs_caches[] = {
  96         {
  97                 .ckd_cache = &dynlock_cachep,
  98                 .ckd_name  = "dynlock_cache",
  99                 .ckd_size  = sizeof(struct dynlock_handle)
 100         },
 101         {
 102                 .ckd_cache = &osd_itea_cachep,
 103                 .ckd_name  = "osd_itea_cache",
 104                 .ckd_size  = sizeof(struct osd_it_ea)
 105         },
 106         {
 107                 .ckd_cache = NULL
 108         }
 109 };
 110
 111 static const char dot[] = ".";
 112 static const char dotdot[] = "..";
 113 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 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_ea_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154
 155 /*
 156  * Concurrency: doesn't matter
 157  */
 158 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 159 {
 160         struct osd_thread_info *oti = osd_oti_get(env);
 161         return oti->oti_w_locks > 0 && o->oo_owner == env;
 162 }
 163
 164 /*
 165  * Concurrency: doesn't access mutable data
 166  */
 167 static int osd_root_get(const struct lu_env *env,
 168                         struct dt_device *dev, struct lu_fid *f)
 169 {
 170         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 171         return 0;
 172 }
 173
 174 /*
 175  * the following set of functions are used to maintain per-thread
 176  * cache of FID->ino mapping. this mechanism is needed to resolve
 177  * FID to inode at dt_insert() which in turn stores ino in the
 178  * directory entries to keep ldiskfs compatible with ext[34].
 179  * due to locking-originated restrictions we can't lookup ino
 180  * using LU cache (deadlock is possible). lookup using OI is quite
 181  * expensive. so instead we maintain this cache and methods like
 182  * dt_create() fill it. so in the majority of cases dt_insert() is
 183  * able to find needed mapping in lockless manner.
 184  */
 185 static struct osd_idmap_cache *
 186 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 187              const struct lu_fid *fid)
 188 {
 189         struct osd_thread_info  *oti   = osd_oti_get(env);
 190         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
 191         int i;
 192         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 193                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 194                         continue;
 195                 if (idc[i].oic_dev != osd)
 196                         continue;
 197
 198                 return idc + i;
 199         }
 200
 201         return NULL;
 202 }
 203
 204 static struct osd_idmap_cache *
 205 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 206             const struct lu_fid *fid)
 207 {
 208         struct osd_thread_info  *oti   = osd_oti_get(env);
 209         struct osd_idmap_cache  *idc;
 210         int i;
 211
 212         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 213                 i = oti->oti_ins_cache_size * 2;
 214                 if (i == 0)
 215                         i = OSD_INS_CACHE_SIZE;
 216                 OBD_ALLOC(idc, sizeof(*idc) * i);
 217                 if (idc == NULL)
 218                         return ERR_PTR(-ENOMEM);
 219                 if (oti->oti_ins_cache != NULL) {
 220                         memcpy(idc, oti->oti_ins_cache,
 221                                oti->oti_ins_cache_used * sizeof(*idc));
 222                         OBD_FREE(oti->oti_ins_cache,
 223                                  oti->oti_ins_cache_used * sizeof(*idc));
 224                 }
 225                 oti->oti_ins_cache = idc;
 226                 oti->oti_ins_cache_size = i;
 227         }
 228
 229         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 230         idc->oic_fid = *fid;
 231         idc->oic_dev = osd;
 232         idc->oic_lid.oii_ino = 0;
 233         idc->oic_lid.oii_gen = 0;
 234         idc->oic_remote = 0;
 235
 236         return idc;
 237 }
 238
 239 /*
 240  * lookup mapping for the given fid in the cache, initialize a
 241  * new one if not found. the initialization checks whether the
 242  * object is local or remote. for local objects, OI is used to
 243  * learn ino/generation. the function is used when the caller
 244  * has no information about the object, e.g. at dt_insert().
 245  */
 246 static struct osd_idmap_cache *
 247 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 248                      const struct lu_fid *fid)
 249 {
 250         struct osd_idmap_cache *idc;
 251         int rc;
 252
 253         idc = osd_idc_find(env, osd, fid);
 254         LASSERT(!IS_ERR(idc));
 255         if (idc != NULL)
 256                 return idc;
 257
 258         /* new mapping is needed */
 259         idc = osd_idc_add(env, osd, fid);
 260         if (IS_ERR(idc))
 261                 return idc;
 262
 263         /* initialize it */
 264         rc = osd_remote_fid(env, osd, fid);
 265         if (unlikely(rc < 0))
 266                 return ERR_PTR(rc);
 267
 268         if (rc == 0) {
 269                 /* the object is local, lookup in OI */
 270                 /* XXX: probably cheaper to lookup in LU first? */
 271                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 272                                    &idc->oic_lid, 0);
 273                 if (unlikely(rc < 0)) {
 274                         CERROR("can't lookup: rc = %d\n", rc);
 275                         return ERR_PTR(rc);
 276                 }
 277         } else {
 278                 /* the object is remote */
 279                 idc->oic_remote = 1;
 280         }
 281
 282         return idc;
 283 }
 284
 285 /*
 286  * lookup mapping for given FID and fill it from the given object.
 287  * the object is lolcal by definition.
 288  */
 289 static int osd_idc_find_and_init(const struct lu_env *env,
 290                                  struct osd_device *osd,
 291                                  struct osd_object *obj)
 292 {
 293         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 294         struct osd_idmap_cache  *idc;
 295
 296         idc = osd_idc_find(env, osd, fid);
 297         LASSERT(!IS_ERR(idc));
 298         if (idc != NULL) {
 299                 if (obj->oo_inode == NULL)
 300                         return 0;
 301                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 302                         LASSERT(idc->oic_lid.oii_ino == 0);
 303                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 304                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 305                 }
 306                 return 0;
 307         }
 308
 309         /* new mapping is needed */
 310         idc = osd_idc_add(env, osd, fid);
 311         if (IS_ERR(idc))
 312                 return PTR_ERR(idc);
 313
 314         if (obj->oo_inode != NULL) {
 315                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 316                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 317         }
 318         return 0;
 319 }
 320
 321 /*
 322  * OSD object methods.
 323  */
 324
 325 /*
 326  * Concurrency: no concurrent access is possible that early in object
 327  * life-cycle.
 328  */
 329 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 330                                           const struct lu_object_header *hdr,
 331                                           struct lu_device *d)
 332 {
 333         struct osd_object *mo;
 334
 335         OBD_ALLOC_PTR(mo);
 336         if (mo != NULL) {
 337                 struct lu_object *l;
 338
 339                 l = &mo->oo_dt.do_lu;
 340                 dt_object_init(&mo->oo_dt, NULL, d);
 341                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 342                 l->lo_ops = &osd_lu_obj_ops;
 343                 init_rwsem(&mo->oo_sem);
 344                 init_rwsem(&mo->oo_ext_idx_sem);
 345                 spin_lock_init(&mo->oo_guard);
 346                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 347                 return l;
 348         } else {
 349                 return NULL;
 350         }
 351 }
 352
 353 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 354                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 355 {
 356         int rc;
 357
 358         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 359         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 360                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 361         if (rc > 0) {
 362                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 363                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 364                 rc = 0;
 365                 lustre_lma_swab(lma);
 366                 /* Check LMA compatibility */
 367                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 368                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 369                               "for fid = "DFID", ino = %lu\n",
 370                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 371                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 372                               PFID(&lma->lma_self_fid), inode->i_ino);
 373                         rc = -EOPNOTSUPP;
 374                 }
 375         } else if (rc == 0) {
 376                 rc = -ENODATA;
 377         }
 378
 379         return rc;
 380 }
 381
 382 /*
 383  * retrieve object from backend ext fs.
 384  **/
 385 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 386                        struct osd_inode_id *id)
 387 {
 388         struct inode *inode = NULL;
 389
 390         /* if we look for an inode withing a running
 391          * transaction, then we risk to deadlock */
 392         /* osd_dirent_check_repair() breaks this */
 393         /*LASSERT(current->journal_info == NULL);*/
 394
 395         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 396         if (IS_ERR(inode)) {
 397                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 398                        id->oii_ino, PTR_ERR(inode));
 399         } else if (id->oii_gen != OSD_OII_NOGEN &&
 400                    inode->i_generation != id->oii_gen) {
 401                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 402                        "i_generation = %u\n",
 403                        id->oii_ino, id->oii_gen, inode->i_generation);
 404                 iput(inode);
 405                 inode = ERR_PTR(-ESTALE);
 406         } else if (inode->i_nlink == 0) {
 407                 /* due to parallel readdir and unlink,
 408                 * we can have dead inode here. */
 409                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 410                 iput(inode);
 411                 inode = ERR_PTR(-ESTALE);
 412         } else if (is_bad_inode(inode)) {
 413                 CWARN("%.16s: bad inode: ino = %u\n",
 414                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 415                 iput(inode);
 416                 inode = ERR_PTR(-ENOENT);
 417         } else {
 418                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 419                 if (id->oii_gen == OSD_OII_NOGEN)
 420                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 421
 422                 /* Do not update file c/mtime in ldiskfs.
 423                  * NB: we don't have any lock to protect this because we don't
 424                  * have reference on osd_object now, but contention with
 425                  * another lookup + attr_set can't happen in the tiny window
 426                  * between if (...) and set S_NOCMTIME. */
 427                 if (!(inode->i_flags & S_NOCMTIME))
 428                         inode->i_flags |= S_NOCMTIME;
 429         }
 430         return inode;
 431 }
 432
 433 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 434                           handle_t *handle, struct dentry *child,
 435                           struct inode *inode, struct htree_lock *hlock)
 436 {
 437         int rc, rc2;
 438
 439         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 440         if (rc == -ENOBUFS || rc == -ENOSPC) {
 441                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
 442                 struct inode *parent = child->d_parent->d_inode;
 443                 struct lu_fid *fid = NULL;
 444
 445                 rc2 = osd_get_lma(info, parent, child->d_parent, lma);
 446                 if (rc2 == 0) {
 447                         fid = &lma->lma_self_fid;
 448                 } else if (rc2 == -ENODATA) {
 449                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 450                                 fid = &info->oti_fid3;
 451                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 452                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 453                                 fid = &info->oti_fid3;
 454                                 lu_igif_build(fid, parent->i_ino,
 455                                               parent->i_generation);
 456                         }
 457                 }
 458
 459                 if (fid != NULL)
 460                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 461                               "maximum entry limit\n",
 462                               osd_name(osd), parent->i_ino, PFID(fid),
 463                               rc == -ENOSPC ? "has reached" : "is approaching");
 464                 else
 465                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 466                               "maximum entry limit\n",
 467                               osd_name(osd), parent->i_ino,
 468                               rc == -ENOSPC ? "has reached" : "is approaching");
 469
 470                 /* ignore such error now */
 471                 if (rc == -ENOBUFS)
 472                         rc = 0;
 473         }
 474
 475         return rc;
 476 }
 477
 478
 479 static struct inode *
 480 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 481              struct osd_inode_id *id, struct lu_fid *fid)
 482 {
 483         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 484         struct inode            *inode;
 485         int                      rc;
 486
 487         inode = osd_iget(info, dev, id);
 488         if (IS_ERR(inode))
 489                 return inode;
 490
 491         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 492         if (rc == 0) {
 493                 *fid = lma->lma_self_fid;
 494         } else if (rc == -ENODATA) {
 495                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 496                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 497                 else
 498                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 499         } else {
 500                 iput(inode);
 501                 inode = ERR_PTR(rc);
 502         }
 503         return inode;
 504 }
 505
 506 static struct inode *osd_iget_check(struct osd_thread_info *info,
 507                                     struct osd_device *dev,
 508                                     const struct lu_fid *fid,
 509                                     struct osd_inode_id *id,
 510                                     bool cached)
 511 {
 512         struct inode    *inode;
 513         int              rc     = 0;
 514         ENTRY;
 515
 516         /* The cached OI mapping is trustable. If we cannot locate the inode
 517          * via the cached OI mapping, then return the failure to the caller
 518          * directly without further OI checking. */
 519
 520         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 521         if (IS_ERR(inode)) {
 522                 rc = PTR_ERR(inode);
 523                 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
 524                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 525                                id->oii_ino, rc);
 526
 527                         GOTO(put, rc);
 528                 }
 529
 530                 goto check_oi;
 531         }
 532
 533         if (is_bad_inode(inode)) {
 534                 rc = -ENOENT;
 535                 if (cached) {
 536                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 537
 538                         GOTO(put, rc);
 539                 }
 540
 541                 goto check_oi;
 542         }
 543
 544         if (id->oii_gen != OSD_OII_NOGEN &&
 545             inode->i_generation != id->oii_gen) {
 546                 rc = -ESTALE;
 547                 if (cached) {
 548                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 549                                "oii_gen = %u, i_generation = %u\n",
 550                                id->oii_ino, id->oii_gen, inode->i_generation);
 551
 552                         GOTO(put, rc);
 553                 }
 554
 555                 goto check_oi;
 556         }
 557
 558         if (inode->i_nlink == 0) {
 559                 rc = -ENOENT;
 560                 if (cached) {
 561                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 562
 563                         GOTO(put, rc);
 564                 }
 565
 566                 goto check_oi;
 567         }
 568
 569         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 570
 571 check_oi:
 572         if (rc != 0) {
 573                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 574
 575                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 576                 /* XXX: There are four possible cases:
 577                  *      1. rc = 0.
 578                  *         Backup/restore caused the OI invalid.
 579                  *      2. rc = 0.
 580                  *         Someone unlinked the object but NOT removed
 581                  *         the OI mapping, such as mount target device
 582                  *         as ldiskfs, and modify something directly.
 583                  *      3. rc = -ENOENT.
 584                  *         Someone just removed the object between the
 585                  *         former oi_lookup and the iget. It is normal.
 586                  *      4. Other failure cases.
 587                  *
 588                  *      Generally, when the device is mounted, it will
 589                  *      auto check whether the system is restored from
 590                  *      file-level backup or not. We trust such detect
 591                  *      to distinguish the 1st case from the 2nd case:
 592                  *      if the OI files are consistent but may contain
 593                  *      stale OI mappings because of case 2, if iget()
 594                  *      returns -ENOENT or -ESTALE, then it should be
 595                  *      the case 2. */
 596                 if (rc != 0)
 597                         /* If the OI mapping was in OI file before the
 598                          * osd_iget_check(), but now, it is disappear,
 599                          * then it must be removed by race. That is a
 600                          * normal race case. */
 601                         GOTO(put, rc);
 602
 603                 if ((!IS_ERR(inode) && inode->i_generation != 0 &&
 604                      inode->i_generation == id->oii_gen) ||
 605                     (IS_ERR(inode) && !(dev->od_scrub.os_file.sf_flags &
 606                                         SF_INCONSISTENT)))
 607                         rc = -ENOENT;
 608                 else
 609                         rc = -EREMCHG;
 610         } else {
 611                 if (id->oii_gen == OSD_OII_NOGEN)
 612                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 613
 614                 /* Do not update file c/mtime in ldiskfs.
 615                  * NB: we don't have any lock to protect this because we don't
 616                  * have reference on osd_object now, but contention with
 617                  * another lookup + attr_set can't happen in the tiny window
 618                  * between if (...) and set S_NOCMTIME. */
 619                 if (!(inode->i_flags & S_NOCMTIME))
 620                         inode->i_flags |= S_NOCMTIME;
 621         }
 622
 623         GOTO(put, rc);
 624
 625 put:
 626         if (rc != 0) {
 627                 if (!IS_ERR(inode))
 628                         iput(inode);
 629
 630                 inode = ERR_PTR(rc);
 631         }
 632
 633         return inode;
 634 }
 635
 636 /**
 637  * \retval +v: new filter_fid, does not contain self-fid
 638  * \retval 0:  filter_fid_old, contains self-fid
 639  * \retval -v: other failure cases
 640  */
 641 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 642                  struct dentry *dentry, struct lu_fid *fid)
 643 {
 644         struct filter_fid_old   *ff     = &info->oti_ff;
 645         struct ost_id           *ostid  = &info->oti_ostid;
 646         int                      rc;
 647
 648         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 649         if (rc == sizeof(*ff)) {
 650                 rc = 0;
 651                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 652                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 653                 /* XXX: use 0 as the index for compatibility, the caller will
 654                  *      handle index related issues when necessarry. */
 655                 ostid_to_fid(fid, ostid, 0);
 656         } else if (rc == sizeof(struct filter_fid)) {
 657                 rc = 1;
 658         } else if (rc >= 0) {
 659                 rc = -EINVAL;
 660         }
 661
 662         return rc;
 663 }
 664
 665 static int osd_lma_self_repair(struct osd_thread_info *info,
 666                                struct osd_device *osd, struct inode *inode,
 667                                const struct lu_fid *fid, __u32 compat)
 668 {
 669         handle_t *jh;
 670         int       rc;
 671
 672         LASSERT(current->journal_info == NULL);
 673
 674         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 675                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 676         if (IS_ERR(jh)) {
 677                 rc = PTR_ERR(jh);
 678                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 679                       osd_name(osd), rc);
 680                 return rc;
 681         }
 682
 683         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 684         if (rc != 0)
 685                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 686                       osd_name(osd), rc);
 687         ldiskfs_journal_stop(jh);
 688         return rc;
 689 }
 690
 691 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 692 {
 693         struct osd_thread_info  *info   = osd_oti_get(env);
 694         struct osd_device       *osd    = osd_obj2dev(obj);
 695         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 696         struct inode            *inode  = obj->oo_inode;
 697         struct dentry           *dentry = &info->oti_obj_dentry;
 698         struct lu_fid           *fid    = NULL;
 699         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 700         int                      rc;
 701         ENTRY;
 702
 703         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 704         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 705                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 706         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 707                 fid = &lma->lma_self_fid;
 708                 rc = osd_get_idif(info, inode, dentry, fid);
 709                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 710                         /* For the given OST-object, if it has neither LMA nor
 711                          * FID in XATTR_NAME_FID, then the given FID (which is
 712                          * contained in the @obj, from client RPC for locating
 713                          * the OST-object) is trusted. We use it to generate
 714                          * the LMA. */
 715                         osd_lma_self_repair(info, osd, inode, rfid,
 716                                             LMAC_FID_ON_OST);
 717                         RETURN(0);
 718                 }
 719         }
 720
 721         if (rc < 0)
 722                 RETURN(rc);
 723
 724         if (rc > 0) {
 725                 rc = 0;
 726                 lustre_lma_swab(lma);
 727                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 728                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 729                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 730                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 731                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 732                               PFID(rfid), inode->i_ino);
 733                         rc = -EOPNOTSUPP;
 734                 } else {
 735                         fid = &lma->lma_self_fid;
 736                 }
 737         }
 738
 739         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 740                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 741                         struct ost_id   *oi   = &info->oti_ostid;
 742                         struct lu_fid   *fid1 = &info->oti_fid3;
 743                         __u32            idx  = fid_idif_ost_idx(rfid);
 744
 745                         /* For old IDIF, the OST index is not part of the IDIF,
 746                          * Means that different OSTs may have the same IDIFs.
 747                          * Under such case, we need to make some compatible
 748                          * check to make sure to trigger OI scrub properly. */
 749                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 750                                 /* Given @rfid is new, LMA is old. */
 751                                 fid_to_ostid(fid, oi);
 752                                 ostid_to_fid(fid1, oi, idx);
 753                                 if (lu_fid_eq(fid1, rfid)) {
 754                                         if (osd->od_index_in_idif)
 755                                                 osd_lma_self_repair(info, osd,
 756                                                         inode, rfid,
 757                                                         LMAC_FID_ON_OST);
 758                                         RETURN(0);
 759                                 }
 760                         }
 761                 }
 762
 763                 rc = -EREMCHG;
 764         }
 765
 766         RETURN(rc);
 767 }
 768
 769 struct osd_check_lmv_buf {
 770 #ifdef HAVE_DIR_CONTEXT
 771         /* please keep it as first member */
 772         struct dir_context       ctx;
 773 #endif
 774         struct osd_thread_info  *oclb_info;
 775         struct osd_device       *oclb_dev;
 776         struct osd_idmap_cache  *oclb_oic;
 777 };
 778
 779 /**
 780  * It is called internally by ->readdir() to filter out the
 781  * local slave object's FID of the striped directory.
 782  *
 783  * \retval      1 found the local slave's FID
 784  * \retval      0 continue to check next item
 785  * \retval      -ve for failure
 786  */
 787 #ifdef HAVE_FILLDIR_USE_CTX
 788 static int osd_stripe_dir_filldir(struct dir_context *buf,
 789 #else
 790 static int osd_stripe_dir_filldir(void *buf,
 791 #endif
 792                                   const char *name, int namelen,
 793                                   loff_t offset, __u64 ino, unsigned d_type)
 794 {
 795         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 796         struct osd_thread_info *oti = oclb->oclb_info;
 797         struct lu_fid *fid = &oti->oti_fid3;
 798         struct osd_inode_id *id = &oti->oti_id3;
 799         struct osd_device *dev = oclb->oclb_dev;
 800         struct osd_idmap_cache *oic = oclb->oclb_oic;
 801         struct inode *inode;
 802         int rc;
 803
 804         if (name[0] == '.')
 805                 return 0;
 806
 807         fid_zero(fid);
 808         sscanf(name + 1, SFID, RFID(fid));
 809         if (!fid_is_sane(fid))
 810                 return 0;
 811
 812         if (osd_remote_fid(oti->oti_env, dev, fid))
 813                 return 0;
 814
 815         osd_id_gen(id, ino, OSD_OII_NOGEN);
 816         inode = osd_iget(oti, dev, id);
 817         if (IS_ERR(inode))
 818                 return PTR_ERR(inode);
 819
 820         iput(inode);
 821         osd_add_oi_cache(oti, dev, id, fid);
 822         oic->oic_fid = *fid;
 823         oic->oic_lid = *id;
 824         oic->oic_dev = dev;
 825         rc = osd_oii_insert(dev, oic, true);
 826
 827         return rc == 0 ? 1 : rc;
 828 }
 829
 830 /* When lookup item under striped directory, we need to locate the master
 831  * MDT-object of the striped directory firstly, then the client will send
 832  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 833  * and the item's name. If the system is restored from MDT file level backup,
 834  * then before the OI scrub completely built the OI files, the OI mappings of
 835  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 836  * not a problem for the master MDT-object. Because when locate the master
 837  * MDT-object, we will do name based lookup (for the striped directory itself)
 838  * firstly, during such process we can setup the correct OI mapping for the
 839  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 840  * the client will not trigger name based lookup on the MDT to locate the slave
 841  * MDT-object before locating item under the striped directory, then when
 842  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 843  * is invalid and does not know what the right OI mapping is, then the MDT has
 844  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 845  * the OI file, related OI mapping is unknown yet, please try again later. And
 846  * then client will re-try the RPC again and again until related OI mapping has
 847  * been updated. That is quite inefficient.
 848  *
 849  * To resolve above trouble, we will handle it as the following two cases:
 850  *
 851  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 852  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 853  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 854  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 855  *    directly. Please check osd_fid_lookup().
 856  *
 857  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 858  *    Under such case, during lookup the master MDT-object, we will lookup the
 859  *    slave MDT-object via readdir against the master MDT-object, because the
 860  *    slave MDT-objects information are stored as sub-directories with the name
 861  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 862  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 863  *    the correct OI mapping for the slave MDT-object. */
 864 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 865                          struct inode *inode, struct osd_idmap_cache *oic)
 866 {
 867         struct lu_buf *buf = &oti->oti_big_buf;
 868         struct dentry *dentry = &oti->oti_obj_dentry;
 869         struct file *filp = &oti->oti_file;
 870         const struct file_operations *fops;
 871         struct lmv_mds_md_v1 *lmv1;
 872         struct osd_check_lmv_buf oclb = {
 873 #ifdef HAVE_DIR_CONTEXT
 874                 .ctx.actor = osd_stripe_dir_filldir,
 875 #endif
 876                 .oclb_info = oti,
 877                 .oclb_dev = dev,
 878                 .oclb_oic = oic
 879         };
 880         int rc = 0;
 881         ENTRY;
 882
 883 again:
 884         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 885                              buf->lb_len);
 886         if (rc == -ERANGE) {
 887                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 888                 if (rc > 0) {
 889                         lu_buf_realloc(buf, rc);
 890                         if (buf->lb_buf == NULL)
 891                                 GOTO(out, rc = -ENOMEM);
 892
 893                         goto again;
 894                 }
 895         }
 896
 897         if (unlikely(rc == 0 || rc == -ENODATA))
 898                 GOTO(out, rc = 0);
 899
 900         if (rc < 0)
 901                 GOTO(out, rc);
 902
 903         if (unlikely(buf->lb_buf == NULL)) {
 904                 lu_buf_realloc(buf, rc);
 905                 if (buf->lb_buf == NULL)
 906                         GOTO(out, rc = -ENOMEM);
 907
 908                 goto again;
 909         }
 910
 911         lmv1 = buf->lb_buf;
 912         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 913                 GOTO(out, rc = 0);
 914
 915         fops = inode->i_fop;
 916         dentry->d_inode = inode;
 917         dentry->d_sb = inode->i_sb;
 918         filp->f_pos = 0;
 919         filp->f_path.dentry = dentry;
 920         filp->f_mode = FMODE_64BITHASH;
 921         filp->f_mapping = inode->i_mapping;
 922         filp->f_op = fops;
 923         filp->private_data = NULL;
 924         set_file_inode(filp, inode);
 925
 926 #ifdef HAVE_DIR_CONTEXT
 927         oclb.ctx.pos = filp->f_pos;
 928         rc = fops->iterate(filp, &oclb.ctx);
 929         filp->f_pos = oclb.ctx.pos;
 930 #else
 931         rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
 932 #endif
 933         fops->release(inode, filp);
 934
 935 out:
 936         if (rc < 0)
 937                 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
 938                        DFID": rc = %d\n",
 939                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 940                        inode->i_ino, inode->i_generation,
 941                        PFID(&oic->oic_fid), rc);
 942         else
 943                 rc = 0;
 944
 945         RETURN(rc);
 946 }
 947
 948 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 949                           const struct lu_fid *fid,
 950                           const struct lu_object_conf *conf)
 951 {
 952         struct osd_thread_info *info;
 953         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 954         struct osd_device      *dev;
 955         struct osd_idmap_cache *oic;
 956         struct osd_inode_id    *id;
 957         struct osd_inode_id    *tid;
 958         struct inode           *inode = NULL;
 959         struct osd_scrub       *scrub;
 960         struct scrub_file      *sf;
 961         __u32                   flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT |
 962                                         SS_AUTO_FULL;
 963         __u32                   saved_ino;
 964         __u32                   saved_gen;
 965         int                     result  = 0;
 966         int                     rc1     = 0;
 967         bool                    cached  = true;
 968         bool                    remote  = false;
 969         ENTRY;
 970
 971         LINVRNT(osd_invariant(obj));
 972         LASSERT(obj->oo_inode == NULL);
 973         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 974
 975         dev = osd_dev(ldev);
 976         scrub = &dev->od_scrub;
 977         sf = &scrub->os_file;
 978         info = osd_oti_get(env);
 979         LASSERT(info);
 980         oic = &info->oti_cache;
 981
 982         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
 983                 RETURN(-ENOENT);
 984
 985         /* For the object is created as locking anchor, or for the object to
 986          * be created on disk. No need to osd_oi_lookup() at here because FID
 987          * shouldn't never be re-used, if it's really a duplicate FID from
 988          * unexpected reason, we should be able to detect it later by calling
 989          * do_create->osd_oi_insert(). */
 990         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 991                 GOTO(out, result = 0);
 992
 993         /* Search order: 1. per-thread cache. */
 994         if (lu_fid_eq(fid, &oic->oic_fid) &&
 995             likely(oic->oic_dev == dev)) {
 996                 id = &oic->oic_lid;
 997                 goto iget;
 998         }
 999
1000         id = &info->oti_id;
1001         if (!list_empty(&scrub->os_inconsistent_items)) {
1002                 /* Search order: 2. OI scrub pending list. */
1003                 result = osd_oii_lookup(dev, fid, id);
1004                 if (result == 0)
1005                         goto iget;
1006         }
1007
1008         cached = false;
1009         /* Search order: 3. OI files. */
1010         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1011         if (result == -ENOENT) {
1012                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1013                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1014                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1015                                       sf->sf_oi_bitmap))
1016                         GOTO(out, result = 0);
1017
1018                 goto trigger;
1019         }
1020
1021         if (result != 0)
1022                 GOTO(out, result);
1023
1024 iget:
1025         inode = osd_iget_check(info, dev, fid, id, cached);
1026         if (IS_ERR(inode)) {
1027                 result = PTR_ERR(inode);
1028                 if (result == -ENOENT || result == -ESTALE)
1029                         GOTO(out, result = 0);
1030
1031                 if (result == -EREMCHG) {
1032
1033 trigger:
1034                         /* We still have chance to get the valid inode: for the
1035                          * object which is referenced by remote name entry, the
1036                          * object on the local MDT will be linked under the dir
1037                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1038                          *
1039                          * We do not know whether the object for the given FID
1040                          * is referenced by some remote name entry or not, and
1041                          * especially for DNE II, a multiple-linked object may
1042                          * have many name entries reside on many MDTs.
1043                          *
1044                          * To simplify the operation, OSD will not distinguish
1045                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1046                          * only happened for the RPC from other MDT during the
1047                          * OI scrub, or for the client side RPC with FID only,
1048                          * such as FID to path, or from old connected client. */
1049                         if (!remote &&
1050                             !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1051                                 rc1 = osd_lookup_in_remote_parent(info, dev,
1052                                                                   fid, id);
1053                                 if (rc1 == 0) {
1054                                         remote = true;
1055                                         cached = true;
1056                                         flags |= SS_AUTO_PARTIAL;
1057                                         flags &= ~SS_AUTO_FULL;
1058                                         goto iget;
1059                                 }
1060                         }
1061
1062                         if (thread_is_running(&scrub->os_thread)) {
1063                                 if (scrub->os_partial_scan &&
1064                                     !scrub->os_in_join) {
1065                                         goto join;
1066                                 } else {
1067                                         if (inode != NULL && !IS_ERR(inode)) {
1068                                                 LASSERT(remote);
1069
1070                                                 osd_add_oi_cache(info, dev, id,
1071                                                                  fid);
1072                                                 osd_oii_insert(dev, oic, true);
1073                                         } else {
1074                                                 result = -EINPROGRESS;
1075                                         }
1076                                 }
1077                         } else if (!dev->od_noscrub) {
1078
1079 join:
1080                                 rc1 = osd_scrub_start(dev, flags);
1081                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1082                                               "for the "DFID" with flags 0x%x,"
1083                                               " rc = %d\n", osd_name(dev),
1084                                               PFID(fid), flags, rc1);
1085                                 if (rc1 == 0 || rc1 == -EALREADY) {
1086                                         if (inode != NULL && !IS_ERR(inode)) {
1087                                                 LASSERT(remote);
1088
1089                                                 osd_add_oi_cache(info, dev, id,
1090                                                                  fid);
1091                                                 osd_oii_insert(dev, oic, true);
1092                                         } else {
1093                                                 result = -EINPROGRESS;
1094                                         }
1095                                 } else {
1096                                         result = -EREMCHG;
1097                                 }
1098                         } else {
1099                                 result = -EREMCHG;
1100                         }
1101                 }
1102
1103                 if (inode == NULL || IS_ERR(inode))
1104                         GOTO(out, result);
1105         } else if (remote) {
1106                 goto trigger;
1107         }
1108
1109         obj->oo_inode = inode;
1110         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1111
1112         result = osd_check_lma(env, obj);
1113         if (result == 0)
1114                 goto found;
1115
1116         tid = &info->oti_id3;
1117         LASSERT(tid != id);
1118
1119         if (result == -ENODATA) {
1120                 if (!cached)
1121                         /* The current OI mapping is from the OI file,
1122                          * since the inode has been found via
1123                          * osd_iget_check(), no need recheck OI. */
1124                         goto found;
1125
1126                 result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1127                 if (result == 0) {
1128                         LASSERTF(tid->oii_ino == id->oii_ino &&
1129                                  tid->oii_gen == id->oii_gen,
1130                                  "OI mapping changed(1): %u/%u => %u/%u",
1131                                  tid->oii_ino, tid->oii_gen,
1132                                  id->oii_ino, id->oii_gen);
1133
1134                         LASSERTF(tid->oii_ino == inode->i_ino &&
1135                                  tid->oii_gen == inode->i_generation,
1136                                  "locate wrong inode(1): %u/%u => %ld/%u",
1137                                  tid->oii_ino, tid->oii_gen,
1138                                  inode->i_ino, inode->i_generation);
1139
1140                         /* "result == 0" means the cached OI mapping is still in
1141                          * the OI file, so the target the inode is valid. */
1142                         goto found;
1143                 }
1144
1145                 /* "result == -ENOENT" means that the OI mappinghas been removed
1146                  * by race, the target inode belongs to other object.
1147                  *
1148                  * Others error can be returned  directly. */
1149                 if (result == -ENOENT)
1150                         result = 0;
1151         }
1152
1153         saved_ino = inode->i_ino;
1154         saved_gen = inode->i_generation;
1155         iput(inode);
1156         inode = NULL;
1157         obj->oo_inode = NULL;
1158
1159         if (result != -EREMCHG)
1160                 GOTO(out, result);
1161
1162         if (!cached)
1163                 /* The current OI mapping is from the OI file,
1164                  * since the inode has been found via
1165                  * osd_iget_check(), no need recheck OI. */
1166                 goto trigger;
1167
1168         result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1169         /* "result == -ENOENT" means the cached OI mapping has been removed from
1170          * the OI file by race, above target inode belongs to other object.
1171          *
1172          * Others error can be returned directly. */
1173         if (result != 0)
1174                 GOTO(out, result = (result == -ENOENT ? 0 : result));
1175
1176         LASSERTF(tid->oii_ino == id->oii_ino && tid->oii_gen == id->oii_gen,
1177                  "OI mapping changed(2): %u/%u => %u/%u",
1178                  tid->oii_ino, tid->oii_gen, id->oii_ino, id->oii_gen);
1179
1180         LASSERTF(tid->oii_ino == saved_ino && tid->oii_gen == saved_gen,
1181                  "locate wrong inode(2): %u/%u => %u/%u",
1182                  tid->oii_ino, tid->oii_gen, saved_ino, saved_gen);
1183
1184         goto trigger;
1185
1186 found:
1187         obj->oo_compat_dot_created = 1;
1188         obj->oo_compat_dotdot_created = 1;
1189
1190         if (S_ISDIR(inode->i_mode) &&
1191             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1192                 osd_check_lmv(info, dev, inode, oic);
1193
1194         if (!ldiskfs_pdo)
1195                 GOTO(out, result = 0);
1196
1197         LASSERT(obj->oo_hl_head == NULL);
1198         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1199         if (obj->oo_hl_head == NULL) {
1200                 obj->oo_inode = NULL;
1201                 iput(inode);
1202                 GOTO(out, result = -ENOMEM);
1203         }
1204         GOTO(out, result = 0);
1205
1206 out:
1207         if (result != 0 && cached)
1208                 fid_zero(&oic->oic_fid);
1209
1210         LINVRNT(osd_invariant(obj));
1211         return result;
1212 }
1213
1214 /*
1215  * Concurrency: shouldn't matter.
1216  */
1217 static void osd_object_init0(struct osd_object *obj)
1218 {
1219         LASSERT(obj->oo_inode != NULL);
1220         obj->oo_dt.do_body_ops = &osd_body_ops;
1221         obj->oo_dt.do_lu.lo_header->loh_attr |=
1222                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1223 }
1224
1225 /*
1226  * Concurrency: no concurrent access is possible that early in object
1227  * life-cycle.
1228  */
1229 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1230                            const struct lu_object_conf *conf)
1231 {
1232         struct osd_object *obj = osd_obj(l);
1233         int result;
1234
1235         LINVRNT(osd_invariant(obj));
1236
1237         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1238                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1239                 l->lo_header->loh_attr |= LOHA_EXISTS;
1240                 return 0;
1241         }
1242
1243         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1244         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1245         if (result == 0 && obj->oo_inode != NULL) {
1246                 struct osd_thread_info *oti = osd_oti_get(env);
1247                 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
1248
1249                 osd_object_init0(obj);
1250                 result = osd_get_lma(oti, obj->oo_inode,
1251                                      &oti->oti_obj_dentry, lma);
1252                 if (result == 0) {
1253                         /* Convert LMAI flags to lustre LMA flags
1254                          * and cache it to oo_lma_flags */
1255                         obj->oo_lma_flags =
1256                                 lma_to_lustre_flags(lma->lma_incompat);
1257                 } else if (result == -ENODATA) {
1258                         result = 0;
1259                 }
1260         }
1261
1262         LINVRNT(osd_invariant(obj));
1263         return result;
1264 }
1265
1266 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1267  * The left part is for value, binary mode. */
1268 struct osd_xattr_entry {
1269         struct list_head        oxe_list;
1270         size_t                  oxe_len;
1271         size_t                  oxe_namelen;
1272         bool                    oxe_exist;
1273         struct rcu_head         oxe_rcu;
1274         char                    oxe_buf[0];
1275 };
1276
1277 static int osd_oxc_get(struct osd_object *obj, const char *name,
1278                        struct lu_buf *buf)
1279 {
1280         struct osd_xattr_entry *tmp;
1281         struct osd_xattr_entry *oxe = NULL;
1282         size_t namelen = strlen(name);
1283         int rc;
1284         ENTRY;
1285
1286         rcu_read_lock();
1287         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1288                 if (namelen == tmp->oxe_namelen &&
1289                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1290                         oxe = tmp;
1291                         break;
1292                 }
1293         }
1294
1295         if (oxe == NULL)
1296                 GOTO(out, rc = -ENOENT);
1297
1298         if (!oxe->oxe_exist)
1299                 GOTO(out, rc = -ENODATA);
1300
1301         /* vallen */
1302         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1303         LASSERT(rc > 0);
1304
1305         if (buf->lb_buf == NULL)
1306                 GOTO(out, rc);
1307
1308         if (buf->lb_len < rc)
1309                 GOTO(out, rc = -ERANGE);
1310
1311         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1312         EXIT;
1313 out:
1314         rcu_read_unlock();
1315
1316         return rc;
1317 }
1318
1319 static void osd_oxc_free(struct rcu_head *head)
1320 {
1321         struct osd_xattr_entry *oxe;
1322
1323         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1324         OBD_FREE(oxe, oxe->oxe_len);
1325 }
1326
1327 static void osd_oxc_add(struct osd_object *obj, const char *name,
1328                         const char *buf, int buflen)
1329 {
1330         struct osd_xattr_entry *oxe;
1331         struct osd_xattr_entry *old = NULL;
1332         struct osd_xattr_entry *tmp;
1333         size_t namelen = strlen(name);
1334         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1335
1336         OBD_ALLOC(oxe, len);
1337         if (oxe == NULL)
1338                 return;
1339
1340         INIT_LIST_HEAD(&oxe->oxe_list);
1341         oxe->oxe_len = len;
1342         oxe->oxe_namelen = namelen;
1343         memcpy(oxe->oxe_buf, name, namelen);
1344         if (buflen > 0) {
1345                 LASSERT(buf != NULL);
1346                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1347                 oxe->oxe_exist = true;
1348         } else {
1349                 oxe->oxe_exist = false;
1350         }
1351
1352         /* this should be rarely called, just remove old and add new */
1353         spin_lock(&obj->oo_guard);
1354         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1355                 if (namelen == tmp->oxe_namelen &&
1356                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1357                         old = tmp;
1358                         break;
1359                 }
1360         }
1361         if (old != NULL) {
1362                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1363                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1364         } else {
1365                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1366         }
1367         spin_unlock(&obj->oo_guard);
1368 }
1369
1370 static void osd_oxc_del(struct osd_object *obj, const char *name)
1371 {
1372         struct osd_xattr_entry *oxe;
1373         size_t namelen = strlen(name);
1374
1375         spin_lock(&obj->oo_guard);
1376         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1377                 if (namelen == oxe->oxe_namelen &&
1378                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1379                         list_del_rcu(&oxe->oxe_list);
1380                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1381                         break;
1382                 }
1383         }
1384         spin_unlock(&obj->oo_guard);
1385 }
1386
1387 static void osd_oxc_fini(struct osd_object *obj)
1388 {
1389         struct osd_xattr_entry *oxe, *next;
1390
1391         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1392                 list_del(&oxe->oxe_list);
1393                 OBD_FREE(oxe, oxe->oxe_len);
1394         }
1395 }
1396
1397 /*
1398  * Concurrency: no concurrent access is possible that late in object
1399  * life-cycle.
1400  */
1401 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1402 {
1403         struct osd_object *obj = osd_obj(l);
1404
1405         LINVRNT(osd_invariant(obj));
1406
1407         osd_oxc_fini(obj);
1408         dt_object_fini(&obj->oo_dt);
1409         if (obj->oo_hl_head != NULL)
1410                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1411         OBD_FREE_PTR(obj);
1412 }
1413
1414 /*
1415  * Concurrency: no concurrent access is possible that late in object
1416  * life-cycle.
1417  */
1418 static void osd_index_fini(struct osd_object *o)
1419 {
1420         struct iam_container *bag;
1421
1422         if (o->oo_dir != NULL) {
1423                 bag = &o->oo_dir->od_container;
1424                 if (o->oo_inode != NULL) {
1425                         if (bag->ic_object == o->oo_inode)
1426                                 iam_container_fini(bag);
1427                 }
1428                 OBD_FREE_PTR(o->oo_dir);
1429                 o->oo_dir = NULL;
1430         }
1431 }
1432
1433 /*
1434  * Concurrency: no concurrent access is possible that late in object
1435  * life-cycle (for all existing callers, that is. New callers have to provide
1436  * their own locking.)
1437  */
1438 static int osd_inode_unlinked(const struct inode *inode)
1439 {
1440         return inode->i_nlink == 0;
1441 }
1442
1443 enum {
1444         OSD_TXN_OI_DELETE_CREDITS    = 20,
1445         OSD_TXN_INODE_DELETE_CREDITS = 20
1446 };
1447
1448 /*
1449  * Journal
1450  */
1451
1452 #if OSD_THANDLE_STATS
1453 /**
1454  * Set time when the handle is allocated
1455  */
1456 static void osd_th_alloced(struct osd_thandle *oth)
1457 {
1458         oth->oth_alloced = cfs_time_current();
1459 }
1460
1461 /**
1462  * Set time when the handle started
1463  */
1464 static void osd_th_started(struct osd_thandle *oth)
1465 {
1466         oth->oth_started = cfs_time_current();
1467 }
1468
1469 /**
1470  * Helper function to convert time interval to microseconds packed in
1471  * long int.
1472  */
1473 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
1474 {
1475         struct timeval val;
1476
1477         cfs_duration_usec(cfs_time_sub(end, start), &val);
1478         return val.tv_sec * 1000000 + val.tv_usec;
1479 }
1480
1481 /**
1482  * Check whether the we deal with this handle for too long.
1483  */
1484 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1485                                 cfs_time_t alloced, cfs_time_t started,
1486                                 cfs_time_t closed)
1487 {
1488         cfs_time_t now = cfs_time_current();
1489
1490         LASSERT(dev != NULL);
1491
1492         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1493                             interval_to_usec(alloced, started));
1494         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1495                             interval_to_usec(started, closed));
1496         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1497                             interval_to_usec(closed, now));
1498
1499         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
1500                 CWARN("transaction handle %p was open for too long: "
1501                       "now "CFS_TIME_T" ,"
1502                       "alloced "CFS_TIME_T" ,"
1503                       "started "CFS_TIME_T" ,"
1504                       "closed "CFS_TIME_T"\n",
1505                       oth, now, alloced, started, closed);
1506                 libcfs_debug_dumpstack(NULL);
1507         }
1508 }
1509
1510 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1511 {                                                                       \
1512         cfs_time_t __closed = cfs_time_current();                       \
1513         cfs_time_t __alloced = oth->oth_alloced;                        \
1514         cfs_time_t __started = oth->oth_started;                        \
1515                                                                         \
1516         expr;                                                           \
1517         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1518 }
1519
1520 #else /* OSD_THANDLE_STATS */
1521
1522 #define osd_th_alloced(h)                  do {} while(0)
1523 #define osd_th_started(h)                  do {} while(0)
1524 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1525
1526 #endif /* OSD_THANDLE_STATS */
1527
1528 /*
1529  * Concurrency: doesn't access mutable data.
1530  */
1531 static int osd_param_is_not_sane(const struct osd_device *dev,
1532                                  const struct thandle *th)
1533 {
1534         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1535
1536         return oh->ot_credits > osd_transaction_size(dev);
1537 }
1538
1539 /*
1540  * Concurrency: shouldn't matter.
1541  */
1542 static void osd_trans_commit_cb(struct super_block *sb,
1543                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1544 {
1545         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1546         struct thandle     *th  = &oh->ot_super;
1547         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1548         struct dt_txn_commit_cb *dcb, *tmp;
1549
1550         LASSERT(oh->ot_handle == NULL);
1551
1552         if (error)
1553                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1554
1555         dt_txn_hook_commit(th);
1556
1557         /* call per-transaction callbacks if any */
1558         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1559                                  dcb_linkage) {
1560                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1561                          "commit callback entry: magic=%x name='%s'\n",
1562                          dcb->dcb_magic, dcb->dcb_name);
1563                 list_del_init(&dcb->dcb_linkage);
1564                 dcb->dcb_func(NULL, th, dcb, error);
1565         }
1566
1567         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1568         lu_device_put(lud);
1569         th->th_dev = NULL;
1570
1571         lu_context_exit(&th->th_ctx);
1572         lu_context_fini(&th->th_ctx);
1573         OBD_FREE_PTR(oh);
1574 }
1575
1576 #ifndef HAVE_SB_START_WRITE
1577 # define sb_start_write(sb) do {} while (0)
1578 # define sb_end_write(sb) do {} while (0)
1579 #endif
1580
1581 static struct thandle *osd_trans_create(const struct lu_env *env,
1582                                         struct dt_device *d)
1583 {
1584         struct osd_thread_info  *oti = osd_oti_get(env);
1585         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1586         struct osd_thandle      *oh;
1587         struct thandle          *th;
1588         ENTRY;
1589
1590         /* on pending IO in this thread should left from prev. request */
1591         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1592
1593         sb_start_write(osd_sb(osd_dt_dev(d)));
1594
1595         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1596         if (oh != NULL) {
1597                 oh->ot_quota_trans = &oti->oti_quota_trans;
1598                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1599                 th = &oh->ot_super;
1600                 th->th_dev = d;
1601                 th->th_result = 0;
1602                 th->th_tags = LCT_TX_HANDLE;
1603                 oh->ot_credits = 0;
1604                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1605                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1606                 osd_th_alloced(oh);
1607
1608                 memset(oti->oti_declare_ops, 0,
1609                        sizeof(oti->oti_declare_ops));
1610                 memset(oti->oti_declare_ops_cred, 0,
1611                        sizeof(oti->oti_declare_ops_cred));
1612                 memset(oti->oti_declare_ops_used, 0,
1613                        sizeof(oti->oti_declare_ops_used));
1614         } else {
1615                 sb_end_write(osd_sb(osd_dt_dev(d)));
1616                 th = ERR_PTR(-ENOMEM);
1617         }
1618         RETURN(th);
1619 }
1620
1621 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1622 {
1623         struct osd_thread_info  *oti = osd_oti_get(env);
1624         struct osd_thandle      *oh;
1625
1626         oh = container_of0(th, struct osd_thandle, ot_super);
1627         LASSERT(oh != NULL);
1628
1629         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1630               oti->oti_declare_ops[OSD_OT_CREATE],
1631               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1632               oti->oti_declare_ops_used[OSD_OT_CREATE],
1633               oti->oti_declare_ops[OSD_OT_DESTROY],
1634               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1635               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1636         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1637               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1638               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1639               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1640               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1641               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1642               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1643         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1644               oti->oti_declare_ops[OSD_OT_WRITE],
1645               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1646               oti->oti_declare_ops_used[OSD_OT_WRITE],
1647               oti->oti_declare_ops[OSD_OT_PUNCH],
1648               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1649               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1650               oti->oti_declare_ops[OSD_OT_QUOTA],
1651               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1652               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1653         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1654               oti->oti_declare_ops[OSD_OT_INSERT],
1655               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1656               oti->oti_declare_ops_used[OSD_OT_INSERT],
1657               oti->oti_declare_ops[OSD_OT_DELETE],
1658               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1659               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1660         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1661               oti->oti_declare_ops[OSD_OT_REF_ADD],
1662               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1663               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1664               oti->oti_declare_ops[OSD_OT_REF_DEL],
1665               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1666               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1667 }
1668
1669 /*
1670  * Concurrency: shouldn't matter.
1671  */
1672 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1673                            struct thandle *th)
1674 {
1675         struct osd_thread_info *oti = osd_oti_get(env);
1676         struct osd_device  *dev = osd_dt_dev(d);
1677         handle_t           *jh;
1678         struct osd_thandle *oh;
1679         int rc;
1680
1681         ENTRY;
1682
1683         LASSERT(current->journal_info == NULL);
1684
1685         oh = container_of0(th, struct osd_thandle, ot_super);
1686         LASSERT(oh != NULL);
1687         LASSERT(oh->ot_handle == NULL);
1688
1689         rc = dt_txn_hook_start(env, d, th);
1690         if (rc != 0)
1691                 GOTO(out, rc);
1692
1693         if (unlikely(osd_param_is_not_sane(dev, th))) {
1694                 static unsigned long last_printed;
1695                 static int last_credits;
1696
1697                 /* don't make noise on a tiny testing systems
1698                  * actual credits misuse will be caught anyway */
1699                 if (last_credits != oh->ot_credits &&
1700                     time_after(jiffies, last_printed +
1701                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1702                     osd_transaction_size(dev) > 512) {
1703                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1704                               oh->ot_credits, osd_transaction_size(dev));
1705                         osd_trans_dump_creds(env, th);
1706                         libcfs_debug_dumpstack(NULL);
1707                         last_credits = oh->ot_credits;
1708                         last_printed = jiffies;
1709                 }
1710                 /* XXX Limit the credits to 'max_transaction_buffers', and
1711                  *     let the underlying filesystem to catch the error if
1712                  *     we really need so many credits.
1713                  *
1714                  *     This should be removed when we can calculate the
1715                  *     credits precisely. */
1716                 oh->ot_credits = osd_transaction_size(dev);
1717         }
1718
1719         /*
1720          * XXX temporary stuff. Some abstraction layer should
1721          * be used.
1722          */
1723         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1724         osd_th_started(oh);
1725         if (!IS_ERR(jh)) {
1726                 oh->ot_handle = jh;
1727                 LASSERT(oti->oti_txns == 0);
1728                 lu_context_init(&th->th_ctx, th->th_tags);
1729                 lu_context_enter(&th->th_ctx);
1730
1731                 lu_device_get(&d->dd_lu_dev);
1732                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1733                               "osd-tx", th);
1734                 oti->oti_txns++;
1735                 rc = 0;
1736         } else {
1737                 rc = PTR_ERR(jh);
1738         }
1739 out:
1740         RETURN(rc);
1741 }
1742
1743 static int osd_seq_exists(const struct lu_env *env,
1744                           struct osd_device *osd, u64 seq)
1745 {
1746         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1747         struct seq_server_site  *ss = osd_seq_site(osd);
1748         int                     rc;
1749         ENTRY;
1750
1751         LASSERT(ss != NULL);
1752         LASSERT(ss->ss_server_fld != NULL);
1753
1754         rc = osd_fld_lookup(env, osd, seq, range);
1755         if (rc != 0) {
1756                 if (rc != -ENOENT)
1757                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1758                                osd_name(osd), seq, rc);
1759                 RETURN(0);
1760         }
1761
1762         RETURN(ss->ss_node_id == range->lsr_index);
1763 }
1764
1765 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1766 {
1767         struct dt_txn_commit_cb *dcb;
1768         struct dt_txn_commit_cb *tmp;
1769
1770         /* call per-transaction stop callbacks if any */
1771         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1772                                  dcb_linkage) {
1773                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1774                          "commit callback entry: magic=%x name='%s'\n",
1775                          dcb->dcb_magic, dcb->dcb_name);
1776                 list_del_init(&dcb->dcb_linkage);
1777                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1778         }
1779 }
1780
1781 /*
1782  * Concurrency: shouldn't matter.
1783  */
1784 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1785                           struct thandle *th)
1786 {
1787         int                     rc = 0, remove_agents = 0;
1788         struct osd_thandle     *oh;
1789         struct osd_thread_info *oti = osd_oti_get(env);
1790         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1791         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1792         struct qsd_instance    *qsd = osd->od_quota_slave;
1793         struct lquota_trans    *qtrans;
1794         ENTRY;
1795
1796         oh = container_of0(th, struct osd_thandle, ot_super);
1797
1798         /* reset OI cache for safety */
1799         oti->oti_ins_cache_used = 0;
1800
1801         remove_agents = oh->ot_remove_agents;
1802
1803         qtrans = oh->ot_quota_trans;
1804         oh->ot_quota_trans = NULL;
1805
1806         if (oh->ot_handle != NULL) {
1807                 handle_t *hdl = oh->ot_handle;
1808
1809                 /*
1810                  * add commit callback
1811                  * notice we don't do this in osd_trans_start()
1812                  * as underlying transaction can change during truncate
1813                  */
1814                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1815                                          &oh->ot_jcb);
1816
1817                 LASSERT(oti->oti_txns == 1);
1818                 oti->oti_txns--;
1819
1820                 rc = dt_txn_hook_stop(env, th);
1821                 if (rc != 0)
1822                         CERROR("%s: failed in transaction hook: rc = %d\n",
1823                                osd_name(osd), rc);
1824
1825                 osd_trans_stop_cb(oh, rc);
1826                 /* hook functions might modify th_sync */
1827                 hdl->h_sync = th->th_sync;
1828
1829                 oh->ot_handle = NULL;
1830                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1831                 if (rc != 0)
1832                         CERROR("%s: failed to stop transaction: rc = %d\n",
1833                                osd_name(osd), rc);
1834         } else {
1835                 osd_trans_stop_cb(oh, th->th_result);
1836                 OBD_FREE_PTR(oh);
1837         }
1838
1839         /* inform the quota slave device that the transaction is stopping */
1840         qsd_op_end(env, qsd, qtrans);
1841
1842         /* as we want IO to journal and data IO be concurrent, we don't block
1843          * awaiting data IO completion in osd_do_bio(), instead we wait here
1844          * once transaction is submitted to the journal. all reqular requests
1845          * don't do direct IO (except read/write), thus this wait_event becomes
1846          * no-op for them.
1847          *
1848          * IMPORTANT: we have to wait till any IO submited by the thread is
1849          * completed otherwise iobuf may be corrupted by different request
1850          */
1851         wait_event(iobuf->dr_wait,
1852                        atomic_read(&iobuf->dr_numreqs) == 0);
1853         osd_fini_iobuf(osd, iobuf);
1854         if (!rc)
1855                 rc = iobuf->dr_error;
1856
1857         if (unlikely(remove_agents != 0))
1858                 osd_process_scheduled_agent_removals(env, osd);
1859
1860         sb_end_write(osd_sb(osd));
1861
1862         RETURN(rc);
1863 }
1864
1865 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1866 {
1867         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1868                                                ot_super);
1869
1870         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1871         LASSERT(&dcb->dcb_func != NULL);
1872         if (dcb->dcb_flags & DCB_TRANS_STOP)
1873                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1874         else
1875                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1876
1877         return 0;
1878 }
1879
1880 /*
1881  * Called just before object is freed. Releases all resources except for
1882  * object itself (that is released by osd_object_free()).
1883  *
1884  * Concurrency: no concurrent access is possible that late in object
1885  * life-cycle.
1886  */
1887 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1888 {
1889         struct osd_object *obj   = osd_obj(l);
1890         struct inode      *inode = obj->oo_inode;
1891
1892         LINVRNT(osd_invariant(obj));
1893
1894         /*
1895          * If object is unlinked remove fid->ino mapping from object index.
1896          */
1897
1898         osd_index_fini(obj);
1899         if (inode != NULL) {
1900                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1901                 qid_t                    uid = i_uid_read(inode);
1902                 qid_t                    gid = i_gid_read(inode);
1903
1904                 iput(inode);
1905                 obj->oo_inode = NULL;
1906
1907                 if (qsd != NULL) {
1908                         struct osd_thread_info  *info = osd_oti_get(env);
1909                         struct lquota_id_info   *qi = &info->oti_qi;
1910
1911                         /* Release granted quota to master if necessary */
1912                         qi->lqi_id.qid_uid = uid;
1913                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1914
1915                         qi->lqi_id.qid_uid = gid;
1916                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1917                 }
1918         }
1919 }
1920
1921 /*
1922  * Concurrency: ->loo_object_release() is called under site spin-lock.
1923  */
1924 static void osd_object_release(const struct lu_env *env,
1925                                struct lu_object *l)
1926 {
1927         struct osd_object *o = osd_obj(l);
1928         /* nobody should be releasing a non-destroyed object with nlink=0
1929          * the API allows this, but ldiskfs doesn't like and then report
1930          * this inode as deleted */
1931         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1932                 LBUG();
1933 }
1934
1935 /*
1936  * Concurrency: shouldn't matter.
1937  */
1938 static int osd_object_print(const struct lu_env *env, void *cookie,
1939                             lu_printer_t p, const struct lu_object *l)
1940 {
1941         struct osd_object *o = osd_obj(l);
1942         struct iam_descr  *d;
1943
1944         if (o->oo_dir != NULL)
1945                 d = o->oo_dir->od_container.ic_descr;
1946         else
1947                 d = NULL;
1948         return (*p)(env, cookie,
1949                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1950                     o, o->oo_inode,
1951                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1952                     o->oo_inode ? o->oo_inode->i_generation : 0,
1953                     d ? d->id_ops->id_name : "plain");
1954 }
1955
1956 /*
1957  * Concurrency: shouldn't matter.
1958  */
1959 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1960                struct obd_statfs *sfs)
1961 {
1962         struct osd_device       *osd = osd_dt_dev(d);
1963         struct super_block      *sb = osd_sb(osd);
1964         struct kstatfs          *ksfs;
1965         __u64                    reserved;
1966         int                      result = 0;
1967
1968         if (unlikely(osd->od_mnt == NULL))
1969                 return -EINPROGRESS;
1970
1971         /* osd_lproc.c call this without env, allocate ksfs for that case */
1972         if (unlikely(env == NULL)) {
1973                 OBD_ALLOC_PTR(ksfs);
1974                 if (ksfs == NULL)
1975                         return -ENOMEM;
1976         } else {
1977                 ksfs = &osd_oti_get(env)->oti_ksfs;
1978         }
1979
1980         result = sb->s_op->statfs(sb->s_root, ksfs);
1981         if (result)
1982                 goto out;
1983
1984         statfs_pack(sfs, ksfs);
1985         if (unlikely(sb->s_flags & MS_RDONLY))
1986                 sfs->os_state |= OS_STATE_READONLY;
1987         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1988                                          LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1989                 sfs->os_maxbytes = sb->s_maxbytes;
1990         else
1991                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1992
1993         /*
1994          * Reserve some space so to avoid fragmenting the filesystem too much.
1995          * Fragmentation not only impacts performance, but can also increase
1996          * metadata overhead significantly, causing grant calculation to be
1997          * wrong.
1998          *
1999          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2000          */
2001         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2002         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2003         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2004                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2005
2006         sfs->os_blocks -= reserved;
2007         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2008         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2009
2010 out:
2011         if (unlikely(env == NULL))
2012                 OBD_FREE_PTR(ksfs);
2013         return result;
2014 }
2015
2016 /**
2017  * Estimate space needed for file creations. We assume the largest filename
2018  * which is 2^64 - 1, hence a filename of 20 chars.
2019  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2020  */
2021 #ifdef __LDISKFS_DIR_REC_LEN
2022 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2023 #else
2024 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2025 #endif
2026
2027 /*
2028  * Concurrency: doesn't access mutable data.
2029  */
2030 static void osd_conf_get(const struct lu_env *env,
2031                          const struct dt_device *dev,
2032                          struct dt_device_param *param)
2033 {
2034         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2035         int                ea_overhead;
2036
2037         /*
2038          * XXX should be taken from not-yet-existing fs abstraction layer.
2039          */
2040         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2041         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2042         param->ddp_symlink_max  = sb->s_blocksize;
2043         param->ddp_mount_type     = LDD_MT_LDISKFS;
2044         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2045                 param->ddp_maxbytes = sb->s_maxbytes;
2046         else
2047                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2048         /* inode are statically allocated, so per-inode space consumption
2049          * is the space consumed by the directory entry */
2050         param->ddp_inodespace     = PER_OBJ_USAGE;
2051         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2052          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2053          * maximum length to not under count the actual number of extents
2054          * needed for writing a file. */
2055         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2056         /* worst-case extent insertion metadata overhead */
2057         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2058         param->ddp_mntopts      = 0;
2059         if (test_opt(sb, XATTR_USER))
2060                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2061         if (test_opt(sb, POSIX_ACL))
2062                 param->ddp_mntopts |= MNTOPT_ACL;
2063
2064         /* LOD might calculate the max stripe count based on max_ea_size,
2065          * so we need take account in the overhead as well,
2066          * xattr_header + magic + xattr_entry_head */
2067         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2068                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2069
2070 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2071         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2072                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2073                                                                 ea_overhead;
2074         else
2075 #endif
2076                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2077 }
2078
2079 /*
2080  * Concurrency: shouldn't matter.
2081  */
2082 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2083 {
2084         int rc;
2085
2086         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2087
2088         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2089
2090         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2091                LUSTRE_OSD_LDISKFS_NAME, rc);
2092
2093         return rc;
2094 }
2095
2096 /**
2097  * Start commit for OSD device.
2098  *
2099  * An implementation of dt_commit_async method for OSD device.
2100  * Asychronously starts underlayng fs sync and thereby a transaction
2101  * commit.
2102  *
2103  * \param env environment
2104  * \param d dt device
2105  *
2106  * \see dt_device_operations
2107  */
2108 static int osd_commit_async(const struct lu_env *env,
2109                             struct dt_device *d)
2110 {
2111         struct super_block *s = osd_sb(osd_dt_dev(d));
2112         ENTRY;
2113
2114         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2115         RETURN(s->s_op->sync_fs(s, 0));
2116 }
2117
2118 /*
2119  * Concurrency: shouldn't matter.
2120  */
2121
2122 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2123 {
2124         struct super_block *sb = osd_sb(osd_dt_dev(d));
2125         struct block_device *dev = sb->s_bdev;
2126 #ifdef HAVE_DEV_SET_RDONLY
2127         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2128         int rc = 0;
2129 #else
2130         int rc = -EOPNOTSUPP;
2131 #endif
2132         ENTRY;
2133
2134 #ifdef HAVE_DEV_SET_RDONLY
2135         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2136
2137         if (sb->s_op->freeze_fs) {
2138                 rc = sb->s_op->freeze_fs(sb);
2139                 if (rc)
2140                         goto out;
2141         }
2142
2143         if (jdev && (jdev != dev)) {
2144                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2145                        (long)jdev);
2146                 dev_set_rdonly(jdev);
2147         }
2148         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2149         dev_set_rdonly(dev);
2150
2151         if (sb->s_op->unfreeze_fs)
2152                 sb->s_op->unfreeze_fs(sb);
2153
2154 out:
2155 #endif
2156         if (rc)
2157                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2158                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2159
2160         RETURN(rc);
2161 }
2162
2163 /**
2164  * Note: we do not count into QUOTA here.
2165  * If we mount with --data_journal we may need more.
2166  */
2167 const int osd_dto_credits_noquota[DTO_NR] = {
2168         /**
2169          * Insert.
2170          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2171          * SINGLEDATA_TRANS_BLOCKS(8)
2172          * XXX Note: maybe iam need more, since iam have more level than
2173          *           EXT3 htree.
2174          */
2175         [DTO_INDEX_INSERT]  = 16,
2176         /**
2177          * Delete
2178          * just modify a single entry, probably merge few within a block
2179          */
2180         [DTO_INDEX_DELETE]  = 1,
2181         /**
2182          * Used for OI scrub
2183          */
2184         [DTO_INDEX_UPDATE]  = 16,
2185         /**
2186          * 4(inode, inode bits, groups, GDT)
2187          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2188          */
2189         [DTO_OBJECT_CREATE] = 4,
2190         /**
2191          * 4(inode, inode bits, groups, GDT)
2192          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2193          */
2194         [DTO_OBJECT_DELETE] = 4,
2195         /**
2196          * Attr set credits (inode)
2197          */
2198         [DTO_ATTR_SET_BASE] = 1,
2199         /**
2200          * Xattr set. The same as xattr of EXT3.
2201          * DATA_TRANS_BLOCKS(14)
2202          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2203          * are also counted in. Do not know why?
2204          */
2205         [DTO_XATTR_SET]     = 14,
2206         /**
2207          * credits for inode change during write.
2208          */
2209         [DTO_WRITE_BASE]    = 3,
2210         /**
2211          * credits for single block write.
2212          */
2213         [DTO_WRITE_BLOCK]   = 14,
2214         /**
2215          * Attr set credits for chown.
2216          * This is extra credits for setattr, and it is null without quota
2217          */
2218         [DTO_ATTR_SET_CHOWN] = 0
2219 };
2220
2221 static const struct dt_device_operations osd_dt_ops = {
2222         .dt_root_get       = osd_root_get,
2223         .dt_statfs         = osd_statfs,
2224         .dt_trans_create   = osd_trans_create,
2225         .dt_trans_start    = osd_trans_start,
2226         .dt_trans_stop     = osd_trans_stop,
2227         .dt_trans_cb_add   = osd_trans_cb_add,
2228         .dt_conf_get       = osd_conf_get,
2229         .dt_sync           = osd_sync,
2230         .dt_ro             = osd_ro,
2231         .dt_commit_async   = osd_commit_async,
2232 };
2233
2234 static void osd_object_read_lock(const struct lu_env *env,
2235                                  struct dt_object *dt, unsigned role)
2236 {
2237         struct osd_object *obj = osd_dt_obj(dt);
2238         struct osd_thread_info *oti = osd_oti_get(env);
2239
2240         LINVRNT(osd_invariant(obj));
2241
2242         LASSERT(obj->oo_owner != env);
2243         down_read_nested(&obj->oo_sem, role);
2244
2245         LASSERT(obj->oo_owner == NULL);
2246         oti->oti_r_locks++;
2247 }
2248
2249 static void osd_object_write_lock(const struct lu_env *env,
2250                                   struct dt_object *dt, unsigned role)
2251 {
2252         struct osd_object *obj = osd_dt_obj(dt);
2253         struct osd_thread_info *oti = osd_oti_get(env);
2254
2255         LINVRNT(osd_invariant(obj));
2256
2257         LASSERT(obj->oo_owner != env);
2258         down_write_nested(&obj->oo_sem, role);
2259
2260         LASSERT(obj->oo_owner == NULL);
2261         obj->oo_owner = env;
2262         oti->oti_w_locks++;
2263 }
2264
2265 static void osd_object_read_unlock(const struct lu_env *env,
2266                                    struct dt_object *dt)
2267 {
2268         struct osd_object *obj = osd_dt_obj(dt);
2269         struct osd_thread_info *oti = osd_oti_get(env);
2270
2271         LINVRNT(osd_invariant(obj));
2272
2273         LASSERT(oti->oti_r_locks > 0);
2274         oti->oti_r_locks--;
2275         up_read(&obj->oo_sem);
2276 }
2277
2278 static void osd_object_write_unlock(const struct lu_env *env,
2279                                     struct dt_object *dt)
2280 {
2281         struct osd_object *obj = osd_dt_obj(dt);
2282         struct osd_thread_info *oti = osd_oti_get(env);
2283
2284         LINVRNT(osd_invariant(obj));
2285
2286         LASSERT(obj->oo_owner == env);
2287         LASSERT(oti->oti_w_locks > 0);
2288         oti->oti_w_locks--;
2289         obj->oo_owner = NULL;
2290         up_write(&obj->oo_sem);
2291 }
2292
2293 static int osd_object_write_locked(const struct lu_env *env,
2294                                    struct dt_object *dt)
2295 {
2296         struct osd_object *obj = osd_dt_obj(dt);
2297
2298         LINVRNT(osd_invariant(obj));
2299
2300         return obj->oo_owner == env;
2301 }
2302
2303 static struct timespec *osd_inode_time(const struct lu_env *env,
2304                                        struct inode *inode, __u64 seconds)
2305 {
2306         struct osd_thread_info  *oti = osd_oti_get(env);
2307         struct timespec         *t   = &oti->oti_time;
2308
2309         t->tv_sec = seconds;
2310         t->tv_nsec = 0;
2311         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2312         return t;
2313 }
2314
2315 static void osd_inode_getattr(const struct lu_env *env,
2316                               struct inode *inode, struct lu_attr *attr)
2317 {
2318         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2319                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2320                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
2321                            LA_TYPE;
2322
2323         attr->la_atime   = LTIME_S(inode->i_atime);
2324         attr->la_mtime   = LTIME_S(inode->i_mtime);
2325         attr->la_ctime   = LTIME_S(inode->i_ctime);
2326         attr->la_mode    = inode->i_mode;
2327         attr->la_size    = i_size_read(inode);
2328         attr->la_blocks  = inode->i_blocks;
2329         attr->la_uid     = i_uid_read(inode);
2330         attr->la_gid     = i_gid_read(inode);
2331         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2332         attr->la_nlink   = inode->i_nlink;
2333         attr->la_rdev    = inode->i_rdev;
2334         attr->la_blksize = 1 << inode->i_blkbits;
2335         attr->la_blkbits = inode->i_blkbits;
2336 }
2337
2338 static int osd_attr_get(const struct lu_env *env,
2339                         struct dt_object *dt,
2340                         struct lu_attr *attr)
2341 {
2342         struct osd_object *obj = osd_dt_obj(dt);
2343
2344         if (unlikely(!dt_object_exists(dt)))
2345                 return -ENOENT;
2346         if (unlikely(obj->oo_destroyed))
2347                 return -ENOENT;
2348
2349         LASSERT(!dt_object_remote(dt));
2350         LINVRNT(osd_invariant(obj));
2351
2352         spin_lock(&obj->oo_guard);
2353         osd_inode_getattr(env, obj->oo_inode, attr);
2354         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2355                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2356         spin_unlock(&obj->oo_guard);
2357
2358         return 0;
2359 }
2360
2361 static int osd_declare_attr_set(const struct lu_env *env,
2362                                 struct dt_object *dt,
2363                                 const struct lu_attr *attr,
2364                                 struct thandle *handle)
2365 {
2366         struct osd_thandle     *oh;
2367         struct osd_object      *obj;
2368         struct osd_thread_info *info = osd_oti_get(env);
2369         struct lquota_id_info  *qi = &info->oti_qi;
2370         qid_t                   uid;
2371         qid_t                   gid;
2372         long long               bspace;
2373         int                     rc = 0;
2374         bool                    enforce;
2375         ENTRY;
2376
2377         LASSERT(dt != NULL);
2378         LASSERT(handle != NULL);
2379
2380         obj = osd_dt_obj(dt);
2381         LASSERT(osd_invariant(obj));
2382
2383         oh = container_of0(handle, struct osd_thandle, ot_super);
2384         LASSERT(oh->ot_handle == NULL);
2385
2386         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2387                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2388
2389         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2390                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2391
2392         if (attr == NULL || obj->oo_inode == NULL)
2393                 RETURN(rc);
2394
2395         bspace   = obj->oo_inode->i_blocks;
2396         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2397         bspace   = toqb(bspace);
2398
2399         /* Changing ownership is always preformed by super user, it should not
2400          * fail with EDQUOT.
2401          *
2402          * We still need to call the osd_declare_qid() to calculate the journal
2403          * credits for updating quota accounting files and to trigger quota
2404          * space adjustment once the operation is completed.*/
2405         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2406                 /* USERQUOTA */
2407                 uid = i_uid_read(obj->oo_inode);
2408                 qi->lqi_type = USRQUOTA;
2409                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2410                 /* inode accounting */
2411                 qi->lqi_is_blk = false;
2412
2413                 /* one more inode for the new uid ... */
2414                 qi->lqi_id.qid_uid = attr->la_uid;
2415                 qi->lqi_space      = 1;
2416                 /* Reserve credits for the new uid */
2417                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2418                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2419                         rc = 0;
2420                 if (rc)
2421                         RETURN(rc);
2422
2423                 /* and one less inode for the current uid */
2424                 qi->lqi_id.qid_uid = uid;
2425                 qi->lqi_space      = -1;
2426                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2427                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2428                         rc = 0;
2429                 if (rc)
2430                         RETURN(rc);
2431
2432                 /* block accounting */
2433                 qi->lqi_is_blk = true;
2434
2435                 /* more blocks for the new uid ... */
2436                 qi->lqi_id.qid_uid = attr->la_uid;
2437                 qi->lqi_space      = bspace;
2438                 /*
2439                  * Credits for the new uid has been reserved, re-use "obj"
2440                  * to save credit reservation.
2441                  */
2442                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2443                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2444                         rc = 0;
2445                 if (rc)
2446                         RETURN(rc);
2447
2448                 /* and finally less blocks for the current uid */
2449                 qi->lqi_id.qid_uid = uid;
2450                 qi->lqi_space      = -bspace;
2451                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2452                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2453                         rc = 0;
2454                 if (rc)
2455                         RETURN(rc);
2456
2457                 /* GROUP QUOTA */
2458                 gid = i_gid_read(obj->oo_inode);
2459                 qi->lqi_type = GRPQUOTA;
2460                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2461
2462                 /* inode accounting */
2463                 qi->lqi_is_blk = false;
2464
2465                 /* one more inode for the new gid ... */
2466                 qi->lqi_id.qid_gid = attr->la_gid;
2467                 qi->lqi_space      = 1;
2468                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2469                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2470                         rc = 0;
2471                 if (rc)
2472                         RETURN(rc);
2473
2474                 /* and one less inode for the current gid */
2475                 qi->lqi_id.qid_gid = gid;
2476                 qi->lqi_space      = -1;
2477                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2478                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2479                         rc = 0;
2480                 if (rc)
2481                         RETURN(rc);
2482
2483                 /* block accounting */
2484                 qi->lqi_is_blk = true;
2485
2486                 /* more blocks for the new gid ... */
2487                 qi->lqi_id.qid_gid = attr->la_gid;
2488                 qi->lqi_space      = bspace;
2489                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2490                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2491                         rc = 0;
2492                 if (rc)
2493                         RETURN(rc);
2494
2495                 /* and finally less blocks for the current gid */
2496                 qi->lqi_id.qid_gid = gid;
2497                 qi->lqi_space      = -bspace;
2498                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2499                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2500                         rc = 0;
2501                 if (rc)
2502                         RETURN(rc);
2503         }
2504
2505         RETURN(rc);
2506 }
2507
2508 static int osd_inode_setattr(const struct lu_env *env,
2509                              struct inode *inode, const struct lu_attr *attr)
2510 {
2511         __u64 bits = attr->la_valid;
2512
2513         /* Only allow set size for regular file */
2514         if (!S_ISREG(inode->i_mode))
2515                 bits &= ~(LA_SIZE | LA_BLOCKS);
2516
2517         if (bits == 0)
2518                 return 0;
2519
2520         if (bits & LA_ATIME)
2521                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2522         if (bits & LA_CTIME)
2523                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2524         if (bits & LA_MTIME)
2525                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2526         if (bits & LA_SIZE) {
2527                 spin_lock(&inode->i_lock);
2528                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2529                 i_size_write(inode, attr->la_size);
2530                 spin_unlock(&inode->i_lock);
2531         }
2532
2533         /* OSD should not change "i_blocks" which is used by quota.
2534          * "i_blocks" should be changed by ldiskfs only. */
2535         if (bits & LA_MODE)
2536                 inode->i_mode = (inode->i_mode & S_IFMT) |
2537                                 (attr->la_mode & ~S_IFMT);
2538         if (bits & LA_UID)
2539                 i_uid_write(inode, attr->la_uid);
2540         if (bits & LA_GID)
2541                 i_gid_write(inode, attr->la_gid);
2542         if (bits & LA_NLINK)
2543                 set_nlink(inode, attr->la_nlink);
2544         if (bits & LA_RDEV)
2545                 inode->i_rdev = attr->la_rdev;
2546
2547         if (bits & LA_FLAGS) {
2548                 /* always keep S_NOCMTIME */
2549                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2550                                  S_NOCMTIME;
2551         }
2552         return 0;
2553 }
2554
2555 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2556 {
2557         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2558             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2559                 struct iattr    iattr;
2560                 int             rc;
2561
2562                 ll_vfs_dq_init(inode);
2563                 iattr.ia_valid = 0;
2564                 if (attr->la_valid & LA_UID)
2565                         iattr.ia_valid |= ATTR_UID;
2566                 if (attr->la_valid & LA_GID)
2567                         iattr.ia_valid |= ATTR_GID;
2568                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2569                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2570
2571                 rc = ll_vfs_dq_transfer(inode, &iattr);
2572                 if (rc) {
2573                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2574                                "enforcement enabled on the ldiskfs "
2575                                "filesystem?\n", inode->i_sb->s_id, rc);
2576                         return rc;
2577                 }
2578         }
2579         return 0;
2580 }
2581
2582 static int osd_attr_set(const struct lu_env *env,
2583                         struct dt_object *dt,
2584                         const struct lu_attr *attr,
2585                         struct thandle *handle)
2586 {
2587         struct osd_object *obj = osd_dt_obj(dt);
2588         struct inode      *inode;
2589         int rc;
2590
2591         if (!dt_object_exists(dt))
2592                 return -ENOENT;
2593
2594         LASSERT(handle != NULL);
2595         LASSERT(!dt_object_remote(dt));
2596         LASSERT(osd_invariant(obj));
2597
2598         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2599
2600         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2601                 struct osd_thread_info  *oti  = osd_oti_get(env);
2602                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2603                 struct lu_fid           *fid1 = &oti->oti_fid;
2604                 struct osd_inode_id     *id   = &oti->oti_id;
2605                 struct iam_path_descr   *ipd;
2606                 struct iam_container    *bag;
2607                 struct osd_thandle      *oh;
2608                 int                      rc;
2609
2610                 fid_cpu_to_be(fid1, fid0);
2611                 memset(id, 1, sizeof(*id));
2612                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2613                                   fid0)->oi_dir.od_container;
2614                 ipd = osd_idx_ipd_get(env, bag);
2615                 if (unlikely(ipd == NULL))
2616                         RETURN(-ENOMEM);
2617
2618                 oh = container_of0(handle, struct osd_thandle, ot_super);
2619                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2620                                 (const struct iam_rec *)id, ipd);
2621                 osd_ipd_put(env, bag, ipd);
2622                 return(rc > 0 ? 0 : rc);
2623         }
2624
2625         inode = obj->oo_inode;
2626
2627         rc = osd_quota_transfer(inode, attr);
2628         if (rc)
2629                 return rc;
2630
2631         spin_lock(&obj->oo_guard);
2632         rc = osd_inode_setattr(env, inode, attr);
2633         spin_unlock(&obj->oo_guard);
2634         if (rc != 0)
2635                 GOTO(out, rc);
2636
2637         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2638
2639         if (!(attr->la_valid & LA_FLAGS))
2640                 GOTO(out, rc);
2641
2642         /* Let's check if there are extra flags need to be set into LMA */
2643         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2644                 struct osd_thread_info *info = osd_oti_get(env);
2645                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2646
2647                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2648                 if (rc != 0)
2649                         GOTO(out, rc);
2650
2651                 lma->lma_incompat |=
2652                         lustre_to_lma_flags(attr->la_flags);
2653                 lustre_lma_swab(lma);
2654                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2655                                      lma, sizeof(*lma), XATTR_REPLACE);
2656                 if (rc != 0) {
2657                         struct osd_device *osd = osd_obj2dev(obj);
2658
2659                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2660                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2661                               lma->lma_incompat, rc);
2662                 } else {
2663                         obj->oo_lma_flags =
2664                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2665                 }
2666                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2667         }
2668 out:
2669         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2670
2671         return rc;
2672 }
2673
2674 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2675                                            struct osd_object *obj,
2676                                            const char *name, const int namelen)
2677 {
2678         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2679 }
2680
2681 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2682                       umode_t mode, struct dt_allocation_hint *hint,
2683                       struct thandle *th)
2684 {
2685         int result;
2686         struct osd_device  *osd = osd_obj2dev(obj);
2687         struct osd_thandle *oth;
2688         struct dt_object   *parent = NULL;
2689         struct inode       *inode;
2690
2691         LINVRNT(osd_invariant(obj));
2692         LASSERT(obj->oo_inode == NULL);
2693         LASSERT(obj->oo_hl_head == NULL);
2694
2695         if (S_ISDIR(mode) && ldiskfs_pdo) {
2696                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2697                 if (obj->oo_hl_head == NULL)
2698                         return -ENOMEM;
2699         }
2700
2701         oth = container_of(th, struct osd_thandle, ot_super);
2702         LASSERT(oth->ot_handle->h_transaction != NULL);
2703
2704         if (hint != NULL && hint->dah_parent != NULL &&
2705             !dt_object_remote(hint->dah_parent))
2706                 parent = hint->dah_parent;
2707
2708         inode = ldiskfs_create_inode(oth->ot_handle,
2709                                      parent ? osd_dt_obj(parent)->oo_inode :
2710                                               osd_sb(osd)->s_root->d_inode,
2711                                      mode);
2712         if (!IS_ERR(inode)) {
2713                 /* Do not update file c/mtime in ldiskfs. */
2714                 inode->i_flags |= S_NOCMTIME;
2715
2716                 /* For new created object, it must be consistent,
2717                  * and it is unnecessary to scrub against it. */
2718                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2719
2720                 obj->oo_inode = inode;
2721                 result = 0;
2722         } else {
2723                 if (obj->oo_hl_head != NULL) {
2724                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2725                         obj->oo_hl_head = NULL;
2726                 }
2727                 result = PTR_ERR(inode);
2728         }
2729         LINVRNT(osd_invariant(obj));
2730         return result;
2731 }
2732
2733 enum {
2734         OSD_NAME_LEN = 255
2735 };
2736
2737 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2738                      struct lu_attr *attr,
2739                      struct dt_allocation_hint *hint,
2740                      struct dt_object_format *dof,
2741                      struct thandle *th)
2742 {
2743         int result;
2744         struct osd_thandle *oth;
2745         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2746
2747         LASSERT(S_ISDIR(attr->la_mode));
2748
2749         oth = container_of(th, struct osd_thandle, ot_super);
2750         LASSERT(oth->ot_handle->h_transaction != NULL);
2751         result = osd_mkfile(info, obj, mode, hint, th);
2752
2753         return result;
2754 }
2755
2756 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2757                         struct lu_attr *attr,
2758                         struct dt_allocation_hint *hint,
2759                         struct dt_object_format *dof,
2760                         struct thandle *th)
2761 {
2762         int result;
2763         struct osd_thandle *oth;
2764         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2765
2766         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2767
2768         LASSERT(S_ISREG(attr->la_mode));
2769
2770         oth = container_of(th, struct osd_thandle, ot_super);
2771         LASSERT(oth->ot_handle->h_transaction != NULL);
2772
2773         result = osd_mkfile(info, obj, mode, hint, th);
2774         if (result == 0) {
2775                 LASSERT(obj->oo_inode != NULL);
2776                 if (feat->dif_flags & DT_IND_VARKEY)
2777                         result = iam_lvar_create(obj->oo_inode,
2778                                                  feat->dif_keysize_max,
2779                                                  feat->dif_ptrsize,
2780                                                  feat->dif_recsize_max,
2781                                                  oth->ot_handle);
2782                 else
2783                         result = iam_lfix_create(obj->oo_inode,
2784                                                  feat->dif_keysize_max,
2785                                                  feat->dif_ptrsize,
2786                                                  feat->dif_recsize_max,
2787                                                  oth->ot_handle);
2788
2789         }
2790         return result;
2791 }
2792
2793 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2794                      struct lu_attr *attr,
2795                      struct dt_allocation_hint *hint,
2796                      struct dt_object_format *dof,
2797                      struct thandle *th)
2798 {
2799         LASSERT(S_ISREG(attr->la_mode));
2800         return osd_mkfile(info, obj, (attr->la_mode &
2801                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2802 }
2803
2804 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2805                      struct lu_attr *attr,
2806                      struct dt_allocation_hint *hint,
2807                      struct dt_object_format *dof,
2808                      struct thandle *th)
2809 {
2810         LASSERT(S_ISLNK(attr->la_mode));
2811         return osd_mkfile(info, obj, (attr->la_mode &
2812                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2813 }
2814
2815 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2816                      struct lu_attr *attr,
2817                      struct dt_allocation_hint *hint,
2818                      struct dt_object_format *dof,
2819                      struct thandle *th)
2820 {
2821         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2822         int result;
2823
2824         LINVRNT(osd_invariant(obj));
2825         LASSERT(obj->oo_inode == NULL);
2826         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2827                 S_ISFIFO(mode) || S_ISSOCK(mode));
2828
2829         result = osd_mkfile(info, obj, mode, hint, th);
2830         if (result == 0) {
2831                 LASSERT(obj->oo_inode != NULL);
2832                 /*
2833                  * This inode should be marked dirty for i_rdev.  Currently
2834                  * that is done in the osd_attr_init().
2835                  */
2836                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2837                                    attr->la_rdev);
2838         }
2839         LINVRNT(osd_invariant(obj));
2840         return result;
2841 }
2842
2843 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2844                               struct lu_attr *,
2845                               struct dt_allocation_hint *hint,
2846                               struct dt_object_format *dof,
2847                               struct thandle *);
2848
2849 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2850 {
2851         osd_obj_type_f result;
2852
2853         switch (type) {
2854         case DFT_DIR:
2855                 result = osd_mkdir;
2856                 break;
2857         case DFT_REGULAR:
2858                 result = osd_mkreg;
2859                 break;
2860         case DFT_SYM:
2861                 result = osd_mksym;
2862                 break;
2863         case DFT_NODE:
2864                 result = osd_mknod;
2865                 break;
2866         case DFT_INDEX:
2867                 result = osd_mk_index;
2868                 break;
2869
2870         default:
2871                 LBUG();
2872                 break;
2873         }
2874         return result;
2875 }
2876
2877
2878 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2879                         struct dt_object *parent, struct dt_object *child,
2880                         umode_t child_mode)
2881 {
2882         LASSERT(ah);
2883
2884         ah->dah_parent = parent;
2885         ah->dah_mode = child_mode;
2886
2887         if (parent != NULL && !dt_object_remote(parent)) {
2888                 /* will help to find FID->ino at dt_insert("..") */
2889                 struct osd_object *pobj = osd_dt_obj(parent);
2890                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2891         }
2892 }
2893
2894 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2895                           struct lu_attr *attr, struct dt_object_format *dof)
2896 {
2897         struct inode   *inode = obj->oo_inode;
2898         __u64           valid = attr->la_valid;
2899         int             result;
2900
2901         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2902
2903         if (dof->dof_type != DFT_NODE)
2904                 attr->la_valid &= ~LA_RDEV;
2905         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2906                 attr->la_valid &= ~LA_ATIME;
2907         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2908                 attr->la_valid &= ~LA_CTIME;
2909         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2910                 attr->la_valid &= ~LA_MTIME;
2911
2912         result = osd_quota_transfer(inode, attr);
2913         if (result)
2914                 return;
2915
2916         if (attr->la_valid != 0) {
2917                 result = osd_inode_setattr(info->oti_env, inode, attr);
2918                 /*
2919                  * The osd_inode_setattr() should always succeed here.  The
2920                  * only error that could be returned is EDQUOT when we are
2921                  * trying to change the UID or GID of the inode. However, this
2922                  * should not happen since quota enforcement is no longer
2923                  * enabled on ldiskfs (lquota takes care of it).
2924                  */
2925                 LASSERTF(result == 0, "%d\n", result);
2926                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2927         }
2928
2929         attr->la_valid = valid;
2930 }
2931
2932 /**
2933  * Helper function for osd_object_create()
2934  *
2935  * \retval 0, on success
2936  */
2937 static int __osd_object_create(struct osd_thread_info *info,
2938                                struct osd_object *obj, struct lu_attr *attr,
2939                                struct dt_allocation_hint *hint,
2940                                struct dt_object_format *dof,
2941                                struct thandle *th)
2942 {
2943         int     result;
2944         __u32   umask;
2945
2946         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2947
2948         /* we drop umask so that permissions we pass are not affected */
2949         umask = current->fs->umask;
2950         current->fs->umask = 0;
2951
2952         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2953                                                   th);
2954         if (likely(obj->oo_inode != NULL)) {
2955                 LASSERT(obj->oo_inode->i_state & I_NEW);
2956
2957                 /* Unlock the inode before attr initialization to avoid
2958                  * unnecessary dqget operations. LU-6378 */
2959                 unlock_new_inode(obj->oo_inode);
2960         }
2961
2962         if (likely(result == 0)) {
2963                 osd_attr_init(info, obj, attr, dof);
2964                 osd_object_init0(obj);
2965         }
2966
2967         /* restore previous umask value */
2968         current->fs->umask = umask;
2969
2970         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2971
2972         return result;
2973 }
2974
2975 /**
2976  * Helper function for osd_object_create()
2977  *
2978  * \retval 0, on success
2979  */
2980 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2981                            const struct lu_fid *fid, struct thandle *th)
2982 {
2983         struct osd_thread_info *info = osd_oti_get(env);
2984         struct osd_inode_id    *id   = &info->oti_id;
2985         struct osd_device      *osd  = osd_obj2dev(obj);
2986         struct osd_thandle     *oh;
2987         int                     rc;
2988
2989         LASSERT(obj->oo_inode != NULL);
2990
2991         oh = container_of0(th, struct osd_thandle, ot_super);
2992         LASSERT(oh->ot_handle);
2993         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2994
2995         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2996         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
2997                            OI_CHECK_FLD, NULL);
2998         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2999
3000         return rc;
3001 }
3002
3003 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3004                    u64 seq, struct lu_seq_range *range)
3005 {
3006         struct seq_server_site  *ss = osd_seq_site(osd);
3007
3008         if (fid_seq_is_idif(seq)) {
3009                 fld_range_set_ost(range);
3010                 range->lsr_index = idif_ost_idx(seq);
3011                 return 0;
3012         }
3013
3014         if (!fid_seq_in_fldb(seq)) {
3015                 fld_range_set_mdt(range);
3016                 if (ss != NULL)
3017                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3018                          * at all */
3019                         range->lsr_index = ss->ss_node_id;
3020                 return 0;
3021         }
3022
3023         LASSERT(ss != NULL);
3024         fld_range_set_any(range);
3025         /* OSD will only do local fld lookup */
3026         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3027 }
3028
3029 /*
3030  * Concurrency: no external locking is necessary.
3031  */
3032 static int osd_declare_object_create(const struct lu_env *env,
3033                                      struct dt_object *dt,
3034                                      struct lu_attr *attr,
3035                                      struct dt_allocation_hint *hint,
3036                                      struct dt_object_format *dof,
3037                                      struct thandle *handle)
3038 {
3039         struct osd_thandle      *oh;
3040         int                      rc;
3041         ENTRY;
3042
3043         LASSERT(handle != NULL);
3044
3045         oh = container_of0(handle, struct osd_thandle, ot_super);
3046         LASSERT(oh->ot_handle == NULL);
3047
3048         /* EA object consumes more credits than regular object: osd_mk_index
3049          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3050          * leaf_node, could involves the block, block bitmap, groups, GDT
3051          * change for each block, so add 4 * 2 credits in that case. */
3052         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3053                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3054                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3055         /* Reuse idle OI block may cause additional one OI block
3056          * to be changed. */
3057         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3058                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3059
3060         if (!attr)
3061                 RETURN(0);
3062
3063         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
3064                                    osd_dt_obj(dt), false, NULL, false);
3065         if (rc != 0)
3066                 RETURN(rc);
3067
3068         /* will help to find FID->ino mapping at dt_insert() */
3069         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3070                                    osd_dt_obj(dt));
3071
3072         RETURN(rc);
3073 }
3074
3075 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3076                              struct lu_attr *attr,
3077                              struct dt_allocation_hint *hint,
3078                              struct dt_object_format *dof, struct thandle *th)
3079 {
3080         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3081         struct osd_object       *obj    = osd_dt_obj(dt);
3082         struct osd_thread_info  *info   = osd_oti_get(env);
3083         int result;
3084         ENTRY;
3085
3086         if (dt_object_exists(dt))
3087                 return -EEXIST;
3088
3089         LINVRNT(osd_invariant(obj));
3090         LASSERT(!dt_object_remote(dt));
3091         LASSERT(osd_write_locked(env, obj));
3092         LASSERT(th != NULL);
3093
3094         if (unlikely(fid_is_acct(fid)))
3095                 /* Quota files can't be created from the kernel any more,
3096                  * 'tune2fs -O quota' will take care of creating them */
3097                 RETURN(-EPERM);
3098
3099         result = __osd_object_create(info, obj, attr, hint, dof, th);
3100         if (result == 0) {
3101                 result = __osd_oi_insert(env, obj, fid, th);
3102                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3103                         obj->oo_dt.do_body_ops = &osd_body_ops;
3104         }
3105         LASSERT(ergo(result == 0,
3106                 dt_object_exists(dt) && !dt_object_remote(dt)));
3107
3108         LASSERT(osd_invariant(obj));
3109         RETURN(result);
3110 }
3111
3112 /**
3113  * Called to destroy on-disk representation of the object
3114  *
3115  * Concurrency: must be locked
3116  */
3117 static int osd_declare_object_destroy(const struct lu_env *env,
3118                                       struct dt_object *dt,
3119                                       struct thandle *th)
3120 {
3121         struct osd_object  *obj = osd_dt_obj(dt);
3122         struct inode       *inode = obj->oo_inode;
3123         struct osd_thandle *oh;
3124         int                 rc;
3125         ENTRY;
3126
3127         if (inode == NULL)
3128                 RETURN(-ENOENT);
3129
3130         oh = container_of0(th, struct osd_thandle, ot_super);
3131         LASSERT(oh->ot_handle == NULL);
3132
3133         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3134                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3135         /* Recycle idle OI leaf may cause additional three OI blocks
3136          * to be changed. */
3137         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3138                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3139                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3140         /* one less inode */
3141         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3142                                    -1, oh, obj, false, NULL, false);
3143         if (rc)
3144                 RETURN(rc);
3145         /* data to be truncated */
3146         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3147                                    0, oh, obj, true, NULL, false);
3148         if (rc)
3149                 RETURN(rc);
3150
3151         /* will help to find FID->ino when this object is being
3152          * added to PENDING/ */
3153         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3154
3155         RETURN(rc);
3156 }
3157
3158 static int osd_object_destroy(const struct lu_env *env,
3159                               struct dt_object *dt,
3160                               struct thandle *th)
3161 {
3162         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
3163         struct osd_object      *obj = osd_dt_obj(dt);
3164         struct inode           *inode = obj->oo_inode;
3165         struct osd_device      *osd = osd_obj2dev(obj);
3166         struct osd_thandle     *oh;
3167         int                     result;
3168         ENTRY;
3169
3170         oh = container_of0(th, struct osd_thandle, ot_super);
3171         LASSERT(oh->ot_handle);
3172         LASSERT(inode);
3173         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3174
3175         if (unlikely(fid_is_acct(fid)))
3176                 RETURN(-EPERM);
3177
3178         if (S_ISDIR(inode->i_mode)) {
3179                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3180                         inode->i_nlink == 2);
3181                 /* it will check/delete the inode from remote parent,
3182                  * how to optimize it? unlink performance impaction XXX */
3183                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3184                 if (result != 0)
3185                         CERROR("%s: delete inode "DFID": rc = %d\n",
3186                                osd_name(osd), PFID(fid), result);
3187
3188                 spin_lock(&obj->oo_guard);
3189                 clear_nlink(inode);
3190                 spin_unlock(&obj->oo_guard);
3191                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3192         }
3193
3194         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3195
3196         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3197
3198         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3199                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3200                                        oh->ot_handle, OI_CHECK_FLD);
3201
3202         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3203         /* XXX: add to ext3 orphan list */
3204         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3205
3206         /* not needed in the cache anymore */
3207         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3208         obj->oo_destroyed = 1;
3209
3210         RETURN(0);
3211 }
3212
3213 /**
3214  * Put the fid into lustre_mdt_attrs, and then place the structure
3215  * inode's ea. This fid should not be altered during the life time
3216  * of the inode.
3217  *
3218  * \retval +ve, on success
3219  * \retval -ve, on error
3220  *
3221  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3222  */
3223 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3224                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3225 {
3226         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3227         int                      rc;
3228         ENTRY;
3229
3230         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3231                 RETURN(0);
3232
3233         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3234                 rc = -ENOMEM;
3235
3236         lustre_lma_init(lma, fid, compat, incompat);
3237         lustre_lma_swab(lma);
3238
3239         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
3240                              XATTR_CREATE);
3241         /* LMA may already exist, but we need to check that all the
3242          * desired compat/incompat flags have been added. */
3243         if (unlikely(rc == -EEXIST)) {
3244                 if (compat == 0 && incompat == 0)
3245                         RETURN(0);
3246
3247                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3248                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
3249                                      LMA_OLD_SIZE);
3250                 if (rc <= 0)
3251                         RETURN(-EINVAL);
3252
3253                 lustre_lma_swab(lma);
3254                 if (!(~lma->lma_compat & compat) &&
3255                     !(~lma->lma_incompat & incompat))
3256                         RETURN(0);
3257
3258                 lma->lma_compat |= compat;
3259                 lma->lma_incompat |= incompat;
3260                 lustre_lma_swab(lma);
3261                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3262                                      sizeof(*lma), XATTR_REPLACE);
3263         }
3264
3265         RETURN(rc);
3266 }
3267
3268 /**
3269  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3270  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3271  * To have compatilibility with 1.8 ldiskfs driver we need to have
3272  * magic number at start of fid data.
3273  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3274  * its inmemory API.
3275  */
3276 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3277                                          const struct lu_fid *fid)
3278 {
3279         if (!fid_is_namespace_visible(fid) ||
3280             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3281                 param->edp_magic = 0;
3282                 return;
3283         }
3284
3285         param->edp_magic = LDISKFS_LUFID_MAGIC;
3286         param->edp_len =  sizeof(struct lu_fid) + 1;
3287         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3288 }
3289
3290 /**
3291  * Try to read the fid from inode ea into dt_rec.
3292  *
3293  * \param fid object fid.
3294  *
3295  * \retval 0 on success
3296  */
3297 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3298                           __u32 ino, struct lu_fid *fid,
3299                           struct osd_inode_id *id)
3300 {
3301         struct osd_thread_info *info  = osd_oti_get(env);
3302         struct inode           *inode;
3303         ENTRY;
3304
3305         osd_id_gen(id, ino, OSD_OII_NOGEN);
3306         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3307         if (IS_ERR(inode))
3308                 RETURN(PTR_ERR(inode));
3309
3310         iput(inode);
3311         RETURN(0);
3312 }
3313
3314 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3315                                         struct inode *dir,
3316                                         struct inode *parent_dir,
3317                                         const struct lu_fid *dot_fid,
3318                                         const struct lu_fid *dot_dot_fid,
3319                                         struct osd_thandle *oth)
3320 {
3321         struct ldiskfs_dentry_param *dot_ldp;
3322         struct ldiskfs_dentry_param *dot_dot_ldp;
3323         __u32 saved_nlink = dir->i_nlink;
3324         int rc;
3325
3326         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3327         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3328
3329         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3330         dot_ldp->edp_magic = 0;
3331
3332         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3333                                     dir, dot_ldp, dot_dot_ldp);
3334         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3335          * the subseqent ref_add() will increase the dir->i_nlink
3336          * as 3. That is incorrect for new created directory.
3337          *
3338          * It looks like hack, because we want to make the OSD API
3339          * to be order-independent for new created directory object
3340          * between dt_insert(..) and ref_add() operations.
3341          *
3342          * Here, we only restore the in-RAM dir-inode's nlink attr,
3343          * becuase if the nlink attr is not 2, then there will be
3344          * ref_add() called following the dt_insert(..), such call
3345          * will make both the in-RAM and on-disk dir-inode's nlink
3346          * attr to be set as 2. LU-7447 */
3347         set_nlink(dir, saved_nlink);
3348         return rc;
3349 }
3350
3351 /**
3352  * Create an local agent inode for remote entry
3353  */
3354 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3355                                                   struct osd_device *osd,
3356                                                   struct osd_object *pobj,
3357                                                   const struct lu_fid *fid,
3358                                                   __u32 type,
3359                                                   struct thandle *th)
3360 {
3361         struct osd_thread_info  *info = osd_oti_get(env);
3362         struct inode            *local;
3363         struct osd_thandle      *oh;
3364         int                     rc;
3365         ENTRY;
3366
3367         LASSERT(th);
3368         oh = container_of(th, struct osd_thandle, ot_super);
3369         LASSERT(oh->ot_handle->h_transaction != NULL);
3370
3371         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3372         if (IS_ERR(local)) {
3373                 CERROR("%s: create local error %d\n", osd_name(osd),
3374                        (int)PTR_ERR(local));
3375                 RETURN(local);
3376         }
3377
3378         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3379         unlock_new_inode(local);
3380
3381         /* Set special LMA flag for local agent inode */
3382         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3383         if (rc != 0) {
3384                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3385                        osd_name(osd), PFID(fid), rc);
3386                 RETURN(ERR_PTR(rc));
3387         }
3388
3389         if (!S_ISDIR(type))
3390                 RETURN(local);
3391
3392         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3393                                          lu_object_fid(&pobj->oo_dt.do_lu),
3394                                          fid, oh);
3395         if (rc != 0) {
3396                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3397                         osd_name(osd), PFID(fid), rc);
3398                 RETURN(ERR_PTR(rc));
3399         }
3400
3401         RETURN(local);
3402 }
3403
3404 /**
3405  * when direntry is deleted, we have to take care of possible agent inode
3406  * referenced by that. unfortunately we can't do this at that point:
3407  * iget() within a running transaction leads to deadlock and we better do
3408  * not call that every delete declaration to save performance. so we put
3409  * a potention agent inode on a list and process that once the transaction
3410  * is over. Notice it's not any worse in terms of real orphans as regular
3411  * object destroy doesn't put inodes on the on-disk orphan list. this should
3412  * be addressed separately
3413  */
3414 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3415                                             struct osd_thandle *oh,
3416                                             __u32 ino)
3417 {
3418         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3419         struct osd_obj_orphan *oor;
3420
3421         OBD_ALLOC_PTR(oor);
3422         if (oor == NULL)
3423                 return -ENOMEM;
3424
3425         oor->oor_ino = ino;
3426         oor->oor_env = (struct lu_env *)env;
3427         spin_lock(&osd->od_osfs_lock);
3428         list_add(&oor->oor_list, &osd->od_orphan_list);
3429         spin_unlock(&osd->od_osfs_lock);
3430
3431         oh->ot_remove_agents = 1;
3432
3433         return 0;
3434
3435 }
3436
3437 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3438                                                 struct osd_device *osd)
3439 {
3440         struct osd_thread_info *info = osd_oti_get(env);
3441         struct osd_obj_orphan *oor, *tmp;
3442         struct osd_inode_id id;
3443         struct list_head list;
3444         struct inode *inode;
3445         struct lu_fid fid;
3446         handle_t *jh;
3447         __u32 ino;
3448
3449         INIT_LIST_HEAD(&list);
3450
3451         spin_lock(&osd->od_osfs_lock);
3452         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3453                 if (oor->oor_env == env) {
3454                         list_del(&oor->oor_list);
3455                         list_add(&oor->oor_list, &list);
3456                 }
3457         }
3458         spin_unlock(&osd->od_osfs_lock);
3459
3460         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3461
3462                 ino = oor->oor_ino;
3463
3464                 list_del(&oor->oor_list);
3465                 OBD_FREE_PTR(oor);
3466
3467                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3468                 inode = osd_iget_fid(info, osd, &id, &fid);
3469                 if (IS_ERR(inode))
3470                         continue;
3471
3472                 if (!osd_remote_fid(env, osd, &fid)) {
3473                         iput(inode);
3474                         continue;
3475                 }
3476
3477                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3478                 clear_nlink(inode);
3479                 mark_inode_dirty(inode);
3480                 ldiskfs_journal_stop(jh);
3481                 iput(inode);
3482         }
3483
3484         return 0;
3485 }
3486
3487 /**
3488  * OSD layer object create function for interoperability mode (b11826).
3489  * This is mostly similar to osd_object_create(). Only difference being, fid is
3490  * inserted into inode ea here.
3491  *
3492  * \retval   0, on success
3493  * \retval -ve, on error
3494  */
3495 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3496                                 struct lu_attr *attr,
3497                                 struct dt_allocation_hint *hint,
3498                                 struct dt_object_format *dof,
3499                                 struct thandle *th)
3500 {
3501         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3502         struct osd_object       *obj    = osd_dt_obj(dt);
3503         struct osd_thread_info  *info   = osd_oti_get(env);
3504         int                      result, on_ost = 0;
3505
3506         ENTRY;
3507
3508         if (dt_object_exists(dt))
3509                 RETURN(-EEXIST);
3510
3511         LASSERT(osd_invariant(obj));
3512         LASSERT(!dt_object_remote(dt));
3513         LASSERT(osd_write_locked(env, obj));
3514         LASSERT(th != NULL);
3515
3516         if (unlikely(fid_is_acct(fid)))
3517                 /* Quota files can't be created from the kernel any more,
3518                  * 'tune2fs -O quota' will take care of creating them */
3519                 RETURN(-EPERM);
3520
3521         result = __osd_object_create(info, obj, attr, hint, dof, th);
3522         if (result == 0) {
3523                 if (fid_is_idif(fid) &&
3524                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3525                         struct lu_fid *tfid = &info->oti_fid;
3526                         struct ost_id *oi   = &info->oti_ostid;
3527
3528                         fid_to_ostid(fid, oi);
3529                         ostid_to_fid(tfid, oi, 0);
3530                         on_ost = 1;
3531                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3532                                                 LMAC_FID_ON_OST, 0);
3533                 } else {
3534                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3535                                                fid, OI_CHECK_FLD);
3536                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3537                                                 on_ost ? LMAC_FID_ON_OST : 0,
3538                                                 0);
3539                 }
3540                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3541                         obj->oo_dt.do_body_ops = &osd_body_ops;
3542         }
3543
3544         if (result == 0)
3545                 result = __osd_oi_insert(env, obj, fid, th);
3546
3547         /* a small optimization - dt_insert() isn't usually applied
3548          * to OST objects, so we don't need to cache OI mapping for
3549          * OST objects */
3550         if (result == 0 && on_ost == 0) {
3551                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3552                 result = osd_idc_find_and_init(env, osd, obj);
3553                 LASSERT(result == 0);
3554         }
3555
3556         LASSERT(ergo(result == 0,
3557                      dt_object_exists(dt) && !dt_object_remote(dt)));
3558         LINVRNT(osd_invariant(obj));
3559         RETURN(result);
3560 }
3561
3562 static int osd_declare_object_ref_add(const struct lu_env *env,
3563                                       struct dt_object *dt,
3564                                       struct thandle *handle)
3565 {
3566         struct osd_thandle       *oh;
3567
3568         /* it's possible that object doesn't exist yet */
3569         LASSERT(handle != NULL);
3570
3571         oh = container_of0(handle, struct osd_thandle, ot_super);
3572         LASSERT(oh->ot_handle == NULL);
3573
3574         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3575                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3576
3577         return 0;
3578 }
3579
3580 /*
3581  * Concurrency: @dt is write locked.
3582  */
3583 static int osd_object_ref_add(const struct lu_env *env,
3584                               struct dt_object *dt, struct thandle *th)
3585 {
3586         struct osd_object  *obj = osd_dt_obj(dt);
3587         struct inode       *inode = obj->oo_inode;
3588         struct osd_thandle *oh;
3589         int                 rc = 0;
3590
3591         if (!dt_object_exists(dt) || obj->oo_destroyed)
3592                 return -ENOENT;
3593
3594         LINVRNT(osd_invariant(obj));
3595         LASSERT(!dt_object_remote(dt));
3596         LASSERT(osd_write_locked(env, obj));
3597         LASSERT(th != NULL);
3598
3599         oh = container_of0(th, struct osd_thandle, ot_super);
3600         LASSERT(oh->ot_handle != NULL);
3601
3602         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3603
3604         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3605                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3606         /*
3607          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3608          * (65000) subdirectories by storing "1" in i_nlink if the link count
3609          * would otherwise overflow. Directory tranversal tools understand
3610          * that (st_nlink == 1) indicates that the filesystem dose not track
3611          * hard links count on the directory, and will not abort subdirectory
3612          * scanning early once (st_nlink - 2) subdirs have been found.
3613          *
3614          * This also has to properly handle the case of inodes with nlink == 0
3615          * in case they are being linked into the PENDING directory
3616          */
3617         spin_lock(&obj->oo_guard);
3618         if (unlikely(inode->i_nlink == 0))
3619                 /* inc_nlink from 0 may cause WARN_ON */
3620                 set_nlink(inode, 1);
3621         else {
3622                 ldiskfs_inc_count(oh->ot_handle, inode);
3623                 if (!S_ISDIR(inode->i_mode))
3624                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3625         }
3626         spin_unlock(&obj->oo_guard);
3627
3628         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3629         LINVRNT(osd_invariant(obj));
3630
3631         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3632
3633         return rc;
3634 }
3635
3636 static int osd_declare_object_ref_del(const struct lu_env *env,
3637                                       struct dt_object *dt,
3638                                       struct thandle *handle)
3639 {
3640         struct osd_thandle *oh;
3641
3642         if (!dt_object_exists(dt))
3643                 return -ENOENT;
3644
3645         LASSERT(!dt_object_remote(dt));
3646         LASSERT(handle != NULL);
3647
3648         oh = container_of0(handle, struct osd_thandle, ot_super);
3649         LASSERT(oh->ot_handle == NULL);
3650
3651         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3652                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3653
3654         return 0;
3655 }
3656
3657 /*
3658  * Concurrency: @dt is write locked.
3659  */
3660 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3661                               struct thandle *th)
3662 {
3663         struct osd_object       *obj = osd_dt_obj(dt);
3664         struct inode            *inode = obj->oo_inode;
3665         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3666         struct osd_thandle      *oh;
3667
3668         if (!dt_object_exists(dt))
3669                 return -ENOENT;
3670
3671         LINVRNT(osd_invariant(obj));
3672         LASSERT(!dt_object_remote(dt));
3673         LASSERT(osd_write_locked(env, obj));
3674         LASSERT(th != NULL);
3675
3676         oh = container_of0(th, struct osd_thandle, ot_super);
3677         LASSERT(oh->ot_handle != NULL);
3678
3679         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3680
3681         spin_lock(&obj->oo_guard);
3682         /* That can be result of upgrade from old Lustre version and
3683          * applied only to local files.  Just skip this ref_del call.
3684          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3685         if (inode->i_nlink == 0) {
3686                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3687                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3688                              ", maybe an upgraded file? (LU-3915)\n",
3689                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3690                 spin_unlock(&obj->oo_guard);
3691                 return 0;
3692         }
3693
3694         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3695                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3696
3697         ldiskfs_dec_count(oh->ot_handle, inode);
3698         spin_unlock(&obj->oo_guard);
3699
3700         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3701         LINVRNT(osd_invariant(obj));
3702
3703         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3704
3705         return 0;
3706 }
3707
3708 /*
3709  * Get the 64-bit version for an inode.
3710  */
3711 static int osd_object_version_get(const struct lu_env *env,
3712                                   struct dt_object *dt, dt_obj_version_t *ver)
3713 {
3714         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3715
3716         CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3717                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3718         *ver = LDISKFS_I(inode)->i_fs_version;
3719         return 0;
3720 }
3721
3722 /*
3723  * Concurrency: @dt is read locked.
3724  */
3725 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3726                          struct lu_buf *buf, const char *name)
3727 {
3728         struct osd_object      *obj    = osd_dt_obj(dt);
3729         struct inode           *inode  = obj->oo_inode;
3730         struct osd_thread_info *info   = osd_oti_get(env);
3731         struct dentry          *dentry = &info->oti_obj_dentry;
3732         bool                    cache_xattr = false;
3733         int                     rc;
3734
3735         /* version get is not real XATTR but uses xattr API */
3736         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3737                 /* for version we are just using xattr API but change inode
3738                  * field instead */
3739                 if (buf->lb_len == 0)
3740                         return sizeof(dt_obj_version_t);
3741
3742                 if (buf->lb_len < sizeof(dt_obj_version_t))
3743                         return -ERANGE;
3744
3745                 osd_object_version_get(env, dt, buf->lb_buf);
3746
3747                 return sizeof(dt_obj_version_t);
3748         }
3749
3750         if (!dt_object_exists(dt))
3751                 return -ENOENT;
3752
3753         LASSERT(!dt_object_remote(dt));
3754         LASSERT(inode->i_op != NULL);
3755         LASSERT(inode->i_op->getxattr != NULL);
3756
3757         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3758             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3759                 cache_xattr = true;
3760
3761         if (cache_xattr) {
3762                 rc = osd_oxc_get(obj, name, buf);
3763                 if (rc != -ENOENT)
3764                         return rc;
3765         }
3766
3767         rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3768         if (cache_xattr) {
3769                 if (rc == -ENOENT || rc == -ENODATA)
3770                         osd_oxc_add(obj, name, NULL, 0);
3771                 else if (rc > 0 && buf->lb_buf != NULL)
3772                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3773         }
3774
3775         return rc;
3776 }
3777
3778 static int osd_declare_xattr_set(const struct lu_env *env,
3779                                  struct dt_object *dt,
3780                                  const struct lu_buf *buf, const char *name,
3781                                  int fl, struct thandle *handle)
3782 {
3783         struct osd_thandle *oh;
3784         int credits;
3785         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3786
3787         LASSERT(handle != NULL);
3788
3789         oh = container_of0(handle, struct osd_thandle, ot_super);
3790         LASSERT(oh->ot_handle == NULL);
3791
3792         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3793                 /* For non-upgrading case, the LMA is set first and
3794                  * usually fit inode. But for upgrade case, the LMA
3795                  * may be in another separated EA block. */
3796                 if (!dt_object_exists(dt))
3797                         credits = 0;
3798                 else if (fl == LU_XATTR_REPLACE)
3799                         credits = 1;
3800                 else
3801                         goto upgrade;
3802         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3803                 credits = 1;
3804         } else {
3805 upgrade:
3806                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3807
3808                 if (buf != NULL) {
3809                         ssize_t buflen;
3810
3811                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3812                                 /* learn xattr size from osd_xattr_get if
3813                                    attribute has not been read yet */
3814                                 buflen = __osd_xattr_get(
3815                                     osd_dt_obj(dt)->oo_inode,
3816                                     &osd_oti_get(env)->oti_obj_dentry,
3817                                     name, NULL, 0);
3818                                 if (buflen < 0)
3819                                         buflen = 0;
3820                         } else {
3821                                 buflen = buf->lb_len;
3822                         }
3823
3824                         if (buflen > sb->s_blocksize) {
3825                                 credits += osd_calc_bkmap_credits(
3826                                     sb, NULL, 0, -1,
3827                                     (buflen + sb->s_blocksize - 1) >>
3828                                     sb->s_blocksize_bits);
3829                         }
3830                 }
3831                 /*
3832                  * xattr set may involve inode quota change, reserve credits for
3833                  * dquot_initialize()
3834                  */
3835                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3836         }
3837
3838         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3839
3840         return 0;
3841 }
3842
3843 /*
3844  * Set the 64-bit version for object
3845  */
3846 static void osd_object_version_set(const struct lu_env *env,
3847                                    struct dt_object *dt,
3848                                    dt_obj_version_t *new_version)
3849 {
3850         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3851
3852         CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3853                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3854
3855         LDISKFS_I(inode)->i_fs_version = *new_version;
3856         /** Version is set after all inode operations are finished,
3857          *  so we should mark it dirty here */
3858         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3859 }
3860
3861 /*
3862  * Concurrency: @dt is write locked.
3863  */
3864 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3865                          const struct lu_buf *buf, const char *name, int fl,
3866                          struct thandle *handle)
3867 {
3868         struct osd_object      *obj      = osd_dt_obj(dt);
3869         struct inode           *inode    = obj->oo_inode;
3870         struct osd_thread_info *info     = osd_oti_get(env);
3871         int                     fs_flags = 0;
3872         int                     rc;
3873         ENTRY;
3874
3875         LASSERT(handle != NULL);
3876
3877         /* version set is not real XATTR */
3878         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3879                 /* for version we are just using xattr API but change inode
3880                  * field instead */
3881                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3882                 osd_object_version_set(env, dt, buf->lb_buf);
3883                 return sizeof(dt_obj_version_t);
3884         }
3885
3886         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3887                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3888
3889         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3890         if (fl & LU_XATTR_REPLACE)
3891                 fs_flags |= XATTR_REPLACE;
3892
3893         if (fl & LU_XATTR_CREATE)
3894                 fs_flags |= XATTR_CREATE;
3895
3896         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3897                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3898                 int                      rc;
3899
3900                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3901                 if (rc != 0)
3902                         RETURN(rc);
3903
3904                 lma->lma_incompat |= LMAI_STRIPED;
3905                 lustre_lma_swab(lma);
3906                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3907                                      sizeof(*lma), XATTR_REPLACE);
3908                 if (rc != 0)
3909                         RETURN(rc);
3910         }
3911
3912         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3913             strcmp(name, XATTR_NAME_LINK) == 0)
3914                 return -ENOSPC;
3915
3916         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3917                                fs_flags);
3918         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3919
3920         if (rc == 0 &&
3921             (strcmp(name, XATTR_NAME_LOV) == 0 ||
3922              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
3923                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
3924
3925         return rc;
3926 }
3927
3928 /*
3929  * Concurrency: @dt is read locked.
3930  */
3931 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3932                           const struct lu_buf *buf)
3933 {
3934         struct osd_object      *obj    = osd_dt_obj(dt);
3935         struct inode           *inode  = obj->oo_inode;
3936         struct osd_thread_info *info   = osd_oti_get(env);
3937         struct dentry          *dentry = &info->oti_obj_dentry;
3938
3939         if (!dt_object_exists(dt))
3940                 return -ENOENT;
3941
3942         LASSERT(!dt_object_remote(dt));
3943         LASSERT(inode->i_op != NULL);
3944         LASSERT(inode->i_op->listxattr != NULL);
3945
3946         dentry->d_inode = inode;
3947         dentry->d_sb = inode->i_sb;
3948         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3949 }
3950
3951 static int osd_declare_xattr_del(const struct lu_env *env,
3952                                  struct dt_object *dt, const char *name,
3953                                  struct thandle *handle)
3954 {
3955         struct osd_thandle *oh;
3956         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3957
3958         LASSERT(!dt_object_remote(dt));
3959         LASSERT(handle != NULL);
3960
3961         oh = container_of0(handle, struct osd_thandle, ot_super);
3962         LASSERT(oh->ot_handle == NULL);
3963
3964         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3965                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3966         /*
3967          * xattr del may involve inode quota change, reserve credits for
3968          * dquot_initialize()
3969          */
3970         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3971
3972         return 0;
3973 }
3974
3975 /*
3976  * Concurrency: @dt is write locked.
3977  */
3978 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3979                          const char *name, struct thandle *handle)
3980 {
3981         struct osd_object      *obj    = osd_dt_obj(dt);
3982         struct inode           *inode  = obj->oo_inode;
3983         struct osd_thread_info *info   = osd_oti_get(env);
3984         struct dentry          *dentry = &info->oti_obj_dentry;
3985         int                     rc;
3986
3987         if (!dt_object_exists(dt))
3988                 return -ENOENT;
3989
3990         LASSERT(!dt_object_remote(dt));
3991         LASSERT(inode->i_op != NULL);
3992         LASSERT(inode->i_op->removexattr != NULL);
3993         LASSERT(handle != NULL);
3994
3995         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3996
3997         ll_vfs_dq_init(inode);
3998         dentry->d_inode = inode;
3999         dentry->d_sb = inode->i_sb;
4000         rc = inode->i_op->removexattr(dentry, name);
4001         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4002
4003         if (rc == 0 &&
4004             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4005              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4006                 osd_oxc_del(obj, name);
4007
4008         return rc;
4009 }
4010
4011 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4012                            __u64 start, __u64 end)
4013 {
4014         struct osd_object       *obj    = osd_dt_obj(dt);
4015         struct inode            *inode  = obj->oo_inode;
4016         struct osd_thread_info  *info   = osd_oti_get(env);
4017         struct dentry           *dentry = &info->oti_obj_dentry;
4018         struct file             *file   = &info->oti_file;
4019         int                     rc;
4020
4021         ENTRY;
4022
4023         dentry->d_inode = inode;
4024         dentry->d_sb = inode->i_sb;
4025         file->f_path.dentry = dentry;
4026         file->f_mapping = inode->i_mapping;
4027         file->f_op = inode->i_fop;
4028         set_file_inode(file, inode);
4029
4030         rc = ll_vfs_fsync_range(file, start, end, 0);
4031
4032         RETURN(rc);
4033 }
4034
4035 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4036 {
4037         return 0;
4038 }
4039
4040 /*
4041  * Index operations.
4042  */
4043
4044 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4045                            const struct dt_index_features *feat)
4046 {
4047         struct iam_descr *descr;
4048
4049         if (osd_object_is_root(o))
4050                 return feat == &dt_directory_features;
4051
4052         LASSERT(o->oo_dir != NULL);
4053
4054         descr = o->oo_dir->od_container.ic_descr;
4055         if (feat == &dt_directory_features) {
4056                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4057                         return 1;
4058                 else
4059                         return 0;
4060         } else {
4061                 return
4062                         feat->dif_keysize_min <= descr->id_key_size &&
4063                         descr->id_key_size <= feat->dif_keysize_max &&
4064                         feat->dif_recsize_min <= descr->id_rec_size &&
4065                         descr->id_rec_size <= feat->dif_recsize_max &&
4066                         !(feat->dif_flags & (DT_IND_VARKEY |
4067                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4068                         ergo(feat->dif_flags & DT_IND_UPDATE,
4069                              1 /* XXX check that object (and file system) is
4070                                 * writable */);
4071         }
4072 }
4073
4074 static int osd_iam_container_init(const struct lu_env *env,
4075                                   struct osd_object *obj,
4076                                   struct osd_directory *dir)
4077 {
4078         struct iam_container *bag = &dir->od_container;
4079         int result;
4080
4081         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4082         if (result != 0)
4083                 return result;
4084
4085         result = iam_container_setup(bag);
4086         if (result == 0)
4087                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4088         else
4089                 iam_container_fini(bag);
4090
4091         return result;
4092 }
4093
4094
4095 /*
4096  * Concurrency: no external locking is necessary.
4097  */
4098 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4099                          const struct dt_index_features *feat)
4100 {
4101         int                      result;
4102         int                      skip_iam = 0;
4103         struct osd_object       *obj = osd_dt_obj(dt);
4104
4105         LINVRNT(osd_invariant(obj));
4106
4107         if (osd_object_is_root(obj)) {
4108                 dt->do_index_ops = &osd_index_ea_ops;
4109                 result = 0;
4110         } else if (feat == &dt_directory_features) {
4111                 dt->do_index_ops = &osd_index_ea_ops;
4112                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4113                         result = 0;
4114                 else
4115                         result = -ENOTDIR;
4116                 skip_iam = 1;
4117         } else if (unlikely(feat == &dt_otable_features)) {
4118                 dt->do_index_ops = &osd_otable_ops;
4119                 return 0;
4120         } else if (unlikely(feat == &dt_acct_features)) {
4121                 dt->do_index_ops = &osd_acct_index_ops;
4122                 result = 0;
4123                 skip_iam = 1;
4124         } else if (!osd_has_index(obj)) {
4125                 struct osd_directory *dir;
4126
4127                 OBD_ALLOC_PTR(dir);
4128                 if (dir != NULL) {
4129
4130                         spin_lock(&obj->oo_guard);
4131                         if (obj->oo_dir == NULL)
4132                                 obj->oo_dir = dir;
4133                         else
4134                                 /*
4135                                  * Concurrent thread allocated container data.
4136                                  */
4137                                 OBD_FREE_PTR(dir);
4138                         spin_unlock(&obj->oo_guard);
4139                         /*
4140                          * Now, that we have container data, serialize its
4141                          * initialization.
4142                          */
4143                         down_write(&obj->oo_ext_idx_sem);
4144                         /*
4145                          * recheck under lock.
4146                          */
4147                         if (!osd_has_index(obj))
4148                                 result = osd_iam_container_init(env, obj,
4149                                                                 obj->oo_dir);
4150                         else
4151                                 result = 0;
4152                         up_write(&obj->oo_ext_idx_sem);
4153                 } else {
4154                         result = -ENOMEM;
4155                 }
4156         } else {
4157                 result = 0;
4158         }
4159
4160         if (result == 0 && skip_iam == 0) {
4161                 if (!osd_iam_index_probe(env, obj, feat))
4162                         result = -ENOTDIR;
4163         }
4164         LINVRNT(osd_invariant(obj));
4165
4166         if (result == 0 && feat == &dt_quota_glb_features &&
4167             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
4168                 result = osd_quota_migration(env, dt);
4169
4170         return result;
4171 }
4172
4173 static int osd_otable_it_attr_get(const struct lu_env *env,
4174                                  struct dt_object *dt,
4175                                  struct lu_attr *attr)
4176 {
4177         attr->la_valid = 0;
4178         return 0;
4179 }
4180
4181 static const struct dt_object_operations osd_obj_ops = {
4182         .do_read_lock         = osd_object_read_lock,
4183         .do_write_lock        = osd_object_write_lock,
4184         .do_read_unlock       = osd_object_read_unlock,
4185         .do_write_unlock      = osd_object_write_unlock,
4186         .do_write_locked      = osd_object_write_locked,
4187         .do_attr_get          = osd_attr_get,
4188         .do_declare_attr_set  = osd_declare_attr_set,
4189         .do_attr_set          = osd_attr_set,
4190         .do_ah_init           = osd_ah_init,
4191         .do_declare_create    = osd_declare_object_create,
4192         .do_create            = osd_object_create,
4193         .do_declare_destroy   = osd_declare_object_destroy,
4194         .do_destroy           = osd_object_destroy,
4195         .do_index_try         = osd_index_try,
4196         .do_declare_ref_add   = osd_declare_object_ref_add,
4197         .do_ref_add           = osd_object_ref_add,
4198         .do_declare_ref_del   = osd_declare_object_ref_del,
4199         .do_ref_del           = osd_object_ref_del,
4200         .do_xattr_get         = osd_xattr_get,
4201         .do_declare_xattr_set = osd_declare_xattr_set,
4202         .do_xattr_set         = osd_xattr_set,
4203         .do_declare_xattr_del = osd_declare_xattr_del,
4204         .do_xattr_del         = osd_xattr_del,
4205         .do_xattr_list        = osd_xattr_list,
4206         .do_object_sync       = osd_object_sync,
4207         .do_invalidate        = osd_invalidate,
4208 };
4209
4210 /**
4211  * dt_object_operations for interoperability mode
4212  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4213  */
4214 static const struct dt_object_operations osd_obj_ea_ops = {
4215         .do_read_lock         = osd_object_read_lock,
4216         .do_write_lock        = osd_object_write_lock,
4217         .do_read_unlock       = osd_object_read_unlock,
4218         .do_write_unlock      = osd_object_write_unlock,
4219         .do_write_locked      = osd_object_write_locked,
4220         .do_attr_get          = osd_attr_get,
4221         .do_declare_attr_set  = osd_declare_attr_set,
4222         .do_attr_set          = osd_attr_set,
4223         .do_ah_init           = osd_ah_init,
4224         .do_declare_create    = osd_declare_object_create,
4225         .do_create            = osd_object_ea_create,
4226         .do_declare_destroy   = osd_declare_object_destroy,
4227         .do_destroy           = osd_object_destroy,
4228         .do_index_try         = osd_index_try,
4229         .do_declare_ref_add   = osd_declare_object_ref_add,
4230         .do_ref_add           = osd_object_ref_add,
4231         .do_declare_ref_del   = osd_declare_object_ref_del,
4232         .do_ref_del           = osd_object_ref_del,
4233         .do_xattr_get         = osd_xattr_get,
4234         .do_declare_xattr_set = osd_declare_xattr_set,
4235         .do_xattr_set         = osd_xattr_set,
4236         .do_declare_xattr_del = osd_declare_xattr_del,
4237         .do_xattr_del         = osd_xattr_del,
4238         .do_xattr_list        = osd_xattr_list,
4239         .do_object_sync       = osd_object_sync,
4240         .do_invalidate        = osd_invalidate,
4241 };
4242
4243 static const struct dt_object_operations osd_obj_otable_it_ops = {
4244         .do_attr_get    = osd_otable_it_attr_get,
4245         .do_index_try   = osd_index_try,
4246 };
4247
4248 static int osd_index_declare_iam_delete(const struct lu_env *env,
4249                                         struct dt_object *dt,
4250                                         const struct dt_key *key,
4251                                         struct thandle *handle)
4252 {
4253         struct osd_thandle    *oh;
4254
4255         oh = container_of0(handle, struct osd_thandle, ot_super);
4256         LASSERT(oh->ot_handle == NULL);
4257
4258         /* Recycle  may cause additional three blocks to be changed. */
4259         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4260                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4261
4262         return 0;
4263 }
4264
4265 /**
4266  *      delete a (key, value) pair from index \a dt specified by \a key
4267  *
4268  *      \param  dt      osd index object
4269  *      \param  key     key for index
4270  *      \param  rec     record reference
4271  *      \param  handle  transaction handler
4272  *
4273  *      \retval  0  success
4274  *      \retval -ve   failure
4275  */
4276 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4277                                 const struct dt_key *key,
4278                                 struct thandle *handle)
4279 {
4280         struct osd_thread_info *oti = osd_oti_get(env);
4281         struct osd_object      *obj = osd_dt_obj(dt);
4282         struct osd_thandle     *oh;
4283         struct iam_path_descr  *ipd;
4284         struct iam_container   *bag = &obj->oo_dir->od_container;
4285         int                     rc;
4286         ENTRY;
4287
4288         if (!dt_object_exists(dt))
4289                 RETURN(-ENOENT);
4290
4291         LINVRNT(osd_invariant(obj));
4292         LASSERT(!dt_object_remote(dt));
4293         LASSERT(bag->ic_object == obj->oo_inode);
4294         LASSERT(handle != NULL);
4295
4296         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4297
4298         ipd = osd_idx_ipd_get(env, bag);
4299         if (unlikely(ipd == NULL))
4300                 RETURN(-ENOMEM);
4301
4302         oh = container_of0(handle, struct osd_thandle, ot_super);
4303         LASSERT(oh->ot_handle != NULL);
4304         LASSERT(oh->ot_handle->h_transaction != NULL);
4305
4306         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4307                 /* swab quota uid/gid provided by caller */
4308                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4309                 key = (const struct dt_key *)&oti->oti_quota_id;
4310         }
4311
4312         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4313         osd_ipd_put(env, bag, ipd);
4314         LINVRNT(osd_invariant(obj));
4315         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4316         RETURN(rc);
4317 }
4318
4319 static int osd_index_declare_ea_delete(const struct lu_env *env,
4320                                        struct dt_object *dt,
4321                                        const struct dt_key *key,
4322                                        struct thandle *handle)
4323 {
4324         struct osd_thandle *oh;
4325         struct inode       *inode;
4326         int                 rc;
4327         ENTRY;
4328
4329         LASSERT(!dt_object_remote(dt));
4330         LASSERT(handle != NULL);
4331
4332         oh = container_of0(handle, struct osd_thandle, ot_super);
4333         LASSERT(oh->ot_handle == NULL);
4334
4335         /* due to DNE we may need to remove an agent inode */
4336         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4337                              osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4338                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
4339
4340         inode = osd_dt_obj(dt)->oo_inode;
4341         if (inode == NULL)
4342                 RETURN(-ENOENT);
4343
4344         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4345                                    0, oh, osd_dt_obj(dt), true, NULL, false);
4346         RETURN(rc);
4347 }
4348
4349 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4350                                           struct dt_rec *fid)
4351 {
4352         struct osd_fid_pack *rec;
4353         int                  rc = -ENODATA;
4354
4355         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4356                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4357                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4358                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4359                         rc = -EINVAL;
4360         }
4361         return rc;
4362 }
4363
4364 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4365                           const struct lu_fid *fid)
4366 {
4367         struct seq_server_site  *ss = osd_seq_site(osd);
4368         ENTRY;
4369
4370         /* FID seqs not in FLDB, must be local seq */
4371         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4372                 RETURN(0);
4373
4374         /* If FLD is not being initialized yet, it only happens during the
4375          * initialization, likely during mgs initialization, and we assume
4376          * this is local FID. */
4377         if (ss == NULL || ss->ss_server_fld == NULL)
4378                 RETURN(0);
4379
4380         /* Only check the local FLDB here */
4381         if (osd_seq_exists(env, osd, fid_seq(fid)))
4382                 RETURN(0);
4383
4384         RETURN(1);
4385 }
4386
4387 /**
4388  * Index delete function for interoperability mode (b11826).
4389  * It will remove the directory entry added by osd_index_ea_insert().
4390  * This entry is needed to maintain name->fid mapping.
4391  *
4392  * \param key,  key i.e. file entry to be deleted
4393  *
4394  * \retval   0, on success
4395  * \retval -ve, on error
4396  */
4397 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4398                                const struct dt_key *key, struct thandle *handle)
4399 {
4400         struct osd_object          *obj = osd_dt_obj(dt);
4401         struct inode               *dir = obj->oo_inode;
4402         struct dentry              *dentry;
4403         struct osd_thandle         *oh;
4404         struct ldiskfs_dir_entry_2 *de = NULL;
4405         struct buffer_head         *bh;
4406         struct htree_lock          *hlock = NULL;
4407         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
4408         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4409         int                        rc;
4410         ENTRY;
4411
4412         if (!dt_object_exists(dt))
4413                 RETURN(-ENOENT);
4414
4415         LINVRNT(osd_invariant(obj));
4416         LASSERT(!dt_object_remote(dt));
4417         LASSERT(handle != NULL);
4418
4419         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4420
4421         oh = container_of(handle, struct osd_thandle, ot_super);
4422         LASSERT(oh->ot_handle != NULL);
4423         LASSERT(oh->ot_handle->h_transaction != NULL);
4424
4425         ll_vfs_dq_init(dir);
4426         dentry = osd_child_dentry_get(env, obj,
4427                                       (char *)key, strlen((char *)key));
4428
4429         if (obj->oo_hl_head != NULL) {
4430                 hlock = osd_oti_get(env)->oti_hlock;
4431                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4432                                    dir, LDISKFS_HLOCK_DEL);
4433         } else {
4434                 down_write(&obj->oo_ext_idx_sem);
4435         }
4436
4437         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4438         if (!IS_ERR(bh)) {
4439                 /* If this is not the ".." entry, it might be a remote DNE
4440                  * entry and  we need to check if the FID is for a remote
4441                  * MDT.  If the FID is  not in the directory entry (e.g.
4442                  * upgraded 1.8 filesystem without dirdata enabled) then
4443                  * we need to get the FID from the LMA. For a remote directory
4444                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4445                  * case.
4446                  *
4447                  * Delete the entry before the agent inode in order to
4448                  * simplify error handling.  At worst an error after deleting
4449                  * the entry first might leak the agent inode afterward. The
4450                  * reverse would need filesystem abort in case of error deleting
4451                  * the entry after the agent had been removed, or leave a
4452                  * dangling entry pointing at a random inode. */
4453                 if (strcmp((char *)key, dotdot) != 0) {
4454                         LASSERT(de != NULL);
4455                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4456                         if (rc == -ENODATA) {
4457                                 /* can't get FID, postpone to the end of the
4458                                  * transaction when iget() is safe */
4459                                 osd_schedule_agent_inode_removal(env, oh,
4460                                                 le32_to_cpu(de->inode));
4461                         } else if (rc == 0 &&
4462                                    unlikely(osd_remote_fid(env, osd, fid))) {
4463                                 osd_schedule_agent_inode_removal(env, oh,
4464                                                 le32_to_cpu(de->inode));
4465                         }
4466                 }
4467                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4468                 brelse(bh);
4469         } else {
4470                 rc = PTR_ERR(bh);
4471         }
4472         if (hlock != NULL)
4473                 ldiskfs_htree_unlock(hlock);
4474         else
4475                 up_write(&obj->oo_ext_idx_sem);
4476
4477         if (rc != 0)
4478                 GOTO(out, rc);
4479
4480         /* For inode on the remote MDT, .. will point to
4481          * /Agent directory, Check whether it needs to delete
4482          * from agent directory */
4483         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4484                 int ret;
4485
4486                 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4487                                                     obj, oh);
4488                 if (ret != 0)
4489                         /* Sigh, the entry has been deleted, and
4490                          * it is not easy to revert it back, so
4491                          * let's keep this error private, and let
4492                          * LFSCK fix it. XXX */
4493                         CERROR("%s: delete remote parent "DFID": rc = %d\n",
4494                                osd_name(osd), PFID(fid), ret);
4495         }
4496 out:
4497         LASSERT(osd_invariant(obj));
4498         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4499         RETURN(rc);
4500 }
4501
4502 /**
4503  *      Lookup index for \a key and copy record to \a rec.
4504  *
4505  *      \param  dt      osd index object
4506  *      \param  key     key for index
4507  *      \param  rec     record reference
4508  *
4509  *      \retval  +ve  success : exact mach
4510  *      \retval  0    return record with key not greater than \a key
4511  *      \retval -ve   failure
4512  */
4513 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4514                                 struct dt_rec *rec, const struct dt_key *key)
4515 {
4516         struct osd_object      *obj = osd_dt_obj(dt);
4517         struct iam_path_descr  *ipd;
4518         struct iam_container   *bag = &obj->oo_dir->od_container;
4519         struct osd_thread_info *oti = osd_oti_get(env);
4520         struct iam_iterator    *it = &oti->oti_idx_it;
4521         struct iam_rec         *iam_rec;
4522         int                     rc;
4523         ENTRY;
4524
4525         if (!dt_object_exists(dt))
4526                 RETURN(-ENOENT);
4527
4528         LASSERT(osd_invariant(obj));
4529         LASSERT(!dt_object_remote(dt));
4530         LASSERT(bag->ic_object == obj->oo_inode);
4531
4532         ipd = osd_idx_ipd_get(env, bag);
4533         if (IS_ERR(ipd))
4534                 RETURN(-ENOMEM);
4535
4536         /* got ipd now we can start iterator. */
4537         iam_it_init(it, bag, 0, ipd);
4538
4539         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4540                 /* swab quota uid/gid provided by caller */
4541                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4542                 key = (const struct dt_key *)&oti->oti_quota_id;
4543         }
4544
4545         rc = iam_it_get(it, (struct iam_key *)key);
4546         if (rc >= 0) {
4547                 if (S_ISDIR(obj->oo_inode->i_mode))
4548                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4549                 else
4550                         iam_rec = (struct iam_rec *) rec;
4551
4552                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4553
4554                 if (S_ISDIR(obj->oo_inode->i_mode))
4555                         osd_fid_unpack((struct lu_fid *) rec,
4556                                        (struct osd_fid_pack *)iam_rec);
4557                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4558                         osd_quota_unpack(obj, rec);
4559         }
4560
4561         iam_it_put(it);
4562         iam_it_fini(it);
4563         osd_ipd_put(env, bag, ipd);
4564
4565         LINVRNT(osd_invariant(obj));
4566
4567         RETURN(rc);
4568 }
4569
4570 static int osd_index_declare_iam_insert(const struct lu_env *env,
4571                                         struct dt_object *dt,
4572                                         const struct dt_rec *rec,
4573                                         const struct dt_key *key,
4574                                         struct thandle *handle)
4575 {
4576         struct osd_thandle *oh;
4577
4578         LASSERT(handle != NULL);
4579
4580         oh = container_of0(handle, struct osd_thandle, ot_super);
4581         LASSERT(oh->ot_handle == NULL);
4582
4583         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4584                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4585
4586         return 0;
4587 }
4588
4589 /**
4590  *      Inserts (key, value) pair in \a dt index object.
4591  *
4592  *      \param  dt      osd index object
4593  *      \param  key     key for index
4594  *      \param  rec     record reference
4595  *      \param  th      transaction handler
4596  *
4597  *      \retval  0  success
4598  *      \retval -ve failure
4599  */
4600 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4601                                 const struct dt_rec *rec,
4602                                 const struct dt_key *key, struct thandle *th,
4603                                 int ignore_quota)
4604 {
4605         struct osd_object     *obj = osd_dt_obj(dt);
4606         struct iam_path_descr *ipd;
4607         struct osd_thandle    *oh;
4608         struct iam_container  *bag;
4609         struct osd_thread_info *oti = osd_oti_get(env);
4610         struct iam_rec         *iam_rec;
4611         int                     rc;
4612         ENTRY;
4613
4614         if (!dt_object_exists(dt))
4615                 RETURN(-ENOENT);
4616
4617         LINVRNT(osd_invariant(obj));
4618         LASSERT(!dt_object_remote(dt));
4619
4620         bag = &obj->oo_dir->od_container;
4621         LASSERT(bag->ic_object == obj->oo_inode);
4622         LASSERT(th != NULL);
4623
4624         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4625
4626         ipd = osd_idx_ipd_get(env, bag);
4627         if (unlikely(ipd == NULL))
4628                 RETURN(-ENOMEM);
4629
4630         oh = container_of0(th, struct osd_thandle, ot_super);
4631         LASSERT(oh->ot_handle != NULL);
4632         LASSERT(oh->ot_handle->h_transaction != NULL);
4633         if (S_ISDIR(obj->oo_inode->i_mode)) {
4634                 iam_rec = (struct iam_rec *)oti->oti_ldp;
4635                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4636         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4637                 /* pack quota uid/gid */
4638                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4639                 key = (const struct dt_key *)&oti->oti_quota_id;
4640                 /* pack quota record */
4641                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4642                 iam_rec = (struct iam_rec *)rec;
4643         } else {
4644                 iam_rec = (struct iam_rec *)rec;
4645         }
4646
4647         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4648                         iam_rec, ipd);
4649         osd_ipd_put(env, bag, ipd);
4650         LINVRNT(osd_invariant(obj));
4651         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4652         RETURN(rc);
4653 }
4654
4655 /**
4656  * Calls ldiskfs_add_entry() to add directory entry
4657  * into the directory. This is required for
4658  * interoperability mode (b11826)
4659  *
4660  * \retval   0, on success
4661  * \retval -ve, on error
4662  */
4663 static int __osd_ea_add_rec(struct osd_thread_info *info,
4664                             struct osd_object *pobj, struct inode  *cinode,
4665                             const char *name, const struct lu_fid *fid,
4666                             struct htree_lock *hlock, struct thandle *th)
4667 {
4668         struct ldiskfs_dentry_param *ldp;
4669         struct dentry               *child;
4670         struct osd_thandle          *oth;
4671         int                          rc;
4672
4673         oth = container_of(th, struct osd_thandle, ot_super);
4674         LASSERT(oth->ot_handle != NULL);
4675         LASSERT(oth->ot_handle->h_transaction != NULL);
4676         LASSERT(pobj->oo_inode);
4677
4678         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4679         if (unlikely(pobj->oo_inode ==
4680                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4681                 ldp->edp_magic = 0;
4682         else
4683                 osd_get_ldiskfs_dirent_param(ldp, fid);
4684         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4685         child->d_fsdata = (void *)ldp;
4686         ll_vfs_dq_init(pobj->oo_inode);
4687         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4688                                    child, cinode, hlock);
4689         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4690                 struct ldiskfs_dir_entry_2      *de;
4691                 struct buffer_head              *bh;
4692                 int                              rc1;
4693
4694                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4695                                             NULL, hlock);
4696                 if (!IS_ERR(bh)) {
4697                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4698                                                                bh);
4699                         if (rc1 == 0) {
4700                                 if (S_ISDIR(cinode->i_mode))
4701                                         de->file_type = LDISKFS_DIRENT_LUFID |
4702                                                         LDISKFS_FT_REG_FILE;
4703                                 else
4704                                         de->file_type = LDISKFS_DIRENT_LUFID |
4705                                                         LDISKFS_FT_DIR;
4706                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4707                                                               NULL, bh);
4708                         }
4709                         brelse(bh);
4710                 }
4711         }
4712
4713         RETURN(rc);
4714 }
4715
4716 /**
4717  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4718  * into the directory.Also sets flags into osd object to
4719  * indicate dot and dotdot are created. This is required for
4720  * interoperability mode (b11826)
4721  *
4722  * \param dir   directory for dot and dotdot fixup.
4723  * \param obj   child object for linking
4724  *
4725  * \retval   0, on success
4726  * \retval -ve, on error
4727  */
4728 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4729                               struct osd_object *dir,
4730                               struct inode *parent_dir, const char *name,
4731                               const struct lu_fid *dot_fid,
4732                               const struct lu_fid *dot_dot_fid,
4733                               struct thandle *th)
4734 {
4735         struct inode                *inode = dir->oo_inode;
4736         struct osd_thandle          *oth;
4737         int result = 0;
4738
4739         oth = container_of(th, struct osd_thandle, ot_super);
4740         LASSERT(oth->ot_handle->h_transaction != NULL);
4741         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4742
4743         if (strcmp(name, dot) == 0) {
4744                 if (dir->oo_compat_dot_created) {
4745                         result = -EEXIST;
4746                 } else {
4747                         LASSERT(inode->i_ino == parent_dir->i_ino);
4748                         dir->oo_compat_dot_created = 1;
4749                         result = 0;
4750                 }
4751         } else if (strcmp(name, dotdot) == 0) {
4752                 if (!dir->oo_compat_dot_created)
4753                         return -EINVAL;
4754                 /* in case of rename, dotdot is already created */
4755                 if (dir->oo_compat_dotdot_created) {
4756                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4757                                                 dot_dot_fid, NULL, th);
4758                 }
4759
4760                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4761                         struct lu_fid tfid = *dot_dot_fid;
4762
4763                         tfid.f_oid--;
4764                         result = osd_add_dot_dotdot_internal(info,
4765                                         dir->oo_inode, parent_dir, dot_fid,
4766                                         &tfid, oth);
4767                 } else {
4768                         result = osd_add_dot_dotdot_internal(info,
4769                                         dir->oo_inode, parent_dir, dot_fid,
4770                                         dot_dot_fid, oth);
4771                 }
4772
4773                 if (result == 0)
4774                         dir->oo_compat_dotdot_created = 1;
4775         }
4776
4777         return result;
4778 }
4779
4780
4781 /**
4782  * It will call the appropriate osd_add* function and return the
4783  * value, return by respective functions.
4784  */
4785 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4786                           struct inode *cinode, const char *name,
4787                           const struct lu_fid *fid, struct thandle *th)
4788 {
4789         struct osd_thread_info *info   = osd_oti_get(env);
4790         struct htree_lock      *hlock;
4791         int                     rc;
4792
4793         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4794
4795         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4796                                                    name[2] =='\0'))) {
4797                 if (hlock != NULL) {
4798                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4799                                            pobj->oo_inode, 0);
4800                 } else {
4801                         down_write(&pobj->oo_ext_idx_sem);
4802                 }
4803
4804                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4805                                         lu_object_fid(&pobj->oo_dt.do_lu),
4806                                         fid, th);
4807         } else {
4808                 if (hlock != NULL) {
4809                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4810                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4811                 } else {
4812                         down_write(&pobj->oo_ext_idx_sem);
4813                 }
4814
4815                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4816                         struct lu_fid *tfid = &info->oti_fid;
4817
4818                         *tfid = *fid;
4819                         tfid->f_ver = ~0;
4820                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4821                                               tfid, hlock, th);
4822                 } else {
4823                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4824                                               hlock, th);
4825                 }
4826         }
4827         if (hlock != NULL)
4828                 ldiskfs_htree_unlock(hlock);
4829         else
4830                 up_write(&pobj->oo_ext_idx_sem);
4831
4832         return rc;
4833 }
4834
4835 static int
4836 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4837                       struct osd_idmap_cache *oic)
4838 {
4839         struct osd_scrub *scrub = &dev->od_scrub;
4840         struct lu_fid *fid = &oic->oic_fid;
4841         struct osd_inode_id *id = &oic->oic_lid;
4842         struct inode *inode = NULL;
4843         int once  = 0;
4844         bool insert;
4845         int rc;
4846         ENTRY;
4847
4848         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4849                 RETURN(0);
4850
4851         if (scrub->os_pos_current > id->oii_ino)
4852                 RETURN(0);
4853
4854 again:
4855         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4856         if (rc == -ENOENT) {
4857                 __u32 gen = id->oii_gen;
4858
4859                 insert = true;
4860                 if (inode != NULL)
4861                         goto trigger;
4862
4863                 inode = osd_iget(oti, dev, id);
4864                 /* The inode has been removed (by race maybe). */
4865                 if (IS_ERR(inode)) {
4866                         rc = PTR_ERR(inode);
4867
4868                         RETURN(rc == -ESTALE ? -ENOENT : rc);
4869                 }
4870
4871                 /* The OI mapping is lost. */
4872                 if (gen != OSD_OII_NOGEN)
4873                         goto trigger;
4874
4875                 iput(inode);
4876                 /* The inode may has been reused by others, we do not know,
4877                  * leave it to be handled by subsequent osd_fid_lookup(). */
4878                 RETURN(0);
4879         } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
4880                 RETURN(rc);
4881         } else {
4882                 insert = false;
4883         }
4884
4885 trigger:
4886         if (thread_is_running(&scrub->os_thread)) {
4887                 if (inode == NULL) {
4888                         inode = osd_iget(oti, dev, id);
4889                         /* The inode has been removed (by race maybe). */
4890                         if (IS_ERR(inode)) {
4891                                 rc = PTR_ERR(inode);
4892
4893                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
4894                         }
4895                 }
4896
4897                 rc = osd_oii_insert(dev, oic, insert);
4898                 /* There is race condition between osd_oi_lookup and OI scrub.
4899                  * The OI scrub finished just after osd_oi_lookup() failure.
4900                  * Under such case, it is unnecessary to trigger OI scrub again,
4901                  * but try to call osd_oi_lookup() again. */
4902                 if (unlikely(rc == -EAGAIN))
4903                         goto again;
4904
4905                 if (!S_ISDIR(inode->i_mode))
4906                         rc = 0;
4907                 else
4908                         rc = osd_check_lmv(oti, dev, inode, oic);
4909
4910                 iput(inode);
4911                 RETURN(rc);
4912         }
4913
4914         if (!dev->od_noscrub && ++once == 1) {
4915                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4916                                      SS_CLEAR_FAILOUT);
4917                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
4918                        "%.16s: trigger partial OI scrub for RPC inconsistency "
4919                        "checking FID "DFID": rc = %d\n",
4920                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4921                        PFID(fid), rc);
4922                 if (rc == 0 || rc == -EALREADY)
4923                         goto again;
4924         }
4925
4926         if (inode != NULL)
4927                 iput(inode);
4928
4929         RETURN(rc);
4930 }
4931
4932 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4933                                struct osd_device *dev,
4934                                struct osd_idmap_cache *oic,
4935                                struct lu_fid *fid, __u32 ino)
4936 {
4937         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4938         struct inode            *inode;
4939         int                      rc;
4940
4941         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4942         inode = osd_iget(oti, dev, &oic->oic_lid);
4943         if (IS_ERR(inode)) {
4944                 fid_zero(&oic->oic_fid);
4945                 return PTR_ERR(inode);
4946         }
4947
4948         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4949         iput(inode);
4950         if (rc != 0)
4951                 fid_zero(&oic->oic_fid);
4952         else
4953                 *fid = oic->oic_fid = lma->lma_self_fid;
4954         return rc;
4955 }
4956
4957 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4958                       struct osd_inode_id *id, const struct lu_fid *fid)
4959 {
4960         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4961                id->oii_ino, id->oii_gen, info);
4962         info->oti_cache.oic_lid = *id;
4963         info->oti_cache.oic_fid = *fid;
4964         info->oti_cache.oic_dev = osd;
4965 }
4966
4967 /**
4968  * Get parent FID from the linkEA.
4969  *
4970  * For a directory which parent resides on remote MDT, to satisfy the
4971  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4972  * the other hand, to make the lookup(..) on the directory can return
4973  * the real parent FID, we append the real parent FID after its ".."
4974  * name entry in the /REMOTE_PARENT_DIR.
4975  *
4976  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4977  * backup. So after the restore, we cannot get the right parent FID from
4978  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4979  * we have stored the real parent FID in the directory object's linkEA,
4980  * we can parse the linkEA for the real parent FID.
4981  *
4982  * \param[in] env       pointer to the thread context
4983  * \param[in] obj       pointer to the object to be handled
4984  * \param[out]fid       pointer to the buffer to hold the parent FID
4985  *
4986  * \retval              0 for getting the real parent FID successfully
4987  * \retval              negative error number on failure
4988  */
4989 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4990                                     struct osd_object *obj,
4991                                     struct lu_fid *fid)
4992 {
4993         struct osd_thread_info  *oti    = osd_oti_get(env);
4994         struct lu_buf           *buf    = &oti->oti_big_buf;
4995         struct dentry           *dentry = &oti->oti_obj_dentry;
4996         struct inode            *inode  = obj->oo_inode;
4997         struct linkea_data       ldata  = { NULL };
4998         int                      rc;
4999         ENTRY;
5000
5001         fid_zero(fid);
5002         if (!S_ISDIR(inode->i_mode))
5003                 RETURN(-EIO);
5004
5005 again:
5006         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5007                              buf->lb_buf, buf->lb_len);
5008         if (rc == -ERANGE) {
5009                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5010                                      NULL, 0);
5011                 if (rc > 0) {
5012                         lu_buf_realloc(buf, rc);
5013                         if (buf->lb_buf == NULL)
5014                                 RETURN(-ENOMEM);
5015
5016                         goto again;
5017                 }
5018         }
5019
5020         if (unlikely(rc == 0))
5021                 RETURN(-ENODATA);
5022
5023         if (rc < 0)
5024                 RETURN(rc);
5025
5026         if (unlikely(buf->lb_buf == NULL)) {
5027                 lu_buf_realloc(buf, rc);
5028                 if (buf->lb_buf == NULL)
5029                         RETURN(-ENOMEM);
5030
5031                 goto again;
5032         }
5033
5034         ldata.ld_buf = buf;
5035         rc = linkea_init(&ldata);
5036         if (rc == 0) {
5037                 linkea_first_entry(&ldata);
5038                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5039         }
5040
5041         RETURN(rc);
5042 }
5043
5044 /**
5045  * Calls ->lookup() to find dentry. From dentry get inode and
5046  * read inode's ea to get fid. This is required for  interoperability
5047  * mode (b11826)
5048  *
5049  * \retval   0, on success
5050  * \retval -ve, on error
5051  */
5052 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5053                              struct dt_rec *rec, const struct dt_key *key)
5054 {
5055         struct inode                    *dir    = obj->oo_inode;
5056         struct dentry                   *dentry;
5057         struct ldiskfs_dir_entry_2      *de;
5058         struct buffer_head              *bh;
5059         struct lu_fid                   *fid = (struct lu_fid *) rec;
5060         struct htree_lock               *hlock = NULL;
5061         int                             ino;
5062         int                             rc;
5063         ENTRY;
5064
5065         LASSERT(dir->i_op != NULL);
5066         LASSERT(dir->i_op->lookup != NULL);
5067
5068         dentry = osd_child_dentry_get(env, obj,
5069                                       (char *)key, strlen((char *)key));
5070
5071         if (obj->oo_hl_head != NULL) {
5072                 hlock = osd_oti_get(env)->oti_hlock;
5073                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5074                                    dir, LDISKFS_HLOCK_LOOKUP);
5075         } else {
5076                 down_read(&obj->oo_ext_idx_sem);
5077         }
5078
5079         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5080         if (!IS_ERR(bh)) {
5081                 struct osd_thread_info *oti = osd_oti_get(env);
5082                 struct osd_inode_id *id = &oti->oti_id;
5083                 struct osd_idmap_cache *oic = &oti->oti_cache;
5084                 struct osd_device *dev = osd_obj2dev(obj);
5085
5086                 ino = le32_to_cpu(de->inode);
5087                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5088                         brelse(bh);
5089                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5090                         GOTO(out, rc);
5091                 }
5092
5093                 rc = osd_get_fid_from_dentry(de, rec);
5094
5095                 /* done with de, release bh */
5096                 brelse(bh);
5097                 if (rc != 0) {
5098                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
5099                                 const char *name = (const char *)key;
5100
5101                                 /* If the parent is on remote MDT, and there
5102                                  * is no FID-in-dirent, then we have to get
5103                                  * the parent FID from the linkEA.  */
5104                                 if (likely(strlen(name) == 2 &&
5105                                            name[0] == '.' && name[1] == '.'))
5106                                         rc = osd_get_pfid_from_linkea(env, obj,
5107                                                                       fid);
5108                         } else {
5109                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5110                         }
5111                 } else {
5112                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5113                 }
5114
5115                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5116                         fid_zero(&oic->oic_fid);
5117
5118                         GOTO(out, rc);
5119                 }
5120
5121                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5122                 rc = osd_consistency_check(oti, dev, oic);
5123                 if (rc != 0)
5124                         fid_zero(&oic->oic_fid);
5125         } else {
5126                 rc = PTR_ERR(bh);
5127         }
5128
5129         GOTO(out, rc);
5130
5131 out:
5132         if (hlock != NULL)
5133                 ldiskfs_htree_unlock(hlock);
5134         else
5135                 up_read(&obj->oo_ext_idx_sem);
5136         return rc;
5137 }
5138
5139 /**
5140  * Put the osd object once done with it.
5141  *
5142  * \param obj osd object that needs to be put
5143  */
5144 static inline void osd_object_put(const struct lu_env *env,
5145                                   struct osd_object *obj)
5146 {
5147         lu_object_put(env, &obj->oo_dt.do_lu);
5148 }
5149
5150 static int osd_index_declare_ea_insert(const struct lu_env *env,
5151                                        struct dt_object *dt,
5152                                        const struct dt_rec *rec,
5153                                        const struct dt_key *key,
5154                                        struct thandle *handle)
5155 {
5156         struct osd_thandle      *oh;
5157         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5158         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5159         const struct lu_fid     *fid    = rec1->rec_fid;
5160         int                      credits, rc = 0;
5161         struct osd_idmap_cache  *idc;
5162         ENTRY;
5163
5164         LASSERT(!dt_object_remote(dt));
5165         LASSERT(handle != NULL);
5166         LASSERT(fid != NULL);
5167         LASSERT(rec1->rec_type != 0);
5168
5169         oh = container_of0(handle, struct osd_thandle, ot_super);
5170         LASSERT(oh->ot_handle == NULL);
5171
5172         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5173
5174         /* we can't call iget() while a transactions is running
5175          * (this can lead to a deadlock), but we need to know
5176          * inum and object type. so we find this information at
5177          * declaration and cache in per-thread info */
5178         idc = osd_idc_find_or_init(env, osd, fid);
5179         if (IS_ERR(idc))
5180                 RETURN(PTR_ERR(idc));
5181         if (idc->oic_remote) {
5182                 /* a reference to remote inode is represented by an
5183                  * agent inode which we have to create */
5184                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5185                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5186         }
5187
5188         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5189
5190         if (osd_dt_obj(dt)->oo_inode != NULL) {
5191                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5192
5193                 /* We ignore block quota on meta pool (MDTs), so needn't
5194                  * calculate how many blocks will be consumed by this index
5195                  * insert */
5196                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5197                                            i_gid_read(inode), 0, oh,
5198                                            osd_dt_obj(dt), true, NULL, false);
5199         }
5200
5201         RETURN(rc);
5202 }
5203
5204 /**
5205  * Index add function for interoperability mode (b11826).
5206  * It will add the directory entry.This entry is needed to
5207  * maintain name->fid mapping.
5208  *
5209  * \param key it is key i.e. file entry to be inserted
5210  * \param rec it is value of given key i.e. fid
5211  *
5212  * \retval   0, on success
5213  * \retval -ve, on error
5214  */
5215 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5216                                const struct dt_rec *rec,
5217                                const struct dt_key *key, struct thandle *th,
5218                                int ignore_quota)
5219 {
5220         struct osd_object       *obj = osd_dt_obj(dt);
5221         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5222         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5223         const struct lu_fid     *fid    = rec1->rec_fid;
5224         const char              *name = (const char *)key;
5225         struct osd_thread_info  *oti   = osd_oti_get(env);
5226         struct inode            *child_inode = NULL;
5227         struct osd_idmap_cache  *idc;
5228         int                     rc;
5229         ENTRY;
5230
5231         if (!dt_object_exists(dt))
5232                 RETURN(-ENOENT);
5233
5234         LASSERT(osd_invariant(obj));
5235         LASSERT(!dt_object_remote(dt));
5236         LASSERT(th != NULL);
5237
5238         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5239
5240         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5241
5242         idc = osd_idc_find(env, osd, fid);
5243         if (unlikely(idc == NULL)) {
5244                 /* this dt_insert() wasn't declared properly, so
5245                  * FID is missing in OI cache. we better do not
5246                  * lookup FID in FLDB/OI and don't risk to deadlock,
5247                  * but in some special cases (lfsck testing, etc)
5248                  * it's much simpler than fixing a caller */
5249                 CERROR("%s: "DFID" wasn't declared for insert\n",
5250                        osd_name(osd), PFID(fid));
5251                 dump_stack();
5252                 idc = osd_idc_find_or_init(env, osd, fid);
5253                 if (IS_ERR(idc))
5254                         RETURN(PTR_ERR(idc));
5255         }
5256
5257         if (idc->oic_remote) {
5258                 /* Insert remote entry */
5259                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5260                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
5261                         struct osd_thandle      *oh;
5262
5263                         /* If parent on remote MDT, we need put this object
5264                          * under AGENT */
5265                         oh = container_of(th, typeof(*oh), ot_super);
5266                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
5267                         if (rc != 0) {
5268                                 CERROR("%s: add "DFID" error: rc = %d\n",
5269                                        osd_name(osd),
5270                                        PFID(lu_object_fid(&dt->do_lu)), rc);
5271                                 RETURN(rc);
5272                         }
5273
5274                         child_inode = igrab(omm->omm_remote_parent->d_inode);
5275                 } else {
5276                         child_inode = osd_create_local_agent_inode(env, osd,
5277                                         obj, fid, rec1->rec_type & S_IFMT, th);
5278                         if (IS_ERR(child_inode))
5279                                 RETURN(PTR_ERR(child_inode));
5280                 }
5281         } else {
5282                 /* Insert local entry */
5283                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5284                         /* for a reason OI cache wasn't filled properly */
5285                         CERROR("%s: OIC for "DFID" isn't filled\n",
5286                                osd_name(osd), PFID(fid));
5287                         RETURN(-EINVAL);
5288                 }
5289                 child_inode = oti->oti_inode;
5290                 if (unlikely(child_inode == NULL)) {
5291                         struct ldiskfs_inode_info *lii;
5292                         OBD_ALLOC_PTR(lii);
5293                         if (lii == NULL)
5294                                 RETURN(-ENOMEM);
5295                         child_inode = oti->oti_inode = &lii->vfs_inode;
5296                 }
5297                 child_inode->i_sb = osd_sb(osd);
5298                 child_inode->i_ino = idc->oic_lid.oii_ino;
5299                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5300         }
5301
5302         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5303
5304         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5305                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5306
5307         if (child_inode && child_inode != oti->oti_inode)
5308                 iput(child_inode);
5309         LASSERT(osd_invariant(obj));
5310         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5311         RETURN(rc);
5312 }
5313
5314 /**
5315  *  Initialize osd Iterator for given osd index object.
5316  *
5317  *  \param  dt      osd index object
5318  */
5319
5320 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5321                                      struct dt_object *dt,
5322                                      __u32 unused)
5323 {
5324         struct osd_it_iam      *it;
5325         struct osd_object      *obj = osd_dt_obj(dt);
5326         struct lu_object       *lo  = &dt->do_lu;
5327         struct iam_path_descr  *ipd;
5328         struct iam_container   *bag = &obj->oo_dir->od_container;
5329
5330         if (!dt_object_exists(dt))
5331                 return ERR_PTR(-ENOENT);
5332
5333         OBD_ALLOC_PTR(it);
5334         if (it == NULL)
5335                 return ERR_PTR(-ENOMEM);
5336
5337         ipd = osd_it_ipd_get(env, bag);
5338         if (likely(ipd != NULL)) {
5339                 it->oi_obj = obj;
5340                 it->oi_ipd = ipd;
5341                 lu_object_get(lo);
5342                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5343                 return (struct dt_it *)it;
5344         } else {
5345                 OBD_FREE_PTR(it);
5346                 return ERR_PTR(-ENOMEM);
5347         }
5348 }
5349
5350 /**
5351  * free given Iterator.
5352  */
5353
5354 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5355 {
5356         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5357         struct osd_object       *obj = it->oi_obj;
5358
5359         iam_it_fini(&it->oi_it);
5360         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5361         lu_object_put(env, &obj->oo_dt.do_lu);
5362         OBD_FREE_PTR(it);
5363 }
5364
5365 /**
5366  *  Move Iterator to record specified by \a key
5367  *
5368  *  \param  di      osd iterator
5369  *  \param  key     key for index
5370  *
5371  *  \retval +ve  di points to record with least key not larger than key
5372  *  \retval  0   di points to exact matched key
5373  *  \retval -ve  failure
5374  */
5375
5376 static int osd_it_iam_get(const struct lu_env *env,
5377                           struct dt_it *di, const struct dt_key *key)
5378 {
5379         struct osd_thread_info  *oti = osd_oti_get(env);
5380         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5381
5382         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5383                 /* swab quota uid/gid */
5384                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5385                 key = (struct dt_key *)&oti->oti_quota_id;
5386         }
5387
5388         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5389 }
5390
5391 /**
5392  *  Release Iterator
5393  *
5394  *  \param  di      osd iterator
5395  */
5396 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5397 {
5398         struct osd_it_iam *it = (struct osd_it_iam *)di;
5399
5400         iam_it_put(&it->oi_it);
5401 }
5402
5403 /**
5404  *  Move iterator by one record
5405  *
5406  *  \param  di      osd iterator
5407  *
5408  *  \retval +1   end of container reached
5409  *  \retval  0   success
5410  *  \retval -ve  failure
5411  */
5412
5413 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5414 {
5415         struct osd_it_iam *it = (struct osd_it_iam *)di;
5416
5417         return iam_it_next(&it->oi_it);
5418 }
5419
5420 /**
5421  * Return pointer to the key under iterator.
5422  */
5423
5424 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5425                                  const struct dt_it *di)
5426 {
5427         struct osd_thread_info *oti = osd_oti_get(env);
5428         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5429         struct osd_object      *obj = it->oi_obj;
5430         struct dt_key          *key;
5431
5432         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5433
5434         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5435                 /* swab quota uid/gid */
5436                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5437                 key = (struct dt_key *)&oti->oti_quota_id;
5438         }
5439
5440         return key;
5441 }
5442
5443 /**
5444  * Return size of key under iterator (in bytes)
5445  */
5446
5447 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5448 {
5449         struct osd_it_iam *it = (struct osd_it_iam *)di;
5450
5451         return iam_it_key_size(&it->oi_it);
5452 }
5453
5454 static inline void
5455 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5456 {
5457         /* check if file type is required */
5458         if (ent->lde_attrs & LUDA_TYPE) {
5459                 struct luda_type *lt;
5460                 int align = sizeof(*lt) - 1;
5461
5462                 len = (len + align) & ~align;
5463                 lt = (struct luda_type *)(ent->lde_name + len);
5464                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5465         }
5466
5467         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5468 }
5469
5470 /**
5471  * build lu direct from backend fs dirent.
5472  */
5473
5474 static inline void
5475 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5476                    char *name, __u16 namelen, __u16 type, __u32 attr)
5477 {
5478         ent->lde_attrs = attr | LUDA_FID;
5479         fid_cpu_to_le(&ent->lde_fid, fid);
5480
5481         ent->lde_hash = cpu_to_le64(offset);
5482         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5483
5484         strncpy(ent->lde_name, name, namelen);
5485         ent->lde_name[namelen] = '\0';
5486         ent->lde_namelen = cpu_to_le16(namelen);
5487
5488         /* append lustre attributes */
5489         osd_it_append_attrs(ent, namelen, type);
5490 }
5491
5492 /**
5493  * Return pointer to the record under iterator.
5494  */
5495 static int osd_it_iam_rec(const struct lu_env *env,
5496                           const struct dt_it *di,
5497                           struct dt_rec *dtrec, __u32 attr)
5498 {
5499         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5500         struct osd_thread_info *info = osd_oti_get(env);
5501         ENTRY;
5502
5503         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5504                 const struct osd_fid_pack *rec;
5505                 struct lu_fid             *fid = &info->oti_fid;
5506                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5507                 char                      *name;
5508                 int                        namelen;
5509                 __u64                      hash;
5510                 int                        rc;
5511
5512                 name = (char *)iam_it_key_get(&it->oi_it);
5513                 if (IS_ERR(name))
5514                         RETURN(PTR_ERR(name));
5515
5516                 namelen = iam_it_key_size(&it->oi_it);
5517
5518                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5519                 if (IS_ERR(rec))
5520                         RETURN(PTR_ERR(rec));
5521
5522                 rc = osd_fid_unpack(fid, rec);
5523                 if (rc)
5524                         RETURN(rc);
5525
5526                 hash = iam_it_store(&it->oi_it);
5527
5528                 /* IAM does not store object type in IAM index (dir) */
5529                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5530                                    0, LUDA_FID);
5531         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5532                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5533                            (struct iam_rec *)dtrec);
5534                 osd_quota_unpack(it->oi_obj, dtrec);
5535         } else {
5536                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5537                            (struct iam_rec *)dtrec);
5538         }
5539
5540         RETURN(0);
5541 }
5542
5543 /**
5544  * Returns cookie for current Iterator position.
5545  */
5546 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5547 {
5548         struct osd_it_iam *it = (struct osd_it_iam *)di;
5549
5550         return iam_it_store(&it->oi_it);
5551 }
5552
5553 /**
5554  * Restore iterator from cookie.
5555  *
5556  * \param  di      osd iterator
5557  * \param  hash    Iterator location cookie
5558  *
5559  * \retval +ve  di points to record with least key not larger than key.
5560  * \retval  0   di points to exact matched key
5561  * \retval -ve  failure
5562  */
5563
5564 static int osd_it_iam_load(const struct lu_env *env,
5565                            const struct dt_it *di, __u64 hash)
5566 {
5567         struct osd_it_iam *it = (struct osd_it_iam *)di;
5568
5569         return iam_it_load(&it->oi_it, hash);
5570 }
5571
5572 static const struct dt_index_operations osd_index_iam_ops = {
5573         .dio_lookup         = osd_index_iam_lookup,
5574         .dio_declare_insert = osd_index_declare_iam_insert,
5575         .dio_insert         = osd_index_iam_insert,
5576         .dio_declare_delete = osd_index_declare_iam_delete,
5577         .dio_delete         = osd_index_iam_delete,
5578         .dio_it     = {
5579                 .init     = osd_it_iam_init,
5580                 .fini     = osd_it_iam_fini,
5581                 .get      = osd_it_iam_get,
5582                 .put      = osd_it_iam_put,
5583                 .next     = osd_it_iam_next,
5584                 .key      = osd_it_iam_key,
5585                 .key_size = osd_it_iam_key_size,
5586                 .rec      = osd_it_iam_rec,
5587                 .store    = osd_it_iam_store,
5588                 .load     = osd_it_iam_load
5589         }
5590 };
5591
5592
5593 /**
5594  * Creates or initializes iterator context.
5595  *
5596  * \retval struct osd_it_ea, iterator structure on success
5597  *
5598  */
5599 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5600                                     struct dt_object *dt,
5601                                     __u32 attr)
5602 {
5603         struct osd_object       *obj  = osd_dt_obj(dt);
5604         struct osd_thread_info  *info = osd_oti_get(env);
5605         struct osd_it_ea        *oie;
5606         struct file             *file;
5607         struct lu_object        *lo   = &dt->do_lu;
5608         struct dentry           *obj_dentry;
5609         ENTRY;
5610
5611         if (!dt_object_exists(dt) || obj->oo_destroyed)
5612                 RETURN(ERR_PTR(-ENOENT));
5613
5614         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5615         if (oie == NULL)
5616                 RETURN(ERR_PTR(-ENOMEM));
5617         obj_dentry = &oie->oie_dentry;
5618
5619         obj_dentry->d_inode = obj->oo_inode;
5620         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5621         obj_dentry->d_name.hash = 0;
5622
5623         oie->oie_rd_dirent       = 0;
5624         oie->oie_it_dirent       = 0;
5625         oie->oie_dirent          = NULL;
5626         if (unlikely(!info->oti_it_ea_buf_used)) {
5627                 oie->oie_buf = info->oti_it_ea_buf;
5628                 info->oti_it_ea_buf_used = 1;
5629         } else {
5630                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5631                 if (oie->oie_buf == NULL)
5632                         RETURN(ERR_PTR(-ENOMEM));
5633         }
5634         oie->oie_obj             = obj;
5635
5636         file = &oie->oie_file;
5637
5638         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5639         if (attr & LUDA_64BITHASH)
5640                 file->f_mode    = FMODE_64BITHASH;
5641         else
5642                 file->f_mode    = FMODE_32BITHASH;
5643         file->f_path.dentry     = obj_dentry;
5644         file->f_mapping         = obj->oo_inode->i_mapping;
5645         file->f_op              = obj->oo_inode->i_fop;
5646         set_file_inode(file, obj->oo_inode);
5647
5648         lu_object_get(lo);
5649         RETURN((struct dt_it *) oie);
5650 }
5651
5652 /**
5653  * Destroy or finishes iterator context.
5654  *
5655  * \param di iterator structure to be destroyed
5656  */
5657 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5658 {
5659         struct osd_thread_info  *info = osd_oti_get(env);
5660         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
5661         struct osd_object       *obj    = oie->oie_obj;
5662         struct inode            *inode  = obj->oo_inode;
5663
5664         ENTRY;
5665         oie->oie_file.f_op->release(inode, &oie->oie_file);
5666         lu_object_put(env, &obj->oo_dt.do_lu);
5667         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5668                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5669         else
5670                 info->oti_it_ea_buf_used = 0;
5671         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5672         EXIT;
5673 }
5674
5675 /**
5676  * It position the iterator at given key, so that next lookup continues from
5677  * that key Or it is similar to dio_it->load() but based on a key,
5678  * rather than file position.
5679  *
5680  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5681  * to the beginning.
5682  *
5683  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5684  */
5685 static int osd_it_ea_get(const struct lu_env *env,
5686                          struct dt_it *di, const struct dt_key *key)
5687 {
5688         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
5689
5690         ENTRY;
5691         LASSERT(((const char *)key)[0] == '\0');
5692         it->oie_file.f_pos      = 0;
5693         it->oie_rd_dirent       = 0;
5694         it->oie_it_dirent       = 0;
5695         it->oie_dirent          = NULL;
5696
5697         RETURN(+1);
5698 }
5699
5700 /**
5701  * Does nothing
5702  */
5703 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5704 {
5705 }
5706
5707 struct osd_filldir_cbs {
5708 #ifdef HAVE_DIR_CONTEXT
5709         struct dir_context ctx;
5710 #endif
5711         struct osd_it_ea  *it;
5712 };
5713 /**
5714  * It is called internally by ->readdir(). It fills the
5715  * iterator's in-memory data structure with required
5716  * information i.e. name, namelen, rec_size etc.
5717  *
5718  * \param buf in which information to be filled in.
5719  * \param name name of the file in given dir
5720  *
5721  * \retval 0 on success
5722  * \retval 1 on buffer full
5723  */
5724 #ifdef HAVE_FILLDIR_USE_CTX
5725 static int osd_ldiskfs_filldir(struct dir_context *buf,
5726 #else
5727 static int osd_ldiskfs_filldir(void *buf,
5728 #endif
5729                                const char *name, int namelen,
5730                                loff_t offset, __u64 ino, unsigned d_type)
5731 {
5732         struct osd_it_ea        *it   =
5733                 ((struct osd_filldir_cbs *)buf)->it;
5734         struct osd_object       *obj  = it->oie_obj;
5735         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5736         struct lu_fid           *fid  = &ent->oied_fid;
5737         struct osd_fid_pack     *rec;
5738         ENTRY;
5739
5740         /* this should never happen */
5741         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5742                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5743                 RETURN(-EIO);
5744         }
5745
5746         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5747             OSD_IT_EA_BUFSIZE)
5748                 RETURN(1);
5749
5750         /* "." is just the object itself. */
5751         if (namelen == 1 && name[0] == '.') {
5752                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5753         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5754                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5755                 if (osd_fid_unpack(fid, rec) != 0)
5756                         fid_zero(fid);
5757         } else {
5758                 fid_zero(fid);
5759         }
5760         d_type &= ~LDISKFS_DIRENT_LUFID;
5761
5762         /* NOT export local root. */
5763         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5764                 ino = obj->oo_inode->i_ino;
5765                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5766         }
5767
5768         ent->oied_ino     = ino;
5769         ent->oied_off     = offset;
5770         ent->oied_namelen = namelen;
5771         ent->oied_type    = d_type;
5772
5773         memcpy(ent->oied_name, name, namelen);
5774
5775         it->oie_rd_dirent++;
5776         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5777         RETURN(0);
5778 }
5779
5780 /**
5781  * Calls ->readdir() to load a directory entry at a time
5782  * and stored it in iterator's in-memory data structure.
5783  *
5784  * \param di iterator's in memory structure
5785  *
5786  * \retval   0 on success
5787  * \retval -ve on error
5788  * \retval +1 reach the end of entry
5789  */
5790 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5791                                const struct dt_it *di)
5792 {
5793         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5794         struct osd_object  *obj   = it->oie_obj;
5795         struct inode       *inode = obj->oo_inode;
5796         struct htree_lock  *hlock = NULL;
5797         struct file        *filp  = &it->oie_file;
5798         int                 rc = 0;
5799         struct osd_filldir_cbs buf = {
5800 #ifdef HAVE_DIR_CONTEXT
5801                 .ctx.actor = osd_ldiskfs_filldir,
5802 #endif
5803                 .it = it
5804         };
5805
5806         ENTRY;
5807         it->oie_dirent = it->oie_buf;
5808         it->oie_rd_dirent = 0;
5809
5810         if (obj->oo_hl_head != NULL) {
5811                 hlock = osd_oti_get(env)->oti_hlock;
5812                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5813                                    inode, LDISKFS_HLOCK_READDIR);
5814         } else {
5815                 down_read(&obj->oo_ext_idx_sem);
5816         }
5817
5818 #ifdef HAVE_DIR_CONTEXT
5819         buf.ctx.pos = filp->f_pos;
5820         rc = inode->i_fop->iterate(filp, &buf.ctx);
5821         filp->f_pos = buf.ctx.pos;
5822 #else
5823         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5824 #endif
5825
5826         if (hlock != NULL)
5827                 ldiskfs_htree_unlock(hlock);
5828         else
5829                 up_read(&obj->oo_ext_idx_sem);
5830
5831         if (it->oie_rd_dirent == 0) {
5832                 /*If it does not get any dirent, it means it has been reached
5833                  *to the end of the dir */
5834                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5835                 if (rc == 0)
5836                         rc = 1;
5837         } else {
5838                 it->oie_dirent = it->oie_buf;
5839                 it->oie_it_dirent = 1;
5840         }
5841
5842         RETURN(rc);
5843 }
5844
5845 /**
5846  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5847  * to load a directory entry at a time and stored it in
5848  * iterator's in-memory data structure.
5849  *
5850  * \param di iterator's in memory structure
5851  *
5852  * \retval +ve iterator reached to end
5853  * \retval   0 iterator not reached to end
5854  * \retval -ve on error
5855  */
5856 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5857 {
5858         struct osd_it_ea *it = (struct osd_it_ea *)di;
5859         int rc;
5860
5861         ENTRY;
5862
5863         if (it->oie_it_dirent < it->oie_rd_dirent) {
5864                 it->oie_dirent =
5865                         (void *) it->oie_dirent +
5866                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5867                                        it->oie_dirent->oied_namelen);
5868                 it->oie_it_dirent++;
5869                 RETURN(0);
5870         } else {
5871                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5872                         rc = +1;
5873                 else
5874                         rc = osd_ldiskfs_it_fill(env, di);
5875         }
5876
5877         RETURN(rc);
5878 }
5879
5880 /**
5881  * Returns the key at current position from iterator's in memory structure.
5882  *
5883  * \param di iterator's in memory structure
5884  *
5885  * \retval key i.e. struct dt_key on success
5886  */
5887 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5888                                     const struct dt_it *di)
5889 {
5890         struct osd_it_ea *it = (struct osd_it_ea *)di;
5891
5892         return (struct dt_key *)it->oie_dirent->oied_name;
5893 }
5894
5895 /**
5896  * Returns the key's size at current position from iterator's in memory structure.
5897  *
5898  * \param di iterator's in memory structure
5899  *
5900  * \retval key_size i.e. struct dt_key on success
5901  */
5902 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5903 {
5904         struct osd_it_ea *it = (struct osd_it_ea *)di;
5905
5906         return it->oie_dirent->oied_namelen;
5907 }
5908
5909 static inline bool
5910 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5911 {
5912         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5913                  "dot_dotdot = %d\n", dot_dotdot);
5914
5915         if (LDISKFS_DIR_REC_LEN(de) >=
5916             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5917                 return true;
5918
5919         return false;
5920 }
5921
5922 static inline bool
5923 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5924                      unsigned blocksize, int dot_dotdot)
5925 {
5926         if (dot_dotdot > 0)
5927                 return osd_dot_dotdot_has_space(de, dot_dotdot);
5928
5929         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5930             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5931                 return true;
5932
5933         return false;
5934 }
5935
5936 static int
5937 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
5938                     handle_t *jh, struct dentry *dentry,
5939                     const struct lu_fid *fid, struct buffer_head *bh,
5940                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
5941                     int dot_dotdot)
5942 {
5943         struct inode                *dir        = dentry->d_parent->d_inode;
5944         struct inode                *inode      = dentry->d_inode;
5945         struct osd_fid_pack         *rec;
5946         struct ldiskfs_dentry_param *ldp;
5947         int                          namelen    = dentry->d_name.len;
5948         int                          rc;
5949         struct osd_thread_info     *info        = osd_oti_get(env);
5950         ENTRY;
5951
5952         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5953                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5954                 RETURN(0);
5955
5956         /* There is enough space to hold the FID-in-dirent. */
5957         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize,
5958                                  dot_dotdot)) {
5959                 rc = ldiskfs_journal_get_write_access(jh, bh);
5960                 if (rc != 0)
5961                         RETURN(rc);
5962
5963                 de->name[namelen] = 0;
5964                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
5965                 rec->fp_len = sizeof(struct lu_fid) + 1;
5966                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5967                 de->file_type |= LDISKFS_DIRENT_LUFID;
5968                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5969
5970                 RETURN(rc);
5971         }
5972
5973         LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot);
5974
5975         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5976         if (rc != 0)
5977                 RETURN(rc);
5978
5979         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5980         osd_get_ldiskfs_dirent_param(ldp, fid);
5981         dentry->d_fsdata = (void *)ldp;
5982         ll_vfs_dq_init(dir);
5983         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
5984         /* It is too bad, we cannot reinsert the name entry back.
5985          * That means we lose it! */
5986         if (rc != 0)
5987                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5988                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5989                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5990                        dir->i_ino, dir->i_generation, namelen,
5991                        dentry->d_name.name, PFID(fid), rc);
5992
5993         RETURN(rc);
5994 }
5995
5996 static int
5997 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5998                         struct osd_it_ea *it, struct lu_fid *fid,
5999                         struct osd_inode_id *id, __u32 *attr)
6000 {
6001         struct osd_thread_info     *info        = osd_oti_get(env);
6002         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
6003         struct osd_device          *dev         = osd_obj2dev(obj);
6004         struct super_block         *sb          = osd_sb(dev);
6005         const char                 *devname     =
6006                                         LDISKFS_SB(sb)->s_es->s_volume_name;
6007         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6008         struct inode               *dir         = obj->oo_inode;
6009         struct htree_lock          *hlock       = NULL;
6010         struct buffer_head         *bh          = NULL;
6011         handle_t                   *jh          = NULL;
6012         struct ldiskfs_dir_entry_2 *de;
6013         struct dentry              *dentry;
6014         struct inode               *inode;
6015         int                         credits;
6016         int                         rc;
6017         int                         dot_dotdot  = 0;
6018         bool                        dirty       = false;
6019         ENTRY;
6020
6021         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6022         inode = osd_iget(info, dev, id);
6023         if (IS_ERR(inode)) {
6024                 rc = PTR_ERR(inode);
6025                 if (rc == -ENOENT || rc == -ESTALE) {
6026                         *attr |= LUDA_UNKNOWN;
6027                         rc = 0;
6028                 } else {
6029                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
6030                                "check_repair, dir = %lu/%u, name = %.*s: "
6031                                "rc = %d\n",
6032                                devname, dir->i_ino, dir->i_generation,
6033                                ent->oied_namelen, ent->oied_name, rc);
6034                 }
6035
6036                 RETURN(rc);
6037         }
6038
6039         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6040                                            ent->oied_namelen);
6041         rc = osd_get_lma(info, inode, dentry, lma);
6042         if (rc == -ENODATA)
6043                 lma = NULL;
6044         else if (rc != 0)
6045                 GOTO(out, rc);
6046
6047         if (ent->oied_name[0] == '.') {
6048                 if (ent->oied_namelen == 1)
6049                         dot_dotdot = 1;
6050                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6051                         dot_dotdot = 2;
6052         }
6053
6054         /* We need to ensure that the name entry is still valid.
6055          * Because it may be removed or renamed by other already.
6056          *
6057          * The unlink or rename operation will start journal before PDO lock,
6058          * so to avoid deadlock, here we need to start journal handle before
6059          * related PDO lock also. But because we do not know whether there
6060          * will be something to be repaired before PDO lock, we just start
6061          * journal without conditions.
6062          *
6063          * We may need to remove the name entry firstly, then insert back.
6064          * One credit is for user quota file update.
6065          * One credit is for group quota file update.
6066          * Two credits are for dirty inode. */
6067         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6068                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6069
6070         if (dev->od_dirent_journal != 0) {
6071
6072 again:
6073                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6074                 if (IS_ERR(jh)) {
6075                         rc = PTR_ERR(jh);
6076                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
6077                                "check_repair, dir = %lu/%u, credits = %d, "
6078                                "name = %.*s: rc = %d\n",
6079                                devname, dir->i_ino, dir->i_generation, credits,
6080                                ent->oied_namelen, ent->oied_name, rc);
6081
6082                         GOTO(out_inode, rc);
6083                 }
6084
6085                 if (obj->oo_hl_head != NULL) {
6086                         hlock = osd_oti_get(env)->oti_hlock;
6087                         /* "0" means exclusive lock for the whole directory.
6088                          * We need to prevent others access such name entry
6089                          * during the delete + insert. Neither HLOCK_ADD nor
6090                          * HLOCK_DEL cannot guarantee the atomicity. */
6091                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6092                 } else {
6093                         down_write(&obj->oo_ext_idx_sem);
6094                 }
6095         } else {
6096                 if (obj->oo_hl_head != NULL) {
6097                         hlock = osd_oti_get(env)->oti_hlock;
6098                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6099                                            LDISKFS_HLOCK_LOOKUP);
6100                 } else {
6101                         down_read(&obj->oo_ext_idx_sem);
6102                 }
6103         }
6104
6105         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6106         /* For dot/dotdot entry, if there is not enough space to hold the
6107          * FID-in-dirent, just keep them there. It only happens when the
6108          * device upgraded from 1.8 or restored from MDT file-level backup.
6109          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
6110          * and needs to get FID from LMA when readdir, it will not affect the
6111          * performance much. */
6112         if (IS_ERR(bh) || (le32_to_cpu(de->inode) != inode->i_ino) ||
6113             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
6114                 *attr |= LUDA_IGNORE;
6115
6116                 GOTO(out, rc = 0);
6117         }
6118
6119         if (lma != NULL) {
6120                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6121                         struct lu_fid *tfid = &lma->lma_self_fid;
6122
6123                         *attr |= LUDA_IGNORE;
6124                         /* It must be REMOTE_PARENT_DIR and as the
6125                          * dotdot entry of remote directory */
6126                         if (unlikely(dot_dotdot != 2 ||
6127                                      fid_seq(tfid) != FID_SEQ_LOCAL_FILE ||
6128                                      fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) {
6129                                 CDEBUG(D_LFSCK, "%.16s: expect remote agent "
6130                                        "parent directory, but got %.*s under "
6131                                        "dir = %lu/%u with the FID "DFID"\n",
6132                                        devname, ent->oied_namelen,
6133                                        ent->oied_name, dir->i_ino,
6134                                        dir->i_generation, PFID(tfid));
6135
6136                                 GOTO(out, rc = -EIO);
6137                         }
6138
6139                         GOTO(out, rc = 0);
6140                 }
6141
6142                 if (fid_is_sane(fid)) {
6143                         /* FID-in-dirent is valid. */
6144                         if (lu_fid_eq(fid, &lma->lma_self_fid))
6145                                 GOTO(out, rc = 0);
6146
6147                         /* Do not repair under dryrun mode. */
6148                         if (*attr & LUDA_VERIFY_DRYRUN) {
6149                                 *attr |= LUDA_REPAIR;
6150
6151                                 GOTO(out, rc = 0);
6152                         }
6153
6154                         if (jh == NULL) {
6155                                 brelse(bh);
6156                                 dev->od_dirent_journal = 1;
6157                                 if (hlock != NULL) {
6158                                         ldiskfs_htree_unlock(hlock);
6159                                         hlock = NULL;
6160                                 } else {
6161                                         up_read(&obj->oo_ext_idx_sem);
6162                                 }
6163
6164                                 goto again;
6165                         }
6166
6167                         *fid = lma->lma_self_fid;
6168                         dirty = true;
6169                         /* Update the FID-in-dirent. */
6170                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6171                                                  bh, de, hlock, dot_dotdot);
6172                         if (rc == 0)
6173                                 *attr |= LUDA_REPAIR;
6174                         else
6175                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
6176                                        "in the dirent, dir = %lu/%u, "
6177                                        "name = %.*s, "DFID": rc = %d\n",
6178                                        devname, dir->i_ino, dir->i_generation,
6179                                        ent->oied_namelen, ent->oied_name,
6180                                        PFID(fid), rc);
6181                 } else {
6182                         /* Do not repair under dryrun mode. */
6183                         if (*attr & LUDA_VERIFY_DRYRUN) {
6184                                 *fid = lma->lma_self_fid;
6185                                 *attr |= LUDA_REPAIR;
6186
6187                                 GOTO(out, rc = 0);
6188                         }
6189
6190                         if (jh == NULL) {
6191                                 brelse(bh);
6192                                 dev->od_dirent_journal = 1;
6193                                 if (hlock != NULL) {
6194                                         ldiskfs_htree_unlock(hlock);
6195                                         hlock = NULL;
6196                                 } else {
6197                                         up_read(&obj->oo_ext_idx_sem);
6198                                 }
6199
6200                                 goto again;
6201                         }
6202
6203                         *fid = lma->lma_self_fid;
6204                         dirty = true;
6205                         /* Append the FID-in-dirent. */
6206                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6207                                                  bh, de, hlock, dot_dotdot);
6208                         if (rc == 0)
6209                                 *attr |= LUDA_REPAIR;
6210                         else
6211                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
6212                                        "after the dirent, dir = %lu/%u, "
6213                                        "name = %.*s, "DFID": rc = %d\n",
6214                                        devname, dir->i_ino, dir->i_generation,
6215                                        ent->oied_namelen, ent->oied_name,
6216                                        PFID(fid), rc);
6217                 }
6218         } else {
6219                 /* Do not repair under dryrun mode. */
6220                 if (*attr & LUDA_VERIFY_DRYRUN) {
6221                         if (fid_is_sane(fid)) {
6222                                 *attr |= LUDA_REPAIR;
6223                         } else {
6224                                 lu_igif_build(fid, inode->i_ino,
6225                                               inode->i_generation);
6226                                 *attr |= LUDA_UPGRADE;
6227                         }
6228
6229                         GOTO(out, rc = 0);
6230                 }
6231
6232                 if (jh == NULL) {
6233                         brelse(bh);
6234                         dev->od_dirent_journal = 1;
6235                         if (hlock != NULL) {
6236                                 ldiskfs_htree_unlock(hlock);
6237                                 hlock = NULL;
6238                         } else {
6239                                 up_read(&obj->oo_ext_idx_sem);
6240                         }
6241
6242                         goto again;
6243                 }
6244
6245                 dirty = true;
6246                 if (unlikely(fid_is_sane(fid))) {
6247                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6248                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6249                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6250                         if (rc == 0)
6251                                 *attr |= LUDA_REPAIR;
6252                         else
6253                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
6254                                        "update dirent, dir = %lu/%u, "
6255                                        "name = %.*s, "DFID": rc = %d\n",
6256                                        devname, dir->i_ino, dir->i_generation,
6257                                        ent->oied_namelen, ent->oied_name,
6258                                        PFID(fid), rc);
6259                 } else {
6260                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6261                         /* It is probably IGIF object. Only aappend the
6262                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6263                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6264                                                  bh, de, hlock, dot_dotdot);
6265                         if (rc == 0)
6266                                 *attr |= LUDA_UPGRADE;
6267                         else
6268                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
6269                                        "after the dirent, dir = %lu/%u, "
6270                                        "name = %.*s, "DFID": rc = %d\n",
6271                                        devname, dir->i_ino, dir->i_generation,
6272                                        ent->oied_namelen, ent->oied_name,
6273                                        PFID(fid), rc);
6274                 }
6275         }
6276
6277         GOTO(out, rc);
6278
6279 out:
6280         if (!IS_ERR(bh))
6281                 brelse(bh);
6282         if (hlock != NULL) {
6283                 ldiskfs_htree_unlock(hlock);
6284         } else {
6285                 if (dev->od_dirent_journal != 0)
6286                         up_write(&obj->oo_ext_idx_sem);
6287                 else
6288                         up_read(&obj->oo_ext_idx_sem);
6289         }
6290
6291         if (jh != NULL)
6292                 ldiskfs_journal_stop(jh);
6293
6294 out_inode:
6295         iput(inode);
6296         if (rc >= 0 && !dirty)
6297                 dev->od_dirent_journal = 0;
6298
6299         return rc;
6300 }
6301
6302 /**
6303  * Returns the value at current position from iterator's in memory structure.
6304  *
6305  * \param di struct osd_it_ea, iterator's in memory structure
6306  * \param attr attr requested for dirent.
6307  * \param lde lustre dirent
6308  *
6309  * \retval   0 no error and \param lde has correct lustre dirent.
6310  * \retval -ve on error
6311  */
6312 static inline int osd_it_ea_rec(const struct lu_env *env,
6313                                 const struct dt_it *di,
6314                                 struct dt_rec *dtrec, __u32 attr)
6315 {
6316         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6317         struct osd_object      *obj   = it->oie_obj;
6318         struct osd_device      *dev   = osd_obj2dev(obj);
6319         struct osd_thread_info *oti   = osd_oti_get(env);
6320         struct osd_inode_id    *id    = &oti->oti_id;
6321         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6322         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6323         __u32                   ino   = it->oie_dirent->oied_ino;
6324         int                     rc    = 0;
6325         ENTRY;
6326
6327         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6328
6329         if (attr & LUDA_VERIFY) {
6330                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6331                         attr |= LUDA_IGNORE;
6332                         /* If the parent is on remote MDT, and there
6333                          * is no FID-in-dirent, then we have to get
6334                          * the parent FID from the linkEA.  */
6335                         if (!fid_is_sane(fid) &&
6336                             it->oie_dirent->oied_namelen == 2 &&
6337                             it->oie_dirent->oied_name[0] == '.' &&
6338                             it->oie_dirent->oied_name[1] == '.')
6339                                 osd_get_pfid_from_linkea(env, obj, fid);
6340                 } else {
6341                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6342                                                      &attr);
6343                 }
6344
6345                 if (!fid_is_sane(fid)) {
6346                         attr &= ~LUDA_IGNORE;
6347                         attr |= LUDA_UNKNOWN;
6348                 }
6349         } else {
6350                 attr &= ~LU_DIRENT_ATTRS_MASK;
6351                 if (!fid_is_sane(fid)) {
6352                         bool is_dotdot = false;
6353                         if (it->oie_dirent->oied_namelen == 2 &&
6354                             it->oie_dirent->oied_name[0] == '.' &&
6355                             it->oie_dirent->oied_name[1] == '.')
6356                                 is_dotdot = true;
6357                         /* If the parent is on remote MDT, and there
6358                          * is no FID-in-dirent, then we have to get
6359                          * the parent FID from the linkEA.  */
6360                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6361                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6362                         } else {
6363                                 if (is_dotdot == false &&
6364                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6365                                         RETURN(-ENOENT);
6366
6367                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6368                         }
6369                 } else {
6370                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6371                 }
6372         }
6373
6374         /* Pack the entry anyway, at least the offset is right. */
6375         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6376                            it->oie_dirent->oied_name,
6377                            it->oie_dirent->oied_namelen,
6378                            it->oie_dirent->oied_type, attr);
6379
6380         if (rc < 0)
6381                 RETURN(rc);
6382
6383         if (osd_remote_fid(env, dev, fid))
6384                 RETURN(0);
6385
6386         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6387                 osd_add_oi_cache(oti, dev, id, fid);
6388
6389         RETURN(rc > 0 ? 0 : rc);
6390 }
6391
6392 /**
6393  * Returns the record size size at current position.
6394  *
6395  * This function will return record(lu_dirent) size in bytes.
6396  *
6397  * \param[in] env       execution environment
6398  * \param[in] di        iterator's in memory structure
6399  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6400  *                      calculate the lu_dirent size.
6401  *
6402  * \retval      record size(in bytes & in memory) of the current lu_dirent
6403  *              entry.
6404  */
6405 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6406                               __u32 attr)
6407 {
6408         struct osd_it_ea *it = (struct osd_it_ea *)di;
6409
6410         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6411 }
6412
6413 /**
6414  * Returns a cookie for current position of the iterator head, so that
6415  * user can use this cookie to load/start the iterator next time.
6416  *
6417  * \param di iterator's in memory structure
6418  *
6419  * \retval cookie for current position, on success
6420  */
6421 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6422 {
6423         struct osd_it_ea *it = (struct osd_it_ea *)di;
6424
6425         return it->oie_dirent->oied_off;
6426 }
6427
6428 /**
6429  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6430  * to load a directory entry at a time and stored it i inn,
6431  * in iterator's in-memory data structure.
6432  *
6433  * \param di struct osd_it_ea, iterator's in memory structure
6434  *
6435  * \retval +ve on success
6436  * \retval -ve on error
6437  */
6438 static int osd_it_ea_load(const struct lu_env *env,
6439                           const struct dt_it *di, __u64 hash)
6440 {
6441         struct osd_it_ea *it = (struct osd_it_ea *)di;
6442         int rc;
6443
6444         ENTRY;
6445         it->oie_file.f_pos = hash;
6446
6447         rc =  osd_ldiskfs_it_fill(env, di);
6448         if (rc > 0)
6449                 rc = -ENODATA;
6450
6451         if (rc == 0)
6452                 rc = +1;
6453
6454         RETURN(rc);
6455 }
6456
6457 /**
6458  * Index lookup function for interoperability mode (b11826).
6459  *
6460  * \param key,  key i.e. file name to be searched
6461  *
6462  * \retval +ve, on success
6463  * \retval -ve, on error
6464  */
6465 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6466                                struct dt_rec *rec, const struct dt_key *key)
6467 {
6468         struct osd_object *obj = osd_dt_obj(dt);
6469         int rc = 0;
6470
6471         ENTRY;
6472
6473         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6474         LINVRNT(osd_invariant(obj));
6475
6476         rc = osd_ea_lookup_rec(env, obj, rec, key);
6477         if (rc == 0)
6478                 rc = +1;
6479         RETURN(rc);
6480 }
6481
6482 /**
6483  * Index and Iterator operations for interoperability
6484  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6485  */
6486 static const struct dt_index_operations osd_index_ea_ops = {
6487         .dio_lookup         = osd_index_ea_lookup,
6488         .dio_declare_insert = osd_index_declare_ea_insert,
6489         .dio_insert         = osd_index_ea_insert,
6490         .dio_declare_delete = osd_index_declare_ea_delete,
6491         .dio_delete         = osd_index_ea_delete,
6492         .dio_it     = {
6493                 .init     = osd_it_ea_init,
6494                 .fini     = osd_it_ea_fini,
6495                 .get      = osd_it_ea_get,
6496                 .put      = osd_it_ea_put,
6497                 .next     = osd_it_ea_next,
6498                 .key      = osd_it_ea_key,
6499                 .key_size = osd_it_ea_key_size,
6500                 .rec      = osd_it_ea_rec,
6501                 .rec_size = osd_it_ea_rec_size,
6502                 .store    = osd_it_ea_store,
6503                 .load     = osd_it_ea_load
6504         }
6505 };
6506
6507 static void *osd_key_init(const struct lu_context *ctx,
6508                           struct lu_context_key *key)
6509 {
6510         struct osd_thread_info *info;
6511
6512         OBD_ALLOC_PTR(info);
6513         if (info == NULL)
6514                 return ERR_PTR(-ENOMEM);
6515
6516         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6517         if (info->oti_it_ea_buf == NULL)
6518                 goto out_free_info;
6519
6520         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6521
6522         info->oti_hlock = ldiskfs_htree_lock_alloc();
6523         if (info->oti_hlock == NULL)
6524                 goto out_free_ea;
6525
6526         return info;
6527
6528  out_free_ea:
6529         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6530  out_free_info:
6531         OBD_FREE_PTR(info);
6532         return ERR_PTR(-ENOMEM);
6533 }
6534
6535 static void osd_key_fini(const struct lu_context *ctx,
6536                          struct lu_context_key *key, void* data)
6537 {
6538         struct osd_thread_info *info = data;
6539         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6540         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6541
6542         if (info->oti_inode != NULL)
6543                 OBD_FREE_PTR(lli);
6544         if (info->oti_hlock != NULL)
6545                 ldiskfs_htree_lock_free(info->oti_hlock);
6546         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6547         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6548         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6549         lu_buf_free(&info->oti_big_buf);
6550         if (idc != NULL) {
6551                 LASSERT(info->oti_ins_cache_size > 0);
6552                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6553                 info->oti_ins_cache = NULL;
6554                 info->oti_ins_cache_size = 0;
6555         }
6556         OBD_FREE_PTR(info);
6557 }
6558
6559 static void osd_key_exit(const struct lu_context *ctx,
6560                          struct lu_context_key *key, void *data)
6561 {
6562         struct osd_thread_info *info = data;
6563
6564         LASSERT(info->oti_r_locks == 0);
6565         LASSERT(info->oti_w_locks == 0);
6566         LASSERT(info->oti_txns    == 0);
6567 }
6568
6569 /* type constructor/destructor: osd_type_init, osd_type_fini */
6570 LU_TYPE_INIT_FINI(osd, &osd_key);
6571
6572 struct lu_context_key osd_key = {
6573         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6574         .lct_init = osd_key_init,
6575         .lct_fini = osd_key_fini,
6576         .lct_exit = osd_key_exit
6577 };
6578
6579
6580 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6581                            const char *name, struct lu_device *next)
6582 {
6583         struct osd_device *osd = osd_dev(d);
6584
6585         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6586             >= sizeof(osd->od_svname))
6587                 return -E2BIG;
6588         return osd_procfs_init(osd, name);
6589 }
6590
6591 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6592 {
6593         struct seq_server_site  *ss = osd_seq_site(osd);
6594         int                     rc;
6595         ENTRY;
6596
6597         if (osd->od_is_ost || osd->od_cl_seq != NULL)
6598                 RETURN(0);
6599
6600         if (unlikely(ss == NULL))
6601                 RETURN(-ENODEV);
6602
6603         OBD_ALLOC_PTR(osd->od_cl_seq);
6604         if (osd->od_cl_seq == NULL)
6605                 RETURN(-ENOMEM);
6606
6607         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6608                              osd->od_svname, ss->ss_server_seq);
6609         if (rc != 0) {
6610                 OBD_FREE_PTR(osd->od_cl_seq);
6611                 osd->od_cl_seq = NULL;
6612                 RETURN(rc);
6613         }
6614
6615         if (ss->ss_node_id == 0) {
6616                 /* If the OSD on the sequence controller(MDT0), then allocate
6617                  * sequence here, otherwise allocate sequence after connected
6618                  * to MDT0 (see mdt_register_lwp_callback()). */
6619                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6620                                    &osd->od_cl_seq->lcs_space, env);
6621         }
6622
6623         RETURN(rc);
6624 }
6625
6626 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6627 {
6628         if (osd->od_cl_seq == NULL)
6629                 return;
6630
6631         seq_client_fini(osd->od_cl_seq);
6632         OBD_FREE_PTR(osd->od_cl_seq);
6633         osd->od_cl_seq = NULL;
6634 }
6635
6636 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6637 {
6638         ENTRY;
6639
6640         /* shutdown quota slave instance associated with the device */
6641         if (o->od_quota_slave != NULL) {
6642                 qsd_fini(env, o->od_quota_slave);
6643                 o->od_quota_slave = NULL;
6644         }
6645
6646         osd_fid_fini(env, o);
6647
6648         RETURN(0);
6649 }
6650
6651 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6652 {
6653         ENTRY;
6654
6655         if (o->od_mnt != NULL) {
6656                 shrink_dcache_sb(osd_sb(o));
6657                 osd_sync(env, &o->od_dt_dev);
6658
6659                 mntput(o->od_mnt);
6660                 o->od_mnt = NULL;
6661         }
6662
6663         EXIT;
6664 }
6665
6666 static int osd_mount(const struct lu_env *env,
6667                      struct osd_device *o, struct lustre_cfg *cfg)
6668 {
6669         const char              *name  = lustre_cfg_string(cfg, 0);
6670         const char              *dev  = lustre_cfg_string(cfg, 1);
6671         const char              *opts;
6672         unsigned long            page, s_flags, lmd_flags = 0;
6673         struct page             *__page;
6674         struct file_system_type *type;
6675         char                    *options = NULL;
6676         char                    *str;
6677         struct osd_thread_info  *info = osd_oti_get(env);
6678         struct lu_fid           *fid = &info->oti_fid;
6679         struct inode            *inode;
6680         int                      rc = 0, force_over_256tb = 0;
6681         ENTRY;
6682
6683         if (o->od_mnt != NULL)
6684                 RETURN(0);
6685
6686         if (strlen(dev) >= sizeof(o->od_mntdev))
6687                 RETURN(-E2BIG);
6688         strcpy(o->od_mntdev, dev);
6689
6690         str = lustre_cfg_string(cfg, 2);
6691         s_flags = simple_strtoul(str, NULL, 0);
6692         str = strstr(str, ":");
6693         if (str)
6694                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6695         opts = lustre_cfg_string(cfg, 3);
6696 #ifdef __BIG_ENDIAN
6697         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6698                 CERROR("%s: device %s extents feature is not guaranteed to "
6699                        "work on big-endian systems. Use \"bigendian_extents\" "
6700                        "mount option to override.\n", name, dev);
6701                 RETURN(-EINVAL);
6702         }
6703 #endif
6704         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6705                 CWARN("force_over_128tb option is depricated."
6706                       "Filesystems less then 256TB can be created without any"
6707                       "force options. Use force_over_256tb option for"
6708                       "filesystems greather then 256TB.\n");
6709         }
6710
6711         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL)
6712                 force_over_256tb = 1;
6713
6714         __page = alloc_page(GFP_KERNEL);
6715         if (__page == NULL)
6716                 GOTO(out, rc = -ENOMEM);
6717         page = (unsigned long)page_address(__page);
6718         options = (char *)page;
6719         *options = '\0';
6720         if (opts != NULL) {
6721                 /* strip out the options for back compatiblity */
6722                 static char *sout[] = {
6723                         "mballoc",
6724                         "iopen",
6725                         "noiopen",
6726                         "iopen_nopriv",
6727                         "extents",
6728                         "noextents",
6729                         /* strip out option we processed in osd */
6730                         "bigendian_extents",
6731 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(3,0,53,0)
6732 #warning "remove force_over_128 option"
6733 #else
6734                         "force_over_128tb (deprecated)",
6735 #endif
6736                         "force_over_256tb",
6737                         NULL
6738                 };
6739                 strcat(options, opts);
6740                 for (rc = 0, str = options; sout[rc]; ) {
6741                         char *op = strstr(str, sout[rc]);
6742                         if (op == NULL) {
6743                                 rc++;
6744                                 str = options;
6745                                 continue;
6746                         }
6747                         if (op == options || *(op - 1) == ',') {
6748                                 str = op + strlen(sout[rc]);
6749                                 if (*str == ',' || *str == '\0') {
6750                                         *str == ',' ? str++ : str;
6751                                         memmove(op, str, strlen(str) + 1);
6752                                 }
6753                         }
6754                         for (str = op; *str != ',' && *str != '\0'; str++)
6755                                 ;
6756                 }
6757         } else {
6758                 strncat(options, "user_xattr,acl", 14);
6759         }
6760
6761         /* Glom up mount options */
6762         if (*options != '\0')
6763                 strcat(options, ",");
6764         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6765
6766         type = get_fs_type("ldiskfs");
6767         if (!type) {
6768                 CERROR("%s: cannot find ldiskfs module\n", name);
6769                 GOTO(out, rc = -ENODEV);
6770         }
6771
6772         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6773         module_put(type->owner);
6774
6775         if (IS_ERR(o->od_mnt)) {
6776                 rc = PTR_ERR(o->od_mnt);
6777                 o->od_mnt = NULL;
6778                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6779                 GOTO(out, rc);
6780         }
6781
6782         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (64ULL << 30) &&
6783             force_over_256tb == 0) {
6784                 CERROR("%s: device %s LDISKFS does not support filesystems "
6785                        "greater than 256TB and can cause data corruption. "
6786                        "Use \"force_over_256tb\" mount option to override.\n",
6787                        name, dev);
6788                 GOTO(out, rc = -EINVAL);
6789         }
6790
6791 #ifdef HAVE_DEV_SET_RDONLY
6792         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6793                 CERROR("%s: underlying device %s is marked as read-only. "
6794                        "Setup failed\n", name, dev);
6795                 GOTO(out_mnt, rc = -EROFS);
6796         }
6797 #endif
6798
6799         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6800                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6801                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6802                 GOTO(out_mnt, rc = -EINVAL);
6803         }
6804
6805 #ifdef LDISKFS_MOUNT_DIRDATA
6806         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6807                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6808                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6809         else if (!o->od_is_ost)
6810                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6811                       "enabling the dirdata feature. If you do not want to "
6812                       "downgrade to Lustre-1.x again, you can enable it via "
6813                       "'tune2fs -O dirdata device'\n", name, dev);
6814 #endif
6815         inode = osd_sb(o)->s_root->d_inode;
6816         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6817         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6818         if (rc != 0) {
6819                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6820                 GOTO(out_mnt, rc);
6821         }
6822
6823         if (lmd_flags & LMD_FLG_NOSCRUB)
6824                 o->od_noscrub = 1;
6825
6826         GOTO(out, rc = 0);
6827
6828 out_mnt:
6829         mntput(o->od_mnt);
6830         o->od_mnt = NULL;
6831
6832 out:
6833         if (__page)
6834                 __free_page(__page);
6835
6836         return rc;
6837 }
6838
6839 static struct lu_device *osd_device_fini(const struct lu_env *env,
6840                                          struct lu_device *d)
6841 {
6842         struct osd_device *o = osd_dev(d);
6843         ENTRY;
6844
6845         osd_shutdown(env, o);
6846         osd_procfs_fini(o);
6847         osd_scrub_cleanup(env, o);
6848         osd_obj_map_fini(o);
6849         osd_umount(env, o);
6850
6851         RETURN(NULL);
6852 }
6853
6854 static int osd_device_init0(const struct lu_env *env,
6855                             struct osd_device *o,
6856                             struct lustre_cfg *cfg)
6857 {
6858         struct lu_device        *l = osd2lu_dev(o);
6859         struct osd_thread_info *info;
6860         int                     rc;
6861         int                     cplen = 0;
6862
6863         /* if the module was re-loaded, env can loose its keys */
6864         rc = lu_env_refill((struct lu_env *) env);
6865         if (rc)
6866                 GOTO(out, rc);
6867         info = osd_oti_get(env);
6868         LASSERT(info);
6869
6870         l->ld_ops = &osd_lu_ops;
6871         o->od_dt_dev.dd_ops = &osd_dt_ops;
6872
6873         spin_lock_init(&o->od_osfs_lock);
6874         mutex_init(&o->od_otable_mutex);
6875         INIT_LIST_HEAD(&o->od_orphan_list);
6876
6877         o->od_read_cache = 1;
6878         o->od_writethrough_cache = 1;
6879         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6880
6881         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6882                         sizeof(o->od_svname));
6883         if (cplen >= sizeof(o->od_svname)) {
6884                 rc = -E2BIG;
6885                 GOTO(out, rc);
6886         }
6887
6888         o->od_index = -1; /* -1 means index is invalid */
6889         rc = server_name2index(o->od_svname, &o->od_index, NULL);
6890         if (rc == LDD_F_SV_TYPE_OST)
6891                 o->od_is_ost = 1;
6892
6893         o->od_full_scrub_ratio = OFSR_DEFAULT;
6894         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6895         rc = osd_mount(env, o, cfg);
6896         if (rc != 0)
6897                 GOTO(out, rc);
6898
6899         rc = osd_obj_map_init(env, o);
6900         if (rc != 0)
6901                 GOTO(out_mnt, rc);
6902
6903         rc = lu_site_init(&o->od_site, l);
6904         if (rc != 0)
6905                 GOTO(out_compat, rc);
6906         o->od_site.ls_bottom_dev = l;
6907
6908         rc = lu_site_init_finish(&o->od_site);
6909         if (rc != 0)
6910                 GOTO(out_site, rc);
6911
6912         INIT_LIST_HEAD(&o->od_ios_list);
6913         /* setup scrub, including OI files initialization */
6914         rc = osd_scrub_setup(env, o);
6915         if (rc < 0)
6916                 GOTO(out_site, rc);
6917
6918         rc = osd_procfs_init(o, o->od_svname);
6919         if (rc != 0) {
6920                 CERROR("%s: can't initialize procfs: rc = %d\n",
6921                        o->od_svname, rc);
6922                 GOTO(out_scrub, rc);
6923         }
6924
6925         LASSERT(l->ld_site->ls_linkage.next != NULL);
6926         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6927
6928         /* initialize quota slave instance */
6929         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6930                                      o->od_proc_entry);
6931         if (IS_ERR(o->od_quota_slave)) {
6932                 rc = PTR_ERR(o->od_quota_slave);
6933                 o->od_quota_slave = NULL;
6934                 GOTO(out_procfs, rc);
6935         }
6936
6937         RETURN(0);
6938
6939 out_procfs:
6940         osd_procfs_fini(o);
6941 out_scrub:
6942         osd_scrub_cleanup(env, o);
6943 out_site:
6944         lu_site_fini(&o->od_site);
6945 out_compat:
6946         osd_obj_map_fini(o);
6947 out_mnt:
6948         osd_umount(env, o);
6949 out:
6950         return rc;
6951 }
6952
6953 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6954                                           struct lu_device_type *t,
6955                                           struct lustre_cfg *cfg)
6956 {
6957         struct osd_device *o;
6958         int                rc;
6959
6960         OBD_ALLOC_PTR(o);
6961         if (o == NULL)
6962                 return ERR_PTR(-ENOMEM);
6963
6964         rc = dt_device_init(&o->od_dt_dev, t);
6965         if (rc == 0) {
6966                 /* Because the ctx might be revived in dt_device_init,
6967                  * refill the env here */
6968                 lu_env_refill((struct lu_env *)env);
6969                 rc = osd_device_init0(env, o, cfg);
6970                 if (rc)
6971                         dt_device_fini(&o->od_dt_dev);
6972         }
6973
6974         if (unlikely(rc != 0))
6975                 OBD_FREE_PTR(o);
6976
6977         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6978 }
6979
6980 static struct lu_device *osd_device_free(const struct lu_env *env,
6981                                          struct lu_device *d)
6982 {
6983         struct osd_device *o = osd_dev(d);
6984         ENTRY;
6985
6986         /* XXX: make osd top device in order to release reference */
6987         d->ld_site->ls_top_dev = d;
6988         lu_site_purge(env, d->ld_site, -1);
6989         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6990                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6991                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6992         }
6993         lu_site_fini(&o->od_site);
6994         dt_device_fini(&o->od_dt_dev);
6995         OBD_FREE_PTR(o);
6996         RETURN(NULL);
6997 }
6998
6999 static int osd_process_config(const struct lu_env *env,
7000                               struct lu_device *d, struct lustre_cfg *cfg)
7001 {
7002         struct osd_device               *o = osd_dev(d);
7003         int                             rc;
7004         ENTRY;
7005
7006         switch (cfg->lcfg_command) {
7007         case LCFG_SETUP:
7008                 rc = osd_mount(env, o, cfg);
7009                 break;
7010         case LCFG_CLEANUP:
7011                 lu_dev_del_linkage(d->ld_site, d);
7012                 rc = osd_shutdown(env, o);
7013                 break;
7014         case LCFG_PARAM:
7015                 LASSERT(&o->od_dt_dev);
7016                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7017                                               cfg, &o->od_dt_dev);
7018                 if (rc > 0 || rc == -ENOSYS)
7019                         rc = class_process_proc_param(PARAM_OST,
7020                                                       lprocfs_osd_obd_vars,
7021                                                       cfg, &o->od_dt_dev);
7022                 break;
7023         default:
7024                 rc = -ENOSYS;
7025         }
7026
7027         RETURN(rc);
7028 }
7029
7030 static int osd_recovery_complete(const struct lu_env *env,
7031                                  struct lu_device *d)
7032 {
7033         struct osd_device       *osd = osd_dev(d);
7034         int                      rc = 0;
7035         ENTRY;
7036
7037         if (osd->od_quota_slave == NULL)
7038                 RETURN(0);
7039
7040         /* start qsd instance on recovery completion, this notifies the quota
7041          * slave code that we are about to process new requests now */
7042         rc = qsd_start(env, osd->od_quota_slave);
7043         RETURN(rc);
7044 }
7045
7046 /*
7047  * we use exports to track all osd users
7048  */
7049 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7050                            struct obd_device *obd, struct obd_uuid *cluuid,
7051                            struct obd_connect_data *data, void *localdata)
7052 {
7053         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7054         struct lustre_handle  conn;
7055         int                   rc;
7056         ENTRY;
7057
7058         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7059
7060         rc = class_connect(&conn, obd, cluuid);
7061         if (rc)
7062                 RETURN(rc);
7063
7064         *exp = class_conn2export(&conn);
7065
7066         spin_lock(&osd->od_osfs_lock);
7067         osd->od_connects++;
7068         spin_unlock(&osd->od_osfs_lock);
7069
7070         RETURN(0);
7071 }
7072
7073 /*
7074  * once last export (we don't count self-export) disappeared
7075  * osd can be released
7076  */
7077 static int osd_obd_disconnect(struct obd_export *exp)
7078 {
7079         struct obd_device *obd = exp->exp_obd;
7080         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7081         int                rc, release = 0;
7082         ENTRY;
7083
7084         /* Only disconnect the underlying layers on the final disconnect. */
7085         spin_lock(&osd->od_osfs_lock);
7086         osd->od_connects--;
7087         if (osd->od_connects == 0)
7088                 release = 1;
7089         spin_unlock(&osd->od_osfs_lock);
7090
7091         rc = class_disconnect(exp); /* bz 9811 */
7092
7093         if (rc == 0 && release)
7094                 class_manual_cleanup(obd);
7095         RETURN(rc);
7096 }
7097
7098 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7099                        struct lu_device *dev)
7100 {
7101         struct osd_device       *osd    = osd_dev(dev);
7102         struct lr_server_data   *lsd    =
7103                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7104         int                      result = 0;
7105         ENTRY;
7106
7107         if (osd->od_quota_slave != NULL) {
7108                 /* set up quota slave objects */
7109                 result = qsd_prepare(env, osd->od_quota_slave);
7110                 if (result != 0)
7111                         RETURN(result);
7112         }
7113
7114         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7115 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7116                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7117                         osd->od_index_in_idif = 1;
7118                 } else {
7119                         osd->od_index_in_idif = 0;
7120                         result = osd_register_proc_index_in_idif(osd);
7121                         if (result != 0)
7122                                 RETURN(result);
7123                 }
7124 #else
7125                 osd->od_index_in_idif = 1;
7126 #endif
7127         }
7128
7129         result = osd_fid_init(env, osd);
7130
7131         RETURN(result);
7132 }
7133
7134 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7135                          struct lu_fid *fid, struct md_op_data *op_data)
7136 {
7137         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7138
7139         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7140 }
7141
7142 static const struct lu_object_operations osd_lu_obj_ops = {
7143         .loo_object_init      = osd_object_init,
7144         .loo_object_delete    = osd_object_delete,
7145         .loo_object_release   = osd_object_release,
7146         .loo_object_free      = osd_object_free,
7147         .loo_object_print     = osd_object_print,
7148         .loo_object_invariant = osd_object_invariant
7149 };
7150
7151 const struct lu_device_operations osd_lu_ops = {
7152         .ldo_object_alloc      = osd_object_alloc,
7153         .ldo_process_config    = osd_process_config,
7154         .ldo_recovery_complete = osd_recovery_complete,
7155         .ldo_prepare           = osd_prepare,
7156 };
7157
7158 static const struct lu_device_type_operations osd_device_type_ops = {
7159         .ldto_init = osd_type_init,
7160         .ldto_fini = osd_type_fini,
7161
7162         .ldto_start = osd_type_start,
7163         .ldto_stop  = osd_type_stop,
7164
7165         .ldto_device_alloc = osd_device_alloc,
7166         .ldto_device_free  = osd_device_free,
7167
7168         .ldto_device_init    = osd_device_init,
7169         .ldto_device_fini    = osd_device_fini
7170 };
7171
7172 static struct lu_device_type osd_device_type = {
7173         .ldt_tags     = LU_DEVICE_DT,
7174         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7175         .ldt_ops      = &osd_device_type_ops,
7176         .ldt_ctx_tags = LCT_LOCAL,
7177 };
7178
7179 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7180 {
7181         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7182         struct super_block *sb = osd_sb(osd);
7183
7184         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7185 }
7186
7187 /*
7188  * lprocfs legacy support.
7189  */
7190 static struct obd_ops osd_obd_device_ops = {
7191         .o_owner = THIS_MODULE,
7192         .o_connect      = osd_obd_connect,
7193         .o_disconnect   = osd_obd_disconnect,
7194         .o_fid_alloc    = osd_fid_alloc,
7195         .o_health_check = osd_health_check,
7196 };
7197
7198 static int __init osd_init(void)
7199 {
7200         int rc;
7201
7202         LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7203 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7204         /* please, try to keep osd_thread_info smaller than a page */
7205         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7206 #endif
7207
7208         osd_oi_mod_init();
7209
7210         rc = lu_kmem_init(ldiskfs_caches);
7211         if (rc)
7212                 return rc;
7213
7214         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7215                                  lprocfs_osd_module_vars,
7216                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7217         if (rc)
7218                 lu_kmem_fini(ldiskfs_caches);
7219         return rc;
7220 }
7221
7222 static void __exit osd_exit(void)
7223 {
7224         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7225         lu_kmem_fini(ldiskfs_caches);
7226 }
7227
7228 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7229 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7230 MODULE_VERSION(LUSTRE_VERSION_STRING);
7231 MODULE_LICENSE("GPL");
7232
7233 module_init(osd_init);
7234 module_exit(osd_exit);