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, 2016, 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 #define PFID_STRIPE_IDX_BITS    16
  82 #define PFID_STRIPE_COUNT_MASK  ((1 << PFID_STRIPE_IDX_BITS) - 1)
  83
  84 int ldiskfs_pdo = 1;
  85 module_param(ldiskfs_pdo, int, 0644);
  86 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  87
  88 int ldiskfs_track_declares_assert;
  89 module_param(ldiskfs_track_declares_assert, int, 0644);
  90 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  91
  92 /* Slab to allocate dynlocks */
  93 struct kmem_cache *dynlock_cachep;
  94
  95 /* Slab to allocate osd_it_ea */
  96 struct kmem_cache *osd_itea_cachep;
  97
  98 static struct lu_kmem_descr ldiskfs_caches[] = {
  99         {
 100                 .ckd_cache = &dynlock_cachep,
 101                 .ckd_name  = "dynlock_cache",
 102                 .ckd_size  = sizeof(struct dynlock_handle)
 103         },
 104         {
 105                 .ckd_cache = &osd_itea_cachep,
 106                 .ckd_name  = "osd_itea_cache",
 107                 .ckd_size  = sizeof(struct osd_it_ea)
 108         },
 109         {
 110                 .ckd_cache = NULL
 111         }
 112 };
 113
 114 static const char dot[] = ".";
 115 static const char dotdot[] = "..";
 116
 117 static const struct lu_object_operations      osd_lu_obj_ops;
 118 static const struct dt_object_operations      osd_obj_ops;
 119 static const struct dt_object_operations      osd_obj_otable_it_ops;
 120 static const struct dt_index_operations       osd_index_iam_ops;
 121 static const struct dt_index_operations       osd_index_ea_ops;
 122
 123 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 124                           const struct lu_fid *fid);
 125 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 126                                                 struct osd_device *osd);
 127
 128 int osd_trans_declare_op2rb[] = {
 129         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 130         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 131         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 132         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 133         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 134         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 135         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 136         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 137         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 138         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 139         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 140 };
 141
 142 static int osd_has_index(const struct osd_object *obj)
 143 {
 144         return obj->oo_dt.do_index_ops != NULL;
 145 }
 146
 147 static int osd_object_invariant(const struct lu_object *l)
 148 {
 149         return osd_invariant(osd_obj(l));
 150 }
 151
 152 /*
 153  * Concurrency: doesn't matter
 154  */
 155 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 156 {
 157         struct osd_thread_info *oti = osd_oti_get(env);
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info  *oti   = osd_oti_get(env);
 187         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
 188         int i;
 189         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 190                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 191                         continue;
 192                 if (idc[i].oic_dev != osd)
 193                         continue;
 194
 195                 return idc + i;
 196         }
 197
 198         return NULL;
 199 }
 200
 201 static struct osd_idmap_cache *
 202 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 203             const struct lu_fid *fid)
 204 {
 205         struct osd_thread_info  *oti   = osd_oti_get(env);
 206         struct osd_idmap_cache  *idc;
 207         int i;
 208
 209         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 210                 i = oti->oti_ins_cache_size * 2;
 211                 if (i == 0)
 212                         i = OSD_INS_CACHE_SIZE;
 213                 OBD_ALLOC(idc, sizeof(*idc) * i);
 214                 if (idc == NULL)
 215                         return ERR_PTR(-ENOMEM);
 216                 if (oti->oti_ins_cache != NULL) {
 217                         memcpy(idc, oti->oti_ins_cache,
 218                                oti->oti_ins_cache_used * sizeof(*idc));
 219                         OBD_FREE(oti->oti_ins_cache,
 220                                  oti->oti_ins_cache_used * sizeof(*idc));
 221                 }
 222                 oti->oti_ins_cache = idc;
 223                 oti->oti_ins_cache_size = i;
 224         }
 225
 226         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 227         idc->oic_fid = *fid;
 228         idc->oic_dev = osd;
 229         idc->oic_lid.oii_ino = 0;
 230         idc->oic_lid.oii_gen = 0;
 231         idc->oic_remote = 0;
 232
 233         return idc;
 234 }
 235
 236 /*
 237  * lookup mapping for the given fid in the cache, initialize a
 238  * new one if not found. the initialization checks whether the
 239  * object is local or remote. for local objects, OI is used to
 240  * learn ino/generation. the function is used when the caller
 241  * has no information about the object, e.g. at dt_insert().
 242  */
 243 static struct osd_idmap_cache *
 244 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 245                      const struct lu_fid *fid)
 246 {
 247         struct osd_idmap_cache *idc;
 248         int rc;
 249
 250         idc = osd_idc_find(env, osd, fid);
 251         LASSERT(!IS_ERR(idc));
 252         if (idc != NULL)
 253                 return idc;
 254
 255         /* new mapping is needed */
 256         idc = osd_idc_add(env, osd, fid);
 257         if (IS_ERR(idc))
 258                 return idc;
 259
 260         /* initialize it */
 261         rc = osd_remote_fid(env, osd, fid);
 262         if (unlikely(rc < 0))
 263                 return ERR_PTR(rc);
 264
 265         if (rc == 0) {
 266                 /* the object is local, lookup in OI */
 267                 /* XXX: probably cheaper to lookup in LU first? */
 268                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 269                                    &idc->oic_lid, 0);
 270                 if (unlikely(rc < 0)) {
 271                         CERROR("can't lookup: rc = %d\n", rc);
 272                         return ERR_PTR(rc);
 273                 }
 274         } else {
 275                 /* the object is remote */
 276                 idc->oic_remote = 1;
 277         }
 278
 279         return idc;
 280 }
 281
 282 /*
 283  * lookup mapping for given FID and fill it from the given object.
 284  * the object is lolcal by definition.
 285  */
 286 static int osd_idc_find_and_init(const struct lu_env *env,
 287                                  struct osd_device *osd,
 288                                  struct osd_object *obj)
 289 {
 290         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 291         struct osd_idmap_cache  *idc;
 292
 293         idc = osd_idc_find(env, osd, fid);
 294         LASSERT(!IS_ERR(idc));
 295         if (idc != NULL) {
 296                 if (obj->oo_inode == NULL)
 297                         return 0;
 298                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 299                         LASSERT(idc->oic_lid.oii_ino == 0);
 300                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 301                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 302                 }
 303                 return 0;
 304         }
 305
 306         /* new mapping is needed */
 307         idc = osd_idc_add(env, osd, fid);
 308         if (IS_ERR(idc))
 309                 return PTR_ERR(idc);
 310
 311         if (obj->oo_inode != NULL) {
 312                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 313                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 314         }
 315         return 0;
 316 }
 317
 318 /*
 319  * OSD object methods.
 320  */
 321
 322 /*
 323  * Concurrency: no concurrent access is possible that early in object
 324  * life-cycle.
 325  */
 326 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 327                                           const struct lu_object_header *hdr,
 328                                           struct lu_device *d)
 329 {
 330         struct osd_object *mo;
 331
 332         OBD_ALLOC_PTR(mo);
 333         if (mo != NULL) {
 334                 struct lu_object *l;
 335
 336                 l = &mo->oo_dt.do_lu;
 337                 dt_object_init(&mo->oo_dt, NULL, d);
 338                 mo->oo_dt.do_ops = &osd_obj_ops;
 339                 l->lo_ops = &osd_lu_obj_ops;
 340                 init_rwsem(&mo->oo_sem);
 341                 init_rwsem(&mo->oo_ext_idx_sem);
 342                 spin_lock_init(&mo->oo_guard);
 343                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 344                 return l;
 345         } else {
 346                 return NULL;
 347         }
 348 }
 349
 350 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 351                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 352 {
 353         int rc;
 354
 355         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 356                              (void *)loa, sizeof(*loa));
 357         if (rc > 0) {
 358                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 359
 360                 if (rc < sizeof(*lma))
 361                         return -EINVAL;
 362
 363                 rc = 0;
 364                 lustre_loa_swab(loa, true);
 365                 /* Check LMA compatibility */
 366                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 367                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 368                               "for fid = "DFID", ino = %lu\n",
 369                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 370                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 371                               PFID(&lma->lma_self_fid), inode->i_ino);
 372                         rc = -EOPNOTSUPP;
 373                 }
 374         } else if (rc == 0) {
 375                 rc = -ENODATA;
 376         }
 377
 378         return rc;
 379 }
 380
 381 /*
 382  * retrieve object from backend ext fs.
 383  **/
 384 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 385                        struct osd_inode_id *id)
 386 {
 387         int rc;
 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 if ((rc = osd_attach_jinode(inode))) {
 418                 iput(inode);
 419                 inode = ERR_PTR(rc);
 420         } else {
 421                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 422                 if (id->oii_gen == OSD_OII_NOGEN)
 423                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 424
 425                 /* Do not update file c/mtime in ldiskfs.
 426                  * NB: we don't have any lock to protect this because we don't
 427                  * have reference on osd_object now, but contention with
 428                  * another lookup + attr_set can't happen in the tiny window
 429                  * between if (...) and set S_NOCMTIME. */
 430                 if (!(inode->i_flags & S_NOCMTIME))
 431                         inode->i_flags |= S_NOCMTIME;
 432         }
 433         return inode;
 434 }
 435
 436 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 437                           handle_t *handle, struct dentry *child,
 438                           struct inode *inode, struct htree_lock *hlock)
 439 {
 440         int rc, rc2;
 441
 442         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 443         if (rc == -ENOBUFS || rc == -ENOSPC) {
 444                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 445                 struct inode *parent = child->d_parent->d_inode;
 446                 struct lu_fid *fid = NULL;
 447
 448                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 449                 if (!rc2) {
 450                         fid = &loa->loa_lma.lma_self_fid;
 451                 } else if (rc2 == -ENODATA) {
 452                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 453                                 fid = &info->oti_fid3;
 454                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 455                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 456                                 fid = &info->oti_fid3;
 457                                 lu_igif_build(fid, parent->i_ino,
 458                                               parent->i_generation);
 459                         }
 460                 }
 461
 462                 if (fid != NULL)
 463                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 464                               "maximum entry limit\n",
 465                               osd_name(osd), parent->i_ino, PFID(fid),
 466                               rc == -ENOSPC ? "has reached" : "is approaching");
 467                 else
 468                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 469                               "maximum entry limit\n",
 470                               osd_name(osd), parent->i_ino,
 471                               rc == -ENOSPC ? "has reached" : "is approaching");
 472
 473                 /* ignore such error now */
 474                 if (rc == -ENOBUFS)
 475                         rc = 0;
 476         }
 477
 478         return rc;
 479 }
 480
 481
 482 static struct inode *
 483 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 484              struct osd_inode_id *id, struct lu_fid *fid)
 485 {
 486         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 487         struct inode *inode;
 488         int rc;
 489
 490         inode = osd_iget(info, dev, id);
 491         if (IS_ERR(inode))
 492                 return inode;
 493
 494         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 495         if (!rc) {
 496                 *fid = loa->loa_lma.lma_self_fid;
 497         } else if (rc == -ENODATA) {
 498                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 499                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 500                 else
 501                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 502         } else {
 503                 iput(inode);
 504                 inode = ERR_PTR(rc);
 505         }
 506         return inode;
 507 }
 508
 509 static struct inode *osd_iget_check(struct osd_thread_info *info,
 510                                     struct osd_device *dev,
 511                                     const struct lu_fid *fid,
 512                                     struct osd_inode_id *id,
 513                                     bool trusted)
 514 {
 515         struct inode *inode;
 516         int rc = 0;
 517         ENTRY;
 518
 519         /* The cached OI mapping is trustable. If we cannot locate the inode
 520          * via the cached OI mapping, then return the failure to the caller
 521          * directly without further OI checking. */
 522
 523 again:
 524         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 525         if (IS_ERR(inode)) {
 526                 rc = PTR_ERR(inode);
 527                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 528                         goto check_oi;
 529
 530                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 531                        PFID(fid), id->oii_ino, rc);
 532                 GOTO(put, rc);
 533         }
 534
 535         if (is_bad_inode(inode)) {
 536                 rc = -ENOENT;
 537                 if (!trusted)
 538                         goto check_oi;
 539
 540                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 541                        PFID(fid), id->oii_ino);
 542                 GOTO(put, rc);
 543         }
 544
 545         if (id->oii_gen != OSD_OII_NOGEN &&
 546             inode->i_generation != id->oii_gen) {
 547                 rc = -ESTALE;
 548                 if (!trusted)
 549                         goto check_oi;
 550
 551                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 552                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 553                        id->oii_ino, id->oii_gen, inode->i_generation);
 554                 GOTO(put, rc);
 555         }
 556
 557         if (inode->i_nlink == 0) {
 558                 rc = -ENOENT;
 559                 if (!trusted)
 560                         goto check_oi;
 561
 562                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 563                        PFID(fid), id->oii_ino);
 564                 GOTO(put, rc);
 565         }
 566
 567         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 568
 569 check_oi:
 570         if (rc != 0) {
 571                 __u32 saved_ino = id->oii_ino;
 572                 __u32 saved_gen = id->oii_gen;
 573
 574                 LASSERT(!trusted);
 575                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 576
 577                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 578                 /* XXX: There are four possible cases:
 579                  *      1. rc = 0.
 580                  *         Backup/restore caused the OI invalid.
 581                  *      2. rc = 0.
 582                  *         Someone unlinked the object but NOT removed
 583                  *         the OI mapping, such as mount target device
 584                  *         as ldiskfs, and modify something directly.
 585                  *      3. rc = -ENOENT.
 586                  *         Someone just removed the object between the
 587                  *         former oi_lookup and the iget. It is normal.
 588                  *      4. Other failure cases.
 589                  *
 590                  *      Generally, when the device is mounted, it will
 591                  *      auto check whether the system is restored from
 592                  *      file-level backup or not. We trust such detect
 593                  *      to distinguish the 1st case from the 2nd case:
 594                  *      if the OI files are consistent but may contain
 595                  *      stale OI mappings because of case 2, if iget()
 596                  *      returns -ENOENT or -ESTALE, then it should be
 597                  *      the case 2. */
 598                 if (rc != 0)
 599                         /* If the OI mapping was in OI file before the
 600                          * osd_iget_check(), but now, it is disappear,
 601                          * then it must be removed by race. That is a
 602                          * normal race case. */
 603                         GOTO(put, rc);
 604
 605                 /* It is the OI scrub updated the OI mapping by race.
 606                  * The new OI mapping must be valid. */
 607                 if (saved_ino != id->oii_ino || saved_gen != id->oii_gen) {
 608                         trusted = true;
 609                         goto again;
 610                 }
 611
 612                 if (IS_ERR(inode)) {
 613                         if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
 614                                 /* It still can be the case 2, but we cannot
 615                                  * distinguish it from the case 1. So return
 616                                  * -EREMCHG to block current operation until
 617                                  *  OI scrub rebuilt the OI mappings. */
 618                                 rc = -EREMCHG;
 619                         else
 620                                 rc = -ENOENT;
 621
 622                         GOTO(put, rc);
 623                 }
 624
 625                 if (inode->i_generation == id->oii_gen)
 626                         rc = -ENOENT;
 627                 else
 628                         rc = -EREMCHG;
 629         } else {
 630                 if (id->oii_gen == OSD_OII_NOGEN)
 631                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 632
 633                 /* Do not update file c/mtime in ldiskfs.
 634                  * NB: we don't have any lock to protect this because we don't
 635                  * have reference on osd_object now, but contention with
 636                  * another lookup + attr_set can't happen in the tiny window
 637                  * between if (...) and set S_NOCMTIME. */
 638                 if (!(inode->i_flags & S_NOCMTIME))
 639                         inode->i_flags |= S_NOCMTIME;
 640         }
 641
 642         GOTO(put, rc);
 643
 644 put:
 645         if (rc != 0) {
 646                 if (!IS_ERR(inode))
 647                         iput(inode);
 648
 649                 inode = ERR_PTR(rc);
 650         }
 651
 652         return inode;
 653 }
 654
 655 /**
 656  * \retval +v: new filter_fid, does not contain self-fid
 657  * \retval 0:  filter_fid_old, contains self-fid
 658  * \retval -v: other failure cases
 659  */
 660 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 661                  struct dentry *dentry, struct lu_fid *fid)
 662 {
 663         struct filter_fid_old   *ff     = &info->oti_ff;
 664         struct ost_id           *ostid  = &info->oti_ostid;
 665         int                      rc;
 666
 667         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 668         if (rc == sizeof(*ff)) {
 669                 rc = 0;
 670                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 671                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 672                 /*
 673                  * XXX: use 0 as the index for compatibility, the caller will
 674                  *      handle index related issues when necessary.
 675                  */
 676                 if (!rc)
 677                         ostid_to_fid(fid, ostid, 0);
 678         } else if (rc == sizeof(struct filter_fid)) {
 679                 rc = 1;
 680         } else if (rc >= 0) {
 681                 rc = -EINVAL;
 682         }
 683
 684         return rc;
 685 }
 686
 687 static int osd_lma_self_repair(struct osd_thread_info *info,
 688                                struct osd_device *osd, struct inode *inode,
 689                                const struct lu_fid *fid, __u32 compat)
 690 {
 691         handle_t *jh;
 692         int       rc;
 693
 694         LASSERT(current->journal_info == NULL);
 695
 696         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 697                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 698         if (IS_ERR(jh)) {
 699                 rc = PTR_ERR(jh);
 700                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 701                       osd_name(osd), rc);
 702                 return rc;
 703         }
 704
 705         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 706         if (rc != 0)
 707                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 708                       osd_name(osd), rc);
 709         ldiskfs_journal_stop(jh);
 710         return rc;
 711 }
 712
 713 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 714 {
 715         struct osd_thread_info  *info   = osd_oti_get(env);
 716         struct osd_device       *osd    = osd_obj2dev(obj);
 717         struct lustre_ost_attrs *loa    = &info->oti_ost_attrs;
 718         struct lustre_mdt_attrs *lma    = &loa->loa_lma;
 719         struct inode            *inode  = obj->oo_inode;
 720         struct dentry           *dentry = &info->oti_obj_dentry;
 721         struct lu_fid           *fid    = NULL;
 722         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 723         int                      rc;
 724         ENTRY;
 725
 726         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 727                              (void *)loa, sizeof(*loa));
 728         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 729                 fid = &lma->lma_self_fid;
 730                 rc = osd_get_idif(info, inode, dentry, fid);
 731                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 732                         /* For the given OST-object, if it has neither LMA nor
 733                          * FID in XATTR_NAME_FID, then the given FID (which is
 734                          * contained in the @obj, from client RPC for locating
 735                          * the OST-object) is trusted. We use it to generate
 736                          * the LMA. */
 737                         osd_lma_self_repair(info, osd, inode, rfid,
 738                                             LMAC_FID_ON_OST);
 739                         RETURN(0);
 740                 }
 741         }
 742
 743         if (rc < 0)
 744                 RETURN(rc);
 745
 746         if (rc > 0) {
 747                 rc = 0;
 748                 lustre_lma_swab(lma);
 749                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 750                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 751                               S_ISREG(inode->i_mode)))) {
 752                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 753                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 754                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 755                               PFID(rfid), inode->i_ino);
 756                         rc = -EOPNOTSUPP;
 757                 } else {
 758                         fid = &lma->lma_self_fid;
 759                 }
 760         }
 761
 762         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 763                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 764                         struct ost_id   *oi   = &info->oti_ostid;
 765                         struct lu_fid   *fid1 = &info->oti_fid3;
 766                         __u32            idx  = fid_idif_ost_idx(rfid);
 767
 768                         /* For old IDIF, the OST index is not part of the IDIF,
 769                          * Means that different OSTs may have the same IDIFs.
 770                          * Under such case, we need to make some compatible
 771                          * check to make sure to trigger OI scrub properly. */
 772                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 773                                 /* Given @rfid is new, LMA is old. */
 774                                 fid_to_ostid(fid, oi);
 775                                 ostid_to_fid(fid1, oi, idx);
 776                                 if (lu_fid_eq(fid1, rfid)) {
 777                                         if (osd->od_index_in_idif)
 778                                                 osd_lma_self_repair(info, osd,
 779                                                         inode, rfid,
 780                                                         LMAC_FID_ON_OST);
 781                                         RETURN(0);
 782                                 }
 783                         }
 784                 }
 785
 786                 rc = -EREMCHG;
 787         }
 788
 789         RETURN(rc);
 790 }
 791
 792 struct osd_check_lmv_buf {
 793 #ifdef HAVE_DIR_CONTEXT
 794         /* please keep it as first member */
 795         struct dir_context       ctx;
 796 #endif
 797         struct osd_thread_info  *oclb_info;
 798         struct osd_device       *oclb_dev;
 799         struct osd_idmap_cache  *oclb_oic;
 800 };
 801
 802 /**
 803  * It is called internally by ->readdir() to filter out the
 804  * local slave object's FID of the striped directory.
 805  *
 806  * \retval      1 found the local slave's FID
 807  * \retval      0 continue to check next item
 808  * \retval      -ve for failure
 809  */
 810 #ifdef HAVE_FILLDIR_USE_CTX
 811 static int osd_stripe_dir_filldir(struct dir_context *buf,
 812 #else
 813 static int osd_stripe_dir_filldir(void *buf,
 814 #endif
 815                                   const char *name, int namelen,
 816                                   loff_t offset, __u64 ino, unsigned d_type)
 817 {
 818         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 819         struct osd_thread_info *oti = oclb->oclb_info;
 820         struct lu_fid *fid = &oti->oti_fid3;
 821         struct osd_inode_id *id = &oti->oti_id3;
 822         struct osd_device *dev = oclb->oclb_dev;
 823         struct osd_idmap_cache *oic = oclb->oclb_oic;
 824         struct inode *inode;
 825         int rc;
 826
 827         if (name[0] == '.')
 828                 return 0;
 829
 830         fid_zero(fid);
 831         sscanf(name + 1, SFID, RFID(fid));
 832         if (!fid_is_sane(fid))
 833                 return 0;
 834
 835         if (osd_remote_fid(oti->oti_env, dev, fid))
 836                 return 0;
 837
 838         osd_id_gen(id, ino, OSD_OII_NOGEN);
 839         inode = osd_iget(oti, dev, id);
 840         if (IS_ERR(inode))
 841                 return PTR_ERR(inode);
 842
 843         iput(inode);
 844         osd_add_oi_cache(oti, dev, id, fid);
 845         oic->oic_fid = *fid;
 846         oic->oic_lid = *id;
 847         oic->oic_dev = dev;
 848         rc = osd_oii_insert(dev, oic, true);
 849
 850         return rc == 0 ? 1 : rc;
 851 }
 852
 853 /* When lookup item under striped directory, we need to locate the master
 854  * MDT-object of the striped directory firstly, then the client will send
 855  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 856  * and the item's name. If the system is restored from MDT file level backup,
 857  * then before the OI scrub completely built the OI files, the OI mappings of
 858  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 859  * not a problem for the master MDT-object. Because when locate the master
 860  * MDT-object, we will do name based lookup (for the striped directory itself)
 861  * firstly, during such process we can setup the correct OI mapping for the
 862  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 863  * the client will not trigger name based lookup on the MDT to locate the slave
 864  * MDT-object before locating item under the striped directory, then when
 865  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 866  * is invalid and does not know what the right OI mapping is, then the MDT has
 867  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 868  * the OI file, related OI mapping is unknown yet, please try again later. And
 869  * then client will re-try the RPC again and again until related OI mapping has
 870  * been updated. That is quite inefficient.
 871  *
 872  * To resolve above trouble, we will handle it as the following two cases:
 873  *
 874  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 875  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 876  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 877  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 878  *    directly. Please check osd_fid_lookup().
 879  *
 880  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 881  *    Under such case, during lookup the master MDT-object, we will lookup the
 882  *    slave MDT-object via readdir against the master MDT-object, because the
 883  *    slave MDT-objects information are stored as sub-directories with the name
 884  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 885  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 886  *    the correct OI mapping for the slave MDT-object. */
 887 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 888                          struct inode *inode, struct osd_idmap_cache *oic)
 889 {
 890         struct lu_buf *buf = &oti->oti_big_buf;
 891         struct dentry *dentry = &oti->oti_obj_dentry;
 892         struct file *filp = &oti->oti_file;
 893         const struct file_operations *fops;
 894         struct lmv_mds_md_v1 *lmv1;
 895         struct osd_check_lmv_buf oclb = {
 896 #ifdef HAVE_DIR_CONTEXT
 897                 .ctx.actor = osd_stripe_dir_filldir,
 898 #endif
 899                 .oclb_info = oti,
 900                 .oclb_dev = dev,
 901                 .oclb_oic = oic
 902         };
 903         int rc = 0;
 904         ENTRY;
 905
 906 again:
 907         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 908                              buf->lb_len);
 909         if (rc == -ERANGE) {
 910                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 911                 if (rc > 0) {
 912                         lu_buf_realloc(buf, rc);
 913                         if (buf->lb_buf == NULL)
 914                                 GOTO(out, rc = -ENOMEM);
 915
 916                         goto again;
 917                 }
 918         }
 919
 920         if (unlikely(rc == 0 || rc == -ENODATA))
 921                 GOTO(out, rc = 0);
 922
 923         if (rc < 0)
 924                 GOTO(out, rc);
 925
 926         if (unlikely(buf->lb_buf == NULL)) {
 927                 lu_buf_realloc(buf, rc);
 928                 if (buf->lb_buf == NULL)
 929                         GOTO(out, rc = -ENOMEM);
 930
 931                 goto again;
 932         }
 933
 934         lmv1 = buf->lb_buf;
 935         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 936                 GOTO(out, rc = 0);
 937
 938         fops = inode->i_fop;
 939         dentry->d_inode = inode;
 940         dentry->d_sb = inode->i_sb;
 941         filp->f_pos = 0;
 942         filp->f_path.dentry = dentry;
 943         filp->f_mode = FMODE_64BITHASH;
 944         filp->f_mapping = inode->i_mapping;
 945         filp->f_op = fops;
 946         filp->private_data = NULL;
 947         set_file_inode(filp, inode);
 948
 949 #ifdef HAVE_DIR_CONTEXT
 950         oclb.ctx.pos = filp->f_pos;
 951         rc = fops->iterate(filp, &oclb.ctx);
 952         filp->f_pos = oclb.ctx.pos;
 953 #else
 954         rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
 955 #endif
 956         fops->release(inode, filp);
 957
 958 out:
 959         if (rc < 0)
 960                 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
 961                        DFID": rc = %d\n",
 962                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 963                        inode->i_ino, inode->i_generation,
 964                        PFID(&oic->oic_fid), rc);
 965         else
 966                 rc = 0;
 967
 968         RETURN(rc);
 969 }
 970
 971 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 972                           const struct lu_fid *fid,
 973                           const struct lu_object_conf *conf)
 974 {
 975         struct osd_thread_info *info;
 976         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
 977         struct osd_device *dev;
 978         struct osd_idmap_cache *oic;
 979         struct osd_inode_id *id;
 980         struct inode *inode = NULL;
 981         struct osd_scrub *scrub;
 982         struct scrub_file *sf;
 983         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
 984         __u32 saved_ino;
 985         __u32 saved_gen;
 986         int result = 0;
 987         int rc1 = 0;
 988         bool remote = false;
 989         bool trusted = true;
 990         bool updated = false;
 991         ENTRY;
 992
 993         LINVRNT(osd_invariant(obj));
 994         LASSERT(obj->oo_inode == NULL);
 995         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 996
 997         dev = osd_dev(ldev);
 998         scrub = &dev->od_scrub;
 999         sf = &scrub->os_file;
1000         info = osd_oti_get(env);
1001         LASSERT(info);
1002         oic = &info->oti_cache;
1003
1004         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1005                 RETURN(-ENOENT);
1006
1007         /* For the object is created as locking anchor, or for the object to
1008          * be created on disk. No need to osd_oi_lookup() at here because FID
1009          * shouldn't never be re-used, if it's really a duplicate FID from
1010          * unexpected reason, we should be able to detect it later by calling
1011          * do_create->osd_oi_insert(). */
1012         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
1013                 GOTO(out, result = 0);
1014
1015         /* Search order: 1. per-thread cache. */
1016         if (lu_fid_eq(fid, &oic->oic_fid) &&
1017             likely(oic->oic_dev == dev)) {
1018                 id = &oic->oic_lid;
1019                 goto iget;
1020         }
1021
1022         id = &info->oti_id;
1023         if (!list_empty(&scrub->os_inconsistent_items)) {
1024                 /* Search order: 2. OI scrub pending list. */
1025                 result = osd_oii_lookup(dev, fid, id);
1026                 if (result == 0)
1027                         goto iget;
1028         }
1029
1030         /* The OI mapping in the OI file can be updated by the OI scrub
1031          * when we locate the inode via FID. So it may be not trustable. */
1032         trusted = false;
1033
1034         /* Search order: 3. OI files. */
1035         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1036         if (result == -ENOENT) {
1037                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1038                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1039                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1040                                       sf->sf_oi_bitmap))
1041                         GOTO(out, result = 0);
1042
1043                 goto trigger;
1044         }
1045
1046         if (result != 0)
1047                 GOTO(out, result);
1048
1049 iget:
1050         inode = osd_iget_check(info, dev, fid, id, trusted);
1051         if (IS_ERR(inode)) {
1052                 result = PTR_ERR(inode);
1053                 if (result == -ENOENT || result == -ESTALE)
1054                         GOTO(out, result = 0);
1055
1056                 if (result == -EREMCHG) {
1057
1058 trigger:
1059                         /* We still have chance to get the valid inode: for the
1060                          * object which is referenced by remote name entry, the
1061                          * object on the local MDT will be linked under the dir
1062                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1063                          *
1064                          * We do not know whether the object for the given FID
1065                          * is referenced by some remote name entry or not, and
1066                          * especially for DNE II, a multiple-linked object may
1067                          * have many name entries reside on many MDTs.
1068                          *
1069                          * To simplify the operation, OSD will not distinguish
1070                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1071                          * only happened for the RPC from other MDT during the
1072                          * OI scrub, or for the client side RPC with FID only,
1073                          * such as FID to path, or from old connected client. */
1074                         if (!remote &&
1075                             !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1076                                 rc1 = osd_lookup_in_remote_parent(info, dev,
1077                                                                   fid, id);
1078                                 if (rc1 == 0) {
1079                                         remote = true;
1080                                         trusted = true;
1081                                         flags |= SS_AUTO_PARTIAL;
1082                                         flags &= ~SS_AUTO_FULL;
1083                                         goto iget;
1084                                 }
1085                         }
1086
1087                         if (thread_is_running(&scrub->os_thread)) {
1088                                 if (scrub->os_partial_scan &&
1089                                     !scrub->os_in_join) {
1090                                         goto join;
1091                                 } else {
1092                                         if (inode != NULL && !IS_ERR(inode)) {
1093                                                 LASSERT(remote);
1094
1095                                                 osd_add_oi_cache(info, dev, id,
1096                                                                  fid);
1097                                                 osd_oii_insert(dev, oic, true);
1098                                         } else {
1099                                                 result = -EINPROGRESS;
1100                                         }
1101                                 }
1102                         } else if (!dev->od_noscrub) {
1103
1104 join:
1105                                 rc1 = osd_scrub_start(dev, flags);
1106                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1107                                               "for the "DFID" with flags 0x%x,"
1108                                               " rc = %d\n", osd_name(dev),
1109                                               PFID(fid), flags, rc1);
1110                                 if (rc1 == 0 || rc1 == -EALREADY) {
1111                                         if (inode != NULL && !IS_ERR(inode)) {
1112                                                 LASSERT(remote);
1113
1114                                                 osd_add_oi_cache(info, dev, id,
1115                                                                  fid);
1116                                                 osd_oii_insert(dev, oic, true);
1117                                         } else {
1118                                                 result = -EINPROGRESS;
1119                                         }
1120                                 } else {
1121                                         result = -EREMCHG;
1122                                 }
1123                         } else {
1124                                 result = -EREMCHG;
1125                         }
1126                 }
1127
1128                 if (inode == NULL || IS_ERR(inode))
1129                         GOTO(out, result);
1130         } else if (remote) {
1131                 goto trigger;
1132         }
1133
1134         obj->oo_inode = inode;
1135         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1136
1137         result = osd_check_lma(env, obj);
1138         if (result == 0)
1139                 goto found;
1140
1141         LASSERTF(id->oii_ino == inode->i_ino &&
1142                  id->oii_gen == inode->i_generation,
1143                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1144                  PFID(fid), id->oii_ino, id->oii_gen,
1145                  inode->i_ino, inode->i_generation);
1146
1147         saved_ino = inode->i_ino;
1148         saved_gen = inode->i_generation;
1149
1150         if (unlikely(result == -ENODATA)) {
1151                 /* If the OI scrub updated the OI mapping by race, it
1152                  * must be valid. Trust the inode that has no LMA EA. */
1153                 if (updated)
1154                         goto found;
1155
1156                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1157                 if (result == 0) {
1158                         /* The OI mapping is still there, the inode is still
1159                          * valid. It is just becaues the inode has no LMA EA. */
1160                         if (saved_ino == id->oii_ino &&
1161                             saved_gen == id->oii_gen)
1162                                 goto found;
1163
1164                         /* It is the OI scrub updated the OI mapping by race.
1165                          * The new OI mapping must be valid. */
1166                         iput(inode);
1167                         inode = NULL;
1168                         obj->oo_inode = NULL;
1169                         trusted = true;
1170                         updated = true;
1171                         goto iget;
1172                 }
1173
1174                 /* "result == -ENOENT" means that the OI mappinghas been
1175                  * removed by race, so the inode belongs to other object.
1176                  *
1177                  * Others error can be returned  directly. */
1178                 if (result == -ENOENT) {
1179                         LASSERT(trusted);
1180
1181                         result = 0;
1182                 }
1183         }
1184
1185         iput(inode);
1186         inode = NULL;
1187         obj->oo_inode = NULL;
1188
1189         if (result != -EREMCHG)
1190                 GOTO(out, result);
1191
1192         LASSERT(!updated);
1193
1194         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1195         /* "result == -ENOENT" means the cached OI mapping has been removed
1196          * from the OI file by race, above inode belongs to other object. */
1197         if (result == -ENOENT) {
1198                 LASSERT(trusted);
1199
1200                 GOTO(out, result = 0);
1201         }
1202
1203         if (result != 0)
1204                 GOTO(out, result);
1205
1206         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen)
1207                 goto trigger;
1208
1209         /* It is the OI scrub updated the OI mapping by race.
1210          * The new OI mapping must be valid. */
1211         trusted = true;
1212         updated = true;
1213         goto iget;
1214
1215 found:
1216         obj->oo_compat_dot_created = 1;
1217         obj->oo_compat_dotdot_created = 1;
1218
1219         if (S_ISDIR(inode->i_mode) &&
1220             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1221                 osd_check_lmv(info, dev, inode, oic);
1222
1223         result = osd_attach_jinode(inode);
1224         if (result) {
1225                 obj->oo_inode = NULL;
1226                 iput(inode);
1227                 GOTO(out, result);
1228         }
1229
1230         if (!ldiskfs_pdo)
1231                 GOTO(out, result = 0);
1232
1233         LASSERT(obj->oo_hl_head == NULL);
1234         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1235         if (obj->oo_hl_head == NULL) {
1236                 obj->oo_inode = NULL;
1237                 iput(inode);
1238                 GOTO(out, result = -ENOMEM);
1239         }
1240         GOTO(out, result = 0);
1241
1242 out:
1243         if (result != 0 && trusted)
1244                 fid_zero(&oic->oic_fid);
1245
1246         LINVRNT(osd_invariant(obj));
1247         return result;
1248 }
1249
1250 /*
1251  * Concurrency: shouldn't matter.
1252  */
1253 static void osd_object_init0(struct osd_object *obj)
1254 {
1255         LASSERT(obj->oo_inode != NULL);
1256         obj->oo_dt.do_body_ops = &osd_body_ops;
1257         obj->oo_dt.do_lu.lo_header->loh_attr |=
1258                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1259 }
1260
1261 /*
1262  * Concurrency: no concurrent access is possible that early in object
1263  * life-cycle.
1264  */
1265 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1266                            const struct lu_object_conf *conf)
1267 {
1268         struct osd_object *obj = osd_obj(l);
1269         int result;
1270
1271         LINVRNT(osd_invariant(obj));
1272
1273         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1274                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1275                 l->lo_header->loh_attr |= LOHA_EXISTS;
1276                 return 0;
1277         }
1278
1279         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1280         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1281         if (result == 0 && obj->oo_inode != NULL) {
1282                 struct osd_thread_info *oti = osd_oti_get(env);
1283                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1284
1285                 osd_object_init0(obj);
1286                 result = osd_get_lma(oti, obj->oo_inode,
1287                                      &oti->oti_obj_dentry, loa);
1288                 if (!result) {
1289                         /* Convert LMAI flags to lustre LMA flags
1290                          * and cache it to oo_lma_flags */
1291                         obj->oo_lma_flags =
1292                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1293                 } else if (result == -ENODATA) {
1294                         result = 0;
1295                 }
1296         }
1297
1298         LINVRNT(osd_invariant(obj));
1299         return result;
1300 }
1301
1302 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1303  * The left part is for value, binary mode. */
1304 struct osd_xattr_entry {
1305         struct list_head        oxe_list;
1306         size_t                  oxe_len;
1307         size_t                  oxe_namelen;
1308         bool                    oxe_exist;
1309         struct rcu_head         oxe_rcu;
1310         char                    oxe_buf[0];
1311 };
1312
1313 static int osd_oxc_get(struct osd_object *obj, const char *name,
1314                        struct lu_buf *buf)
1315 {
1316         struct osd_xattr_entry *tmp;
1317         struct osd_xattr_entry *oxe = NULL;
1318         size_t namelen = strlen(name);
1319         int rc;
1320         ENTRY;
1321
1322         rcu_read_lock();
1323         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1324                 if (namelen == tmp->oxe_namelen &&
1325                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1326                         oxe = tmp;
1327                         break;
1328                 }
1329         }
1330
1331         if (oxe == NULL)
1332                 GOTO(out, rc = -ENOENT);
1333
1334         if (!oxe->oxe_exist)
1335                 GOTO(out, rc = -ENODATA);
1336
1337         /* vallen */
1338         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1339         LASSERT(rc > 0);
1340
1341         if (buf->lb_buf == NULL)
1342                 GOTO(out, rc);
1343
1344         if (buf->lb_len < rc)
1345                 GOTO(out, rc = -ERANGE);
1346
1347         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1348         EXIT;
1349 out:
1350         rcu_read_unlock();
1351
1352         return rc;
1353 }
1354
1355 static void osd_oxc_free(struct rcu_head *head)
1356 {
1357         struct osd_xattr_entry *oxe;
1358
1359         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1360         OBD_FREE(oxe, oxe->oxe_len);
1361 }
1362
1363 static void osd_oxc_add(struct osd_object *obj, const char *name,
1364                         const char *buf, int buflen)
1365 {
1366         struct osd_xattr_entry *oxe;
1367         struct osd_xattr_entry *old = NULL;
1368         struct osd_xattr_entry *tmp;
1369         size_t namelen = strlen(name);
1370         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1371
1372         OBD_ALLOC(oxe, len);
1373         if (oxe == NULL)
1374                 return;
1375
1376         INIT_LIST_HEAD(&oxe->oxe_list);
1377         oxe->oxe_len = len;
1378         oxe->oxe_namelen = namelen;
1379         memcpy(oxe->oxe_buf, name, namelen);
1380         if (buflen > 0) {
1381                 LASSERT(buf != NULL);
1382                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1383                 oxe->oxe_exist = true;
1384         } else {
1385                 oxe->oxe_exist = false;
1386         }
1387
1388         /* this should be rarely called, just remove old and add new */
1389         spin_lock(&obj->oo_guard);
1390         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1391                 if (namelen == tmp->oxe_namelen &&
1392                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1393                         old = tmp;
1394                         break;
1395                 }
1396         }
1397         if (old != NULL) {
1398                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1399                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1400         } else {
1401                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1402         }
1403         spin_unlock(&obj->oo_guard);
1404 }
1405
1406 static void osd_oxc_del(struct osd_object *obj, const char *name)
1407 {
1408         struct osd_xattr_entry *oxe;
1409         size_t namelen = strlen(name);
1410
1411         spin_lock(&obj->oo_guard);
1412         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1413                 if (namelen == oxe->oxe_namelen &&
1414                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1415                         list_del_rcu(&oxe->oxe_list);
1416                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1417                         break;
1418                 }
1419         }
1420         spin_unlock(&obj->oo_guard);
1421 }
1422
1423 static void osd_oxc_fini(struct osd_object *obj)
1424 {
1425         struct osd_xattr_entry *oxe, *next;
1426
1427         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1428                 list_del(&oxe->oxe_list);
1429                 OBD_FREE(oxe, oxe->oxe_len);
1430         }
1431 }
1432
1433 /*
1434  * Concurrency: no concurrent access is possible that late in object
1435  * life-cycle.
1436  */
1437 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1438 {
1439         struct osd_object *obj = osd_obj(l);
1440
1441         LINVRNT(osd_invariant(obj));
1442
1443         osd_oxc_fini(obj);
1444         dt_object_fini(&obj->oo_dt);
1445         if (obj->oo_hl_head != NULL)
1446                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1447         OBD_FREE_PTR(obj);
1448 }
1449
1450 /*
1451  * Concurrency: no concurrent access is possible that late in object
1452  * life-cycle.
1453  */
1454 static void osd_index_fini(struct osd_object *o)
1455 {
1456         struct iam_container *bag;
1457
1458         if (o->oo_dir != NULL) {
1459                 bag = &o->oo_dir->od_container;
1460                 if (o->oo_inode != NULL) {
1461                         if (bag->ic_object == o->oo_inode)
1462                                 iam_container_fini(bag);
1463                 }
1464                 OBD_FREE_PTR(o->oo_dir);
1465                 o->oo_dir = NULL;
1466         }
1467 }
1468
1469 /*
1470  * Concurrency: no concurrent access is possible that late in object
1471  * life-cycle (for all existing callers, that is. New callers have to provide
1472  * their own locking.)
1473  */
1474 static int osd_inode_unlinked(const struct inode *inode)
1475 {
1476         return inode->i_nlink == 0;
1477 }
1478
1479 enum {
1480         OSD_TXN_OI_DELETE_CREDITS    = 20,
1481         OSD_TXN_INODE_DELETE_CREDITS = 20
1482 };
1483
1484 /*
1485  * Journal
1486  */
1487
1488 #if OSD_THANDLE_STATS
1489 /**
1490  * Set time when the handle is allocated
1491  */
1492 static void osd_th_alloced(struct osd_thandle *oth)
1493 {
1494         oth->oth_alloced = ktime_get();
1495 }
1496
1497 /**
1498  * Set time when the handle started
1499  */
1500 static void osd_th_started(struct osd_thandle *oth)
1501 {
1502         oth->oth_started = ktime_get();
1503 }
1504
1505 /**
1506  * Check whether the we deal with this handle for too long.
1507  */
1508 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1509                                 ktime_t alloced, ktime_t started,
1510                                 ktime_t closed)
1511 {
1512         ktime_t now = ktime_get();
1513
1514         LASSERT(dev != NULL);
1515
1516         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1517                             ktime_us_delta(started, alloced));
1518         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1519                             ktime_us_delta(closed, started));
1520         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1521                             ktime_us_delta(now, closed));
1522
1523         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1524                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1525                       oth, now, alloced, started, closed);
1526                 libcfs_debug_dumpstack(NULL);
1527         }
1528 }
1529
1530 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1531 {                                                                       \
1532         ktime_t __closed = ktime_get();                                 \
1533         ktime_t __alloced = oth->oth_alloced;                           \
1534         ktime_t __started = oth->oth_started;                           \
1535                                                                         \
1536         expr;                                                           \
1537         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1538 }
1539
1540 #else /* OSD_THANDLE_STATS */
1541
1542 #define osd_th_alloced(h)                  do {} while(0)
1543 #define osd_th_started(h)                  do {} while(0)
1544 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1545
1546 #endif /* OSD_THANDLE_STATS */
1547
1548 /*
1549  * Concurrency: doesn't access mutable data.
1550  */
1551 static int osd_param_is_not_sane(const struct osd_device *dev,
1552                                  const struct thandle *th)
1553 {
1554         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1555
1556         return oh->ot_credits > osd_transaction_size(dev);
1557 }
1558
1559 /*
1560  * Concurrency: shouldn't matter.
1561  */
1562 static void osd_trans_commit_cb(struct super_block *sb,
1563                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1564 {
1565         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1566         struct thandle     *th  = &oh->ot_super;
1567         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1568         struct dt_txn_commit_cb *dcb, *tmp;
1569
1570         LASSERT(oh->ot_handle == NULL);
1571
1572         if (error)
1573                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1574
1575         dt_txn_hook_commit(th);
1576
1577         /* call per-transaction callbacks if any */
1578         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1579                                  dcb_linkage) {
1580                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1581                          "commit callback entry: magic=%x name='%s'\n",
1582                          dcb->dcb_magic, dcb->dcb_name);
1583                 list_del_init(&dcb->dcb_linkage);
1584                 dcb->dcb_func(NULL, th, dcb, error);
1585         }
1586
1587         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1588         lu_device_put(lud);
1589         th->th_dev = NULL;
1590
1591         lu_context_exit(&th->th_ctx);
1592         lu_context_fini(&th->th_ctx);
1593         OBD_FREE_PTR(oh);
1594 }
1595
1596 #ifndef HAVE_SB_START_WRITE
1597 # define sb_start_write(sb) do {} while (0)
1598 # define sb_end_write(sb) do {} while (0)
1599 #endif
1600
1601 static struct thandle *osd_trans_create(const struct lu_env *env,
1602                                         struct dt_device *d)
1603 {
1604         struct osd_thread_info  *oti = osd_oti_get(env);
1605         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1606         struct osd_thandle      *oh;
1607         struct thandle          *th;
1608         ENTRY;
1609
1610         if (d->dd_rdonly) {
1611                 CERROR("%s: someone try to start transaction under "
1612                        "readonly mode, should be disabled.\n",
1613                        osd_name(osd_dt_dev(d)));
1614                 dump_stack();
1615                 RETURN(ERR_PTR(-EROFS));
1616         }
1617
1618         /* on pending IO in this thread should left from prev. request */
1619         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1620
1621         sb_start_write(osd_sb(osd_dt_dev(d)));
1622
1623         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1624         if (oh != NULL) {
1625                 oh->ot_quota_trans = &oti->oti_quota_trans;
1626                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1627                 th = &oh->ot_super;
1628                 th->th_dev = d;
1629                 th->th_result = 0;
1630                 th->th_tags = LCT_TX_HANDLE;
1631                 oh->ot_credits = 0;
1632                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1633                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1634                 osd_th_alloced(oh);
1635
1636                 memset(oti->oti_declare_ops, 0,
1637                        sizeof(oti->oti_declare_ops));
1638                 memset(oti->oti_declare_ops_cred, 0,
1639                        sizeof(oti->oti_declare_ops_cred));
1640                 memset(oti->oti_declare_ops_used, 0,
1641                        sizeof(oti->oti_declare_ops_used));
1642         } else {
1643                 sb_end_write(osd_sb(osd_dt_dev(d)));
1644                 th = ERR_PTR(-ENOMEM);
1645         }
1646         RETURN(th);
1647 }
1648
1649 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1650 {
1651         struct osd_thread_info  *oti = osd_oti_get(env);
1652         struct osd_thandle      *oh;
1653
1654         oh = container_of0(th, struct osd_thandle, ot_super);
1655         LASSERT(oh != NULL);
1656
1657         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1658               oti->oti_declare_ops[OSD_OT_CREATE],
1659               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1660               oti->oti_declare_ops_used[OSD_OT_CREATE],
1661               oti->oti_declare_ops[OSD_OT_DESTROY],
1662               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1663               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1664         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1665               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1666               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1667               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1668               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1669               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1670               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1671         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1672               oti->oti_declare_ops[OSD_OT_WRITE],
1673               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1674               oti->oti_declare_ops_used[OSD_OT_WRITE],
1675               oti->oti_declare_ops[OSD_OT_PUNCH],
1676               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1677               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1678               oti->oti_declare_ops[OSD_OT_QUOTA],
1679               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1680               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1681         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1682               oti->oti_declare_ops[OSD_OT_INSERT],
1683               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1684               oti->oti_declare_ops_used[OSD_OT_INSERT],
1685               oti->oti_declare_ops[OSD_OT_DELETE],
1686               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1687               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1688         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1689               oti->oti_declare_ops[OSD_OT_REF_ADD],
1690               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1691               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1692               oti->oti_declare_ops[OSD_OT_REF_DEL],
1693               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1694               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1695 }
1696
1697 /*
1698  * Concurrency: shouldn't matter.
1699  */
1700 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1701                            struct thandle *th)
1702 {
1703         struct osd_thread_info *oti = osd_oti_get(env);
1704         struct osd_device  *dev = osd_dt_dev(d);
1705         handle_t           *jh;
1706         struct osd_thandle *oh;
1707         int rc;
1708
1709         ENTRY;
1710
1711         LASSERT(current->journal_info == NULL);
1712
1713         oh = container_of0(th, struct osd_thandle, ot_super);
1714         LASSERT(oh != NULL);
1715         LASSERT(oh->ot_handle == NULL);
1716
1717         rc = dt_txn_hook_start(env, d, th);
1718         if (rc != 0)
1719                 GOTO(out, rc);
1720
1721         if (unlikely(osd_param_is_not_sane(dev, th))) {
1722                 static unsigned long last_printed;
1723                 static int last_credits;
1724
1725                 /* don't make noise on a tiny testing systems
1726                  * actual credits misuse will be caught anyway */
1727                 if (last_credits != oh->ot_credits &&
1728                     time_after(jiffies, last_printed +
1729                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1730                     osd_transaction_size(dev) > 512) {
1731                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1732                               oh->ot_credits, osd_transaction_size(dev));
1733                         osd_trans_dump_creds(env, th);
1734                         libcfs_debug_dumpstack(NULL);
1735                         last_credits = oh->ot_credits;
1736                         last_printed = jiffies;
1737                 }
1738                 /* XXX Limit the credits to 'max_transaction_buffers', and
1739                  *     let the underlying filesystem to catch the error if
1740                  *     we really need so many credits.
1741                  *
1742                  *     This should be removed when we can calculate the
1743                  *     credits precisely. */
1744                 oh->ot_credits = osd_transaction_size(dev);
1745         } else if (ldiskfs_track_declares_assert != 0) {
1746                 /* reserve few credits to prevent an assertion in JBD
1747                  * our debugging mechanism will be able to detected
1748                  * overuse. this can help to debug single-update
1749                  * transactions */
1750                 oh->ot_credits += 10;
1751                 if (unlikely(osd_param_is_not_sane(dev, th)))
1752                         oh->ot_credits = osd_transaction_size(dev);
1753         }
1754
1755         /*
1756          * XXX temporary stuff. Some abstraction layer should
1757          * be used.
1758          */
1759         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1760         osd_th_started(oh);
1761         if (!IS_ERR(jh)) {
1762                 oh->ot_handle = jh;
1763                 LASSERT(oti->oti_txns == 0);
1764                 lu_context_init(&th->th_ctx, th->th_tags);
1765                 lu_context_enter(&th->th_ctx);
1766
1767                 lu_device_get(&d->dd_lu_dev);
1768                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1769                               "osd-tx", th);
1770                 oti->oti_txns++;
1771                 rc = 0;
1772         } else {
1773                 rc = PTR_ERR(jh);
1774         }
1775 out:
1776         RETURN(rc);
1777 }
1778
1779 static int osd_seq_exists(const struct lu_env *env,
1780                           struct osd_device *osd, u64 seq)
1781 {
1782         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1783         struct seq_server_site  *ss = osd_seq_site(osd);
1784         int                     rc;
1785         ENTRY;
1786
1787         LASSERT(ss != NULL);
1788         LASSERT(ss->ss_server_fld != NULL);
1789
1790         rc = osd_fld_lookup(env, osd, seq, range);
1791         if (rc != 0) {
1792                 if (rc != -ENOENT)
1793                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1794                                osd_name(osd), seq, rc);
1795                 RETURN(0);
1796         }
1797
1798         RETURN(ss->ss_node_id == range->lsr_index);
1799 }
1800
1801 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1802 {
1803         struct dt_txn_commit_cb *dcb;
1804         struct dt_txn_commit_cb *tmp;
1805
1806         /* call per-transaction stop callbacks if any */
1807         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1808                                  dcb_linkage) {
1809                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1810                          "commit callback entry: magic=%x name='%s'\n",
1811                          dcb->dcb_magic, dcb->dcb_name);
1812                 list_del_init(&dcb->dcb_linkage);
1813                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1814         }
1815 }
1816
1817 /*
1818  * Concurrency: shouldn't matter.
1819  */
1820 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1821                           struct thandle *th)
1822 {
1823         int                     rc = 0, remove_agents = 0;
1824         struct osd_thandle     *oh;
1825         struct osd_thread_info *oti = osd_oti_get(env);
1826         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1827         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1828         struct qsd_instance    *qsd = osd->od_quota_slave;
1829         struct lquota_trans    *qtrans;
1830         ENTRY;
1831
1832         oh = container_of0(th, struct osd_thandle, ot_super);
1833
1834         /* reset OI cache for safety */
1835         oti->oti_ins_cache_used = 0;
1836
1837         remove_agents = oh->ot_remove_agents;
1838
1839         qtrans = oh->ot_quota_trans;
1840         oh->ot_quota_trans = NULL;
1841
1842         if (oh->ot_handle != NULL) {
1843                 int rc2;
1844                 handle_t *hdl = oh->ot_handle;
1845
1846                 /*
1847                  * add commit callback
1848                  * notice we don't do this in osd_trans_start()
1849                  * as underlying transaction can change during truncate
1850                  */
1851                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1852                                          &oh->ot_jcb);
1853
1854                 LASSERT(oti->oti_txns == 1);
1855                 oti->oti_txns--;
1856
1857                 rc = dt_txn_hook_stop(env, th);
1858                 if (rc != 0)
1859                         CERROR("%s: failed in transaction hook: rc = %d\n",
1860                                osd_name(osd), rc);
1861
1862                 osd_trans_stop_cb(oh, rc);
1863                 /* hook functions might modify th_sync */
1864                 hdl->h_sync = th->th_sync;
1865
1866                 oh->ot_handle = NULL;
1867                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1868                 if (rc2 != 0)
1869                         CERROR("%s: failed to stop transaction: rc = %d\n",
1870                                osd_name(osd), rc2);
1871                 if (!rc)
1872                         rc = rc2;
1873         } else {
1874                 osd_trans_stop_cb(oh, th->th_result);
1875                 OBD_FREE_PTR(oh);
1876         }
1877
1878         /* inform the quota slave device that the transaction is stopping */
1879         qsd_op_end(env, qsd, qtrans);
1880
1881         /* as we want IO to journal and data IO be concurrent, we don't block
1882          * awaiting data IO completion in osd_do_bio(), instead we wait here
1883          * once transaction is submitted to the journal. all reqular requests
1884          * don't do direct IO (except read/write), thus this wait_event becomes
1885          * no-op for them.
1886          *
1887          * IMPORTANT: we have to wait till any IO submited by the thread is
1888          * completed otherwise iobuf may be corrupted by different request
1889          */
1890         wait_event(iobuf->dr_wait,
1891                        atomic_read(&iobuf->dr_numreqs) == 0);
1892         osd_fini_iobuf(osd, iobuf);
1893         if (!rc)
1894                 rc = iobuf->dr_error;
1895
1896         if (unlikely(remove_agents != 0))
1897                 osd_process_scheduled_agent_removals(env, osd);
1898
1899         sb_end_write(osd_sb(osd));
1900
1901         RETURN(rc);
1902 }
1903
1904 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1905 {
1906         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1907                                                ot_super);
1908
1909         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1910         LASSERT(&dcb->dcb_func != NULL);
1911         if (dcb->dcb_flags & DCB_TRANS_STOP)
1912                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1913         else
1914                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1915
1916         return 0;
1917 }
1918
1919 /*
1920  * Called just before object is freed. Releases all resources except for
1921  * object itself (that is released by osd_object_free()).
1922  *
1923  * Concurrency: no concurrent access is possible that late in object
1924  * life-cycle.
1925  */
1926 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1927 {
1928         struct osd_object *obj   = osd_obj(l);
1929         struct inode      *inode = obj->oo_inode;
1930
1931         LINVRNT(osd_invariant(obj));
1932
1933         /*
1934          * If object is unlinked remove fid->ino mapping from object index.
1935          */
1936
1937         osd_index_fini(obj);
1938         if (inode != NULL) {
1939                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1940                 qid_t                    uid = i_uid_read(inode);
1941                 qid_t                    gid = i_gid_read(inode);
1942
1943                 iput(inode);
1944                 obj->oo_inode = NULL;
1945
1946                 if (qsd != NULL) {
1947                         struct osd_thread_info  *info = osd_oti_get(env);
1948                         struct lquota_id_info   *qi = &info->oti_qi;
1949
1950                         /* Release granted quota to master if necessary */
1951                         qi->lqi_id.qid_uid = uid;
1952                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1953
1954                         qi->lqi_id.qid_uid = gid;
1955                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1956
1957                         qi->lqi_id.qid_uid = i_projid_read(inode);
1958                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1959                 }
1960         }
1961 }
1962
1963 /*
1964  * Concurrency: ->loo_object_release() is called under site spin-lock.
1965  */
1966 static void osd_object_release(const struct lu_env *env,
1967                                struct lu_object *l)
1968 {
1969         struct osd_object *o = osd_obj(l);
1970         /* nobody should be releasing a non-destroyed object with nlink=0
1971          * the API allows this, but ldiskfs doesn't like and then report
1972          * this inode as deleted */
1973         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1974                 LBUG();
1975 }
1976
1977 /*
1978  * Concurrency: shouldn't matter.
1979  */
1980 static int osd_object_print(const struct lu_env *env, void *cookie,
1981                             lu_printer_t p, const struct lu_object *l)
1982 {
1983         struct osd_object *o = osd_obj(l);
1984         struct iam_descr  *d;
1985
1986         if (o->oo_dir != NULL)
1987                 d = o->oo_dir->od_container.ic_descr;
1988         else
1989                 d = NULL;
1990         return (*p)(env, cookie,
1991                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1992                     o, o->oo_inode,
1993                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1994                     o->oo_inode ? o->oo_inode->i_generation : 0,
1995                     d ? d->id_ops->id_name : "plain");
1996 }
1997
1998 /*
1999  * Concurrency: shouldn't matter.
2000  */
2001 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2002                struct obd_statfs *sfs)
2003 {
2004         struct osd_device       *osd = osd_dt_dev(d);
2005         struct super_block      *sb = osd_sb(osd);
2006         struct kstatfs          *ksfs;
2007         __u64                    reserved;
2008         int                      result = 0;
2009
2010         if (unlikely(osd->od_mnt == NULL))
2011                 return -EINPROGRESS;
2012
2013         /* osd_lproc.c call this without env, allocate ksfs for that case */
2014         if (unlikely(env == NULL)) {
2015                 OBD_ALLOC_PTR(ksfs);
2016                 if (ksfs == NULL)
2017                         return -ENOMEM;
2018         } else {
2019                 ksfs = &osd_oti_get(env)->oti_ksfs;
2020         }
2021
2022         result = sb->s_op->statfs(sb->s_root, ksfs);
2023         if (result)
2024                 goto out;
2025
2026         statfs_pack(sfs, ksfs);
2027         if (unlikely(sb->s_flags & MS_RDONLY))
2028                 sfs->os_state |= OS_STATE_READONLY;
2029         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2030                                          LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2031                 sfs->os_maxbytes = sb->s_maxbytes;
2032         else
2033                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2034
2035         /*
2036          * Reserve some space so to avoid fragmenting the filesystem too much.
2037          * Fragmentation not only impacts performance, but can also increase
2038          * metadata overhead significantly, causing grant calculation to be
2039          * wrong.
2040          *
2041          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2042          */
2043         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2044         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2045         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2046                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2047
2048         sfs->os_blocks -= reserved;
2049         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2050         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2051
2052 out:
2053         if (unlikely(env == NULL))
2054                 OBD_FREE_PTR(ksfs);
2055         return result;
2056 }
2057
2058 /**
2059  * Estimate space needed for file creations. We assume the largest filename
2060  * which is 2^64 - 1, hence a filename of 20 chars.
2061  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2062  */
2063 #ifdef __LDISKFS_DIR_REC_LEN
2064 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2065 #else
2066 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2067 #endif
2068
2069 /*
2070  * Concurrency: doesn't access mutable data.
2071  */
2072 static void osd_conf_get(const struct lu_env *env,
2073                          const struct dt_device *dev,
2074                          struct dt_device_param *param)
2075 {
2076         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2077         int                ea_overhead;
2078
2079         /*
2080          * XXX should be taken from not-yet-existing fs abstraction layer.
2081          */
2082         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2083         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2084         param->ddp_symlink_max  = sb->s_blocksize;
2085         param->ddp_mount_type     = LDD_MT_LDISKFS;
2086         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2087                 param->ddp_maxbytes = sb->s_maxbytes;
2088         else
2089                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2090         /* inode are statically allocated, so per-inode space consumption
2091          * is the space consumed by the directory entry */
2092         param->ddp_inodespace     = PER_OBJ_USAGE;
2093         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2094          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2095          * maximum length to not under count the actual number of extents
2096          * needed for writing a file. */
2097         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2098         /* worst-case extent insertion metadata overhead */
2099         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2100         param->ddp_mntopts      = 0;
2101         if (test_opt(sb, XATTR_USER))
2102                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2103         if (test_opt(sb, POSIX_ACL))
2104                 param->ddp_mntopts |= MNTOPT_ACL;
2105
2106         /* LOD might calculate the max stripe count based on max_ea_size,
2107          * so we need take account in the overhead as well,
2108          * xattr_header + magic + xattr_entry_head */
2109         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2110                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2111
2112 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2113         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2114                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2115                                                                 ea_overhead;
2116         else
2117 #endif
2118                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2119 }
2120
2121 /*
2122  * Concurrency: shouldn't matter.
2123  */
2124 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2125 {
2126         int rc;
2127
2128         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2129
2130         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2131
2132         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2133                rc);
2134
2135         return rc;
2136 }
2137
2138 /**
2139  * Start commit for OSD device.
2140  *
2141  * An implementation of dt_commit_async method for OSD device.
2142  * Asychronously starts underlayng fs sync and thereby a transaction
2143  * commit.
2144  *
2145  * \param env environment
2146  * \param d dt device
2147  *
2148  * \see dt_device_operations
2149  */
2150 static int osd_commit_async(const struct lu_env *env,
2151                             struct dt_device *d)
2152 {
2153         struct super_block *s = osd_sb(osd_dt_dev(d));
2154         ENTRY;
2155
2156         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2157         RETURN(s->s_op->sync_fs(s, 0));
2158 }
2159
2160 /*
2161  * Concurrency: shouldn't matter.
2162  */
2163
2164 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2165 {
2166         struct super_block *sb = osd_sb(osd_dt_dev(d));
2167         struct block_device *dev = sb->s_bdev;
2168 #ifdef HAVE_DEV_SET_RDONLY
2169         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2170         int rc = 0;
2171 #else
2172         int rc = -EOPNOTSUPP;
2173 #endif
2174         ENTRY;
2175
2176 #ifdef HAVE_DEV_SET_RDONLY
2177         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2178
2179         if (sb->s_op->freeze_fs) {
2180                 rc = sb->s_op->freeze_fs(sb);
2181                 if (rc)
2182                         goto out;
2183         }
2184
2185         if (jdev && (jdev != dev)) {
2186                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2187                        (long)jdev);
2188                 dev_set_rdonly(jdev);
2189         }
2190         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2191         dev_set_rdonly(dev);
2192
2193         if (sb->s_op->unfreeze_fs)
2194                 sb->s_op->unfreeze_fs(sb);
2195
2196 out:
2197 #endif
2198         if (rc)
2199                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2200                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2201
2202         RETURN(rc);
2203 }
2204
2205 /**
2206  * Note: we do not count into QUOTA here.
2207  * If we mount with --data_journal we may need more.
2208  */
2209 const int osd_dto_credits_noquota[DTO_NR] = {
2210         /**
2211          * Insert.
2212          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2213          * SINGLEDATA_TRANS_BLOCKS(8)
2214          * XXX Note: maybe iam need more, since iam have more level than
2215          *           EXT3 htree.
2216          */
2217         [DTO_INDEX_INSERT]  = 16,
2218         /**
2219          * Delete
2220          * just modify a single entry, probably merge few within a block
2221          */
2222         [DTO_INDEX_DELETE]  = 1,
2223         /**
2224          * Used for OI scrub
2225          */
2226         [DTO_INDEX_UPDATE]  = 16,
2227         /**
2228          * 4(inode, inode bits, groups, GDT)
2229          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2230          */
2231         [DTO_OBJECT_CREATE] = 4,
2232         /**
2233          * 4(inode, inode bits, groups, GDT)
2234          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2235          */
2236         [DTO_OBJECT_DELETE] = 4,
2237         /**
2238          * Attr set credits (inode)
2239          */
2240         [DTO_ATTR_SET_BASE] = 1,
2241         /**
2242          * Xattr set. The same as xattr of EXT3.
2243          * DATA_TRANS_BLOCKS(14)
2244          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2245          * are also counted in. Do not know why?
2246          */
2247         [DTO_XATTR_SET]     = 14,
2248         /**
2249          * credits for inode change during write.
2250          */
2251         [DTO_WRITE_BASE]    = 3,
2252         /**
2253          * credits for single block write.
2254          */
2255         [DTO_WRITE_BLOCK]   = 14,
2256         /**
2257          * Attr set credits for chown.
2258          * This is extra credits for setattr, and it is null without quota
2259          */
2260         [DTO_ATTR_SET_CHOWN] = 0
2261 };
2262
2263 static const struct dt_device_operations osd_dt_ops = {
2264         .dt_root_get       = osd_root_get,
2265         .dt_statfs         = osd_statfs,
2266         .dt_trans_create   = osd_trans_create,
2267         .dt_trans_start    = osd_trans_start,
2268         .dt_trans_stop     = osd_trans_stop,
2269         .dt_trans_cb_add   = osd_trans_cb_add,
2270         .dt_conf_get       = osd_conf_get,
2271         .dt_sync           = osd_sync,
2272         .dt_ro             = osd_ro,
2273         .dt_commit_async   = osd_commit_async,
2274 };
2275
2276 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2277                           unsigned role)
2278 {
2279         struct osd_object *obj = osd_dt_obj(dt);
2280         struct osd_thread_info *oti = osd_oti_get(env);
2281
2282         LINVRNT(osd_invariant(obj));
2283
2284         LASSERT(obj->oo_owner != env);
2285         down_read_nested(&obj->oo_sem, role);
2286
2287         LASSERT(obj->oo_owner == NULL);
2288         oti->oti_r_locks++;
2289 }
2290
2291 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2292                            unsigned role)
2293 {
2294         struct osd_object *obj = osd_dt_obj(dt);
2295         struct osd_thread_info *oti = osd_oti_get(env);
2296
2297         LINVRNT(osd_invariant(obj));
2298
2299         LASSERT(obj->oo_owner != env);
2300         down_write_nested(&obj->oo_sem, role);
2301
2302         LASSERT(obj->oo_owner == NULL);
2303         obj->oo_owner = env;
2304         oti->oti_w_locks++;
2305 }
2306
2307 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2308 {
2309         struct osd_object *obj = osd_dt_obj(dt);
2310         struct osd_thread_info *oti = osd_oti_get(env);
2311
2312         LINVRNT(osd_invariant(obj));
2313
2314         LASSERT(oti->oti_r_locks > 0);
2315         oti->oti_r_locks--;
2316         up_read(&obj->oo_sem);
2317 }
2318
2319 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2320 {
2321         struct osd_object *obj = osd_dt_obj(dt);
2322         struct osd_thread_info *oti = osd_oti_get(env);
2323
2324         LINVRNT(osd_invariant(obj));
2325
2326         LASSERT(obj->oo_owner == env);
2327         LASSERT(oti->oti_w_locks > 0);
2328         oti->oti_w_locks--;
2329         obj->oo_owner = NULL;
2330         up_write(&obj->oo_sem);
2331 }
2332
2333 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2334 {
2335         struct osd_object *obj = osd_dt_obj(dt);
2336
2337         LINVRNT(osd_invariant(obj));
2338
2339         return obj->oo_owner == env;
2340 }
2341
2342 static struct timespec *osd_inode_time(const struct lu_env *env,
2343                                        struct inode *inode, __u64 seconds)
2344 {
2345         struct osd_thread_info  *oti = osd_oti_get(env);
2346         struct timespec         *t   = &oti->oti_time;
2347
2348         t->tv_sec = seconds;
2349         t->tv_nsec = 0;
2350         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2351         return t;
2352 }
2353
2354 static void osd_inode_getattr(const struct lu_env *env,
2355                               struct inode *inode, struct lu_attr *attr)
2356 {
2357         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2358                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2359                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2360                            LA_BLKSIZE | LA_TYPE;
2361
2362         attr->la_atime   = LTIME_S(inode->i_atime);
2363         attr->la_mtime   = LTIME_S(inode->i_mtime);
2364         attr->la_ctime   = LTIME_S(inode->i_ctime);
2365         attr->la_mode    = inode->i_mode;
2366         attr->la_size    = i_size_read(inode);
2367         attr->la_blocks  = inode->i_blocks;
2368         attr->la_uid     = i_uid_read(inode);
2369         attr->la_gid     = i_gid_read(inode);
2370         attr->la_projid  = i_projid_read(inode);
2371         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2372         attr->la_nlink   = inode->i_nlink;
2373         attr->la_rdev    = inode->i_rdev;
2374         attr->la_blksize = 1 << inode->i_blkbits;
2375         attr->la_blkbits = inode->i_blkbits;
2376         /*
2377          * Ext4 did not transfer inherit flags from raw inode
2378          * to inode flags, and ext4 internally test raw inode
2379          * @i_flags directly. Instead of patching ext4, we do it here.
2380          */
2381         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2382                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2383 }
2384
2385 static int osd_attr_get(const struct lu_env *env,
2386                         struct dt_object *dt,
2387                         struct lu_attr *attr)
2388 {
2389         struct osd_object *obj = osd_dt_obj(dt);
2390
2391         if (unlikely(!dt_object_exists(dt)))
2392                 return -ENOENT;
2393         if (unlikely(obj->oo_destroyed))
2394                 return -ENOENT;
2395
2396         LASSERT(!dt_object_remote(dt));
2397         LINVRNT(osd_invariant(obj));
2398
2399         spin_lock(&obj->oo_guard);
2400         osd_inode_getattr(env, obj->oo_inode, attr);
2401         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2402                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2403         spin_unlock(&obj->oo_guard);
2404
2405         return 0;
2406 }
2407
2408 static int osd_declare_attr_qid(const struct lu_env *env,
2409                                 struct osd_object *obj,
2410                                 struct osd_thandle *oh, long long bspace,
2411                                 qid_t old_id, qid_t new_id, bool enforce,
2412                                 unsigned type)
2413 {
2414         int rc;
2415         struct osd_thread_info *info = osd_oti_get(env);
2416         struct lquota_id_info  *qi = &info->oti_qi;
2417
2418         qi->lqi_type = type;
2419         /* inode accounting */
2420         qi->lqi_is_blk = false;
2421
2422         /* one more inode for the new id ... */
2423         qi->lqi_id.qid_uid = new_id;
2424         qi->lqi_space      = 1;
2425         /* Reserve credits for the new id */
2426         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2427         if (rc == -EDQUOT || rc == -EINPROGRESS)
2428                 rc = 0;
2429         if (rc)
2430                 RETURN(rc);
2431
2432         /* and one less inode for the current id */
2433         qi->lqi_id.qid_uid = old_id;
2434         qi->lqi_space      = -1;
2435         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2436         if (rc == -EDQUOT || rc == -EINPROGRESS)
2437                 rc = 0;
2438         if (rc)
2439                 RETURN(rc);
2440
2441         /* block accounting */
2442         qi->lqi_is_blk = true;
2443
2444         /* more blocks for the new id ... */
2445         qi->lqi_id.qid_uid = new_id;
2446         qi->lqi_space      = bspace;
2447         /*
2448          * Credits for the new uid has been reserved, re-use "obj"
2449          * to save credit reservation.
2450          */
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         /* and finally less blocks for the current uid */
2458         qi->lqi_id.qid_uid = old_id;
2459         qi->lqi_space      = -bspace;
2460         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2461         if (rc == -EDQUOT || rc == -EINPROGRESS)
2462                 rc = 0;
2463
2464         RETURN(rc);
2465 }
2466
2467 static int osd_declare_attr_set(const struct lu_env *env,
2468                                 struct dt_object *dt,
2469                                 const struct lu_attr *attr,
2470                                 struct thandle *handle)
2471 {
2472         struct osd_thandle     *oh;
2473         struct osd_object      *obj;
2474         qid_t                   uid;
2475         qid_t                   gid;
2476         long long               bspace;
2477         int                     rc = 0;
2478         bool                    enforce;
2479         ENTRY;
2480
2481         LASSERT(dt != NULL);
2482         LASSERT(handle != NULL);
2483
2484         obj = osd_dt_obj(dt);
2485         LASSERT(osd_invariant(obj));
2486
2487         oh = container_of0(handle, struct osd_thandle, ot_super);
2488         LASSERT(oh->ot_handle == NULL);
2489
2490         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2491                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2492
2493         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2494                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2495
2496         if (attr == NULL || obj->oo_inode == NULL)
2497                 RETURN(rc);
2498
2499         bspace   = obj->oo_inode->i_blocks;
2500         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2501         bspace   = toqb(bspace);
2502
2503         /* Changing ownership is always preformed by super user, it should not
2504          * fail with EDQUOT.
2505          *
2506          * We still need to call the osd_declare_qid() to calculate the journal
2507          * credits for updating quota accounting files and to trigger quota
2508          * space adjustment once the operation is completed.*/
2509         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2510                 /* USERQUOTA */
2511                 uid = i_uid_read(obj->oo_inode);
2512                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2513                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2514                                           attr->la_uid, enforce, USRQUOTA);
2515                 if (rc)
2516                         RETURN(rc);
2517
2518                 gid = i_gid_read(obj->oo_inode);
2519                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2520                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2521                                           i_gid_read(obj->oo_inode),
2522                                           attr->la_gid, enforce, GRPQUOTA);
2523                 if (rc)
2524                         RETURN(rc);
2525
2526         }
2527 #ifdef HAVE_PROJECT_QUOTA
2528         if (attr->la_valid & LA_PROJID) {
2529                 __u32 projid = i_projid_read(obj->oo_inode);
2530                 enforce = (attr->la_valid & LA_PROJID) &&
2531                                         (attr->la_projid != projid);
2532                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2533                                           (qid_t)projid, (qid_t)attr->la_projid,
2534                                           enforce, PRJQUOTA);
2535                 if (rc)
2536                         RETURN(rc);
2537         }
2538 #endif
2539         RETURN(rc);
2540 }
2541
2542 static int osd_inode_setattr(const struct lu_env *env,
2543                              struct inode *inode, const struct lu_attr *attr)
2544 {
2545         __u64 bits = attr->la_valid;
2546
2547         /* Only allow set size for regular file */
2548         if (!S_ISREG(inode->i_mode))
2549                 bits &= ~(LA_SIZE | LA_BLOCKS);
2550
2551         if (bits == 0)
2552                 return 0;
2553
2554         if (bits & LA_ATIME)
2555                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2556         if (bits & LA_CTIME)
2557                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2558         if (bits & LA_MTIME)
2559                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2560         if (bits & LA_SIZE) {
2561                 spin_lock(&inode->i_lock);
2562                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2563                 i_size_write(inode, attr->la_size);
2564                 spin_unlock(&inode->i_lock);
2565         }
2566
2567         /* OSD should not change "i_blocks" which is used by quota.
2568          * "i_blocks" should be changed by ldiskfs only. */
2569         if (bits & LA_MODE)
2570                 inode->i_mode = (inode->i_mode & S_IFMT) |
2571                                 (attr->la_mode & ~S_IFMT);
2572         if (bits & LA_UID)
2573                 i_uid_write(inode, attr->la_uid);
2574         if (bits & LA_GID)
2575                 i_gid_write(inode, attr->la_gid);
2576         if (bits & LA_PROJID)
2577                 i_projid_write(inode, attr->la_projid);
2578         if (bits & LA_NLINK)
2579                 set_nlink(inode, attr->la_nlink);
2580         if (bits & LA_RDEV)
2581                 inode->i_rdev = attr->la_rdev;
2582
2583         if (bits & LA_FLAGS) {
2584                 /* always keep S_NOCMTIME */
2585                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2586                                  S_NOCMTIME;
2587                 /*
2588                  * Ext4 did not transfer inherit flags from
2589                  * @inode->i_flags to raw inode i_flags when writing
2590                  * flags, we do it explictly here.
2591                  */
2592                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2593                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2594         }
2595         return 0;
2596 }
2597
2598 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2599 {
2600         int rc;
2601
2602         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2603             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2604                 struct iattr    iattr;
2605
2606                 ll_vfs_dq_init(inode);
2607                 iattr.ia_valid = 0;
2608                 if (attr->la_valid & LA_UID)
2609                         iattr.ia_valid |= ATTR_UID;
2610                 if (attr->la_valid & LA_GID)
2611                         iattr.ia_valid |= ATTR_GID;
2612                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2613                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2614
2615                 rc = ll_vfs_dq_transfer(inode, &iattr);
2616                 if (rc) {
2617                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2618                                "enforcement enabled on the ldiskfs "
2619                                "filesystem?\n", inode->i_sb->s_id, rc);
2620                         return rc;
2621                 }
2622         }
2623
2624 #ifdef HAVE_PROJECT_QUOTA
2625         /* Handle project id transfer here properly */
2626         if (attr->la_valid & LA_PROJID &&
2627             attr->la_projid != i_projid_read(inode)) {
2628                 rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid);
2629                 if (rc) {
2630                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2631                                "enforcement enabled on the ldiskfs "
2632                                "filesystem?\n", inode->i_sb->s_id, rc);
2633                         return rc;
2634                 }
2635         }
2636 #endif
2637         return 0;
2638 }
2639
2640 static int osd_attr_set(const struct lu_env *env,
2641                         struct dt_object *dt,
2642                         const struct lu_attr *attr,
2643                         struct thandle *handle)
2644 {
2645         struct osd_object *obj = osd_dt_obj(dt);
2646         struct inode      *inode;
2647         int rc;
2648
2649         if (!dt_object_exists(dt))
2650                 return -ENOENT;
2651
2652         LASSERT(handle != NULL);
2653         LASSERT(!dt_object_remote(dt));
2654         LASSERT(osd_invariant(obj));
2655
2656         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2657
2658         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2659                 struct osd_thread_info  *oti  = osd_oti_get(env);
2660                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2661                 struct lu_fid           *fid1 = &oti->oti_fid;
2662                 struct osd_inode_id     *id   = &oti->oti_id;
2663                 struct iam_path_descr   *ipd;
2664                 struct iam_container    *bag;
2665                 struct osd_thandle      *oh;
2666                 int                      rc;
2667
2668                 fid_cpu_to_be(fid1, fid0);
2669                 memset(id, 1, sizeof(*id));
2670                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2671                                   fid0)->oi_dir.od_container;
2672                 ipd = osd_idx_ipd_get(env, bag);
2673                 if (unlikely(ipd == NULL))
2674                         RETURN(-ENOMEM);
2675
2676                 oh = container_of0(handle, struct osd_thandle, ot_super);
2677                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2678                                 (const struct iam_rec *)id, ipd);
2679                 osd_ipd_put(env, bag, ipd);
2680                 return(rc > 0 ? 0 : rc);
2681         }
2682
2683         inode = obj->oo_inode;
2684
2685         rc = osd_quota_transfer(inode, attr);
2686         if (rc)
2687                 return rc;
2688
2689         spin_lock(&obj->oo_guard);
2690         rc = osd_inode_setattr(env, inode, attr);
2691         spin_unlock(&obj->oo_guard);
2692         if (rc != 0)
2693                 GOTO(out, rc);
2694
2695         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2696
2697         if (!(attr->la_valid & LA_FLAGS))
2698                 GOTO(out, rc);
2699
2700         /* Let's check if there are extra flags need to be set into LMA */
2701         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2702                 struct osd_thread_info *info = osd_oti_get(env);
2703                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2704
2705                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2706                                  &info->oti_ost_attrs);
2707                 if (rc)
2708                         GOTO(out, rc);
2709
2710                 lma->lma_incompat |=
2711                         lustre_to_lma_flags(attr->la_flags);
2712                 lustre_lma_swab(lma);
2713                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2714                                      lma, sizeof(*lma), XATTR_REPLACE);
2715                 if (rc != 0) {
2716                         struct osd_device *osd = osd_obj2dev(obj);
2717
2718                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2719                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2720                               lma->lma_incompat, rc);
2721                 } else {
2722                         obj->oo_lma_flags =
2723                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2724                 }
2725                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2726         }
2727 out:
2728         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2729
2730         return rc;
2731 }
2732
2733 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2734                                            struct osd_object *obj,
2735                                            const char *name, const int namelen)
2736 {
2737         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2738 }
2739
2740 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2741                       umode_t mode, struct dt_allocation_hint *hint,
2742                       struct thandle *th)
2743 {
2744         int result;
2745         struct osd_device  *osd = osd_obj2dev(obj);
2746         struct osd_thandle *oth;
2747         struct dt_object   *parent = NULL;
2748         struct inode       *inode;
2749
2750         LINVRNT(osd_invariant(obj));
2751         LASSERT(obj->oo_inode == NULL);
2752         LASSERT(obj->oo_hl_head == NULL);
2753
2754         if (S_ISDIR(mode) && ldiskfs_pdo) {
2755                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2756                 if (obj->oo_hl_head == NULL)
2757                         return -ENOMEM;
2758         }
2759
2760         oth = container_of(th, struct osd_thandle, ot_super);
2761         LASSERT(oth->ot_handle->h_transaction != NULL);
2762
2763         if (hint != NULL && hint->dah_parent != NULL &&
2764             !dt_object_remote(hint->dah_parent))
2765                 parent = hint->dah_parent;
2766
2767         inode = ldiskfs_create_inode(oth->ot_handle,
2768                                      parent ? osd_dt_obj(parent)->oo_inode :
2769                                               osd_sb(osd)->s_root->d_inode,
2770                                      mode);
2771         if (!IS_ERR(inode)) {
2772                 /* Do not update file c/mtime in ldiskfs. */
2773                 inode->i_flags |= S_NOCMTIME;
2774
2775                 /* For new created object, it must be consistent,
2776                  * and it is unnecessary to scrub against it. */
2777                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2778
2779                 obj->oo_inode = inode;
2780                 result = 0;
2781         } else {
2782                 if (obj->oo_hl_head != NULL) {
2783                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2784                         obj->oo_hl_head = NULL;
2785                 }
2786                 result = PTR_ERR(inode);
2787         }
2788         LINVRNT(osd_invariant(obj));
2789         return result;
2790 }
2791
2792 enum {
2793         OSD_NAME_LEN = 255
2794 };
2795
2796 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2797                      struct lu_attr *attr,
2798                      struct dt_allocation_hint *hint,
2799                      struct dt_object_format *dof,
2800                      struct thandle *th)
2801 {
2802         int result;
2803         struct osd_thandle *oth;
2804         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2805
2806         LASSERT(S_ISDIR(attr->la_mode));
2807
2808         oth = container_of(th, struct osd_thandle, ot_super);
2809         LASSERT(oth->ot_handle->h_transaction != NULL);
2810         result = osd_mkfile(info, obj, mode, hint, th);
2811
2812         return result;
2813 }
2814
2815 static int osd_mk_index(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         int result;
2822         struct osd_thandle *oth;
2823         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2824
2825         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2826
2827         LASSERT(S_ISREG(attr->la_mode));
2828
2829         oth = container_of(th, struct osd_thandle, ot_super);
2830         LASSERT(oth->ot_handle->h_transaction != NULL);
2831
2832         result = osd_mkfile(info, obj, mode, hint, th);
2833         if (result == 0) {
2834                 LASSERT(obj->oo_inode != NULL);
2835                 if (feat->dif_flags & DT_IND_VARKEY)
2836                         result = iam_lvar_create(obj->oo_inode,
2837                                                  feat->dif_keysize_max,
2838                                                  feat->dif_ptrsize,
2839                                                  feat->dif_recsize_max,
2840                                                  oth->ot_handle);
2841                 else
2842                         result = iam_lfix_create(obj->oo_inode,
2843                                                  feat->dif_keysize_max,
2844                                                  feat->dif_ptrsize,
2845                                                  feat->dif_recsize_max,
2846                                                  oth->ot_handle);
2847
2848         }
2849         return result;
2850 }
2851
2852 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2853                      struct lu_attr *attr,
2854                      struct dt_allocation_hint *hint,
2855                      struct dt_object_format *dof,
2856                      struct thandle *th)
2857 {
2858         LASSERT(S_ISREG(attr->la_mode));
2859         return osd_mkfile(info, obj, (attr->la_mode &
2860                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2861 }
2862
2863 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2864                      struct lu_attr *attr,
2865                      struct dt_allocation_hint *hint,
2866                      struct dt_object_format *dof,
2867                      struct thandle *th)
2868 {
2869         LASSERT(S_ISLNK(attr->la_mode));
2870         return osd_mkfile(info, obj, (attr->la_mode &
2871                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2872 }
2873
2874 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2875                      struct lu_attr *attr,
2876                      struct dt_allocation_hint *hint,
2877                      struct dt_object_format *dof,
2878                      struct thandle *th)
2879 {
2880         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2881         int result;
2882
2883         LINVRNT(osd_invariant(obj));
2884         LASSERT(obj->oo_inode == NULL);
2885         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2886                 S_ISFIFO(mode) || S_ISSOCK(mode));
2887
2888         result = osd_mkfile(info, obj, mode, hint, th);
2889         if (result == 0) {
2890                 LASSERT(obj->oo_inode != NULL);
2891                 /*
2892                  * This inode should be marked dirty for i_rdev.  Currently
2893                  * that is done in the osd_attr_init().
2894                  */
2895                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2896                                    attr->la_rdev);
2897         }
2898         LINVRNT(osd_invariant(obj));
2899         return result;
2900 }
2901
2902 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2903                               struct lu_attr *,
2904                               struct dt_allocation_hint *hint,
2905                               struct dt_object_format *dof,
2906                               struct thandle *);
2907
2908 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2909 {
2910         osd_obj_type_f result;
2911
2912         switch (type) {
2913         case DFT_DIR:
2914                 result = osd_mkdir;
2915                 break;
2916         case DFT_REGULAR:
2917                 result = osd_mkreg;
2918                 break;
2919         case DFT_SYM:
2920                 result = osd_mksym;
2921                 break;
2922         case DFT_NODE:
2923                 result = osd_mknod;
2924                 break;
2925         case DFT_INDEX:
2926                 result = osd_mk_index;
2927                 break;
2928
2929         default:
2930                 LBUG();
2931                 break;
2932         }
2933         return result;
2934 }
2935
2936
2937 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2938                         struct dt_object *parent, struct dt_object *child,
2939                         umode_t child_mode)
2940 {
2941         LASSERT(ah);
2942
2943         ah->dah_parent = parent;
2944         ah->dah_mode = child_mode;
2945
2946         if (parent != NULL && !dt_object_remote(parent)) {
2947                 /* will help to find FID->ino at dt_insert("..") */
2948                 struct osd_object *pobj = osd_dt_obj(parent);
2949                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2950         }
2951 }
2952
2953 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2954                           struct lu_attr *attr, struct dt_object_format *dof)
2955 {
2956         struct inode   *inode = obj->oo_inode;
2957         __u64           valid = attr->la_valid;
2958         int             result;
2959
2960         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2961
2962         if (dof->dof_type != DFT_NODE)
2963                 attr->la_valid &= ~LA_RDEV;
2964         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2965                 attr->la_valid &= ~LA_ATIME;
2966         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2967                 attr->la_valid &= ~LA_CTIME;
2968         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2969                 attr->la_valid &= ~LA_MTIME;
2970
2971         result = osd_quota_transfer(inode, attr);
2972         if (result)
2973                 return;
2974
2975         if (attr->la_valid != 0) {
2976                 result = osd_inode_setattr(info->oti_env, inode, attr);
2977                 /*
2978                  * The osd_inode_setattr() should always succeed here.  The
2979                  * only error that could be returned is EDQUOT when we are
2980                  * trying to change the UID or GID of the inode. However, this
2981                  * should not happen since quota enforcement is no longer
2982                  * enabled on ldiskfs (lquota takes care of it).
2983                  */
2984                 LASSERTF(result == 0, "%d\n", result);
2985                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2986         }
2987
2988         attr->la_valid = valid;
2989 }
2990
2991 /**
2992  * Helper function for osd_create()
2993  *
2994  * \retval 0, on success
2995  */
2996 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
2997                         struct lu_attr *attr, struct dt_allocation_hint *hint,
2998                         struct dt_object_format *dof, struct thandle *th)
2999 {
3000         int     result;
3001         __u32   umask;
3002
3003         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3004
3005         /* we drop umask so that permissions we pass are not affected */
3006         umask = current->fs->umask;
3007         current->fs->umask = 0;
3008
3009         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3010                                                   th);
3011         if (likely(obj->oo_inode != NULL)) {
3012                 LASSERT(obj->oo_inode->i_state & I_NEW);
3013
3014                 /* Unlock the inode before attr initialization to avoid
3015                  * unnecessary dqget operations. LU-6378 */
3016                 unlock_new_inode(obj->oo_inode);
3017         }
3018
3019         if (likely(result == 0)) {
3020                 osd_attr_init(info, obj, attr, dof);
3021                 osd_object_init0(obj);
3022         }
3023
3024         /* restore previous umask value */
3025         current->fs->umask = umask;
3026
3027         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3028
3029         return result;
3030 }
3031
3032 /**
3033  * Helper function for osd_create()
3034  *
3035  * \retval 0, on success
3036  */
3037 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3038                            const struct lu_fid *fid, struct thandle *th)
3039 {
3040         struct osd_thread_info *info = osd_oti_get(env);
3041         struct osd_inode_id    *id   = &info->oti_id;
3042         struct osd_device      *osd  = osd_obj2dev(obj);
3043         struct osd_thandle     *oh;
3044         int                     rc;
3045
3046         LASSERT(obj->oo_inode != NULL);
3047
3048         oh = container_of0(th, struct osd_thandle, ot_super);
3049         LASSERT(oh->ot_handle);
3050         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3051
3052         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3053         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3054                            OI_CHECK_FLD, NULL);
3055         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3056
3057         return rc;
3058 }
3059
3060 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3061                    u64 seq, struct lu_seq_range *range)
3062 {
3063         struct seq_server_site  *ss = osd_seq_site(osd);
3064
3065         if (fid_seq_is_idif(seq)) {
3066                 fld_range_set_ost(range);
3067                 range->lsr_index = idif_ost_idx(seq);
3068                 return 0;
3069         }
3070
3071         if (!fid_seq_in_fldb(seq)) {
3072                 fld_range_set_mdt(range);
3073                 if (ss != NULL)
3074                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3075                          * at all */
3076                         range->lsr_index = ss->ss_node_id;
3077                 return 0;
3078         }
3079
3080         LASSERT(ss != NULL);
3081         fld_range_set_any(range);
3082         /* OSD will only do local fld lookup */
3083         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3084 }
3085
3086 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3087                               struct lu_attr *attr,
3088                               struct dt_allocation_hint *hint,
3089                               struct dt_object_format *dof,
3090                               struct thandle *handle)
3091 {
3092         struct osd_thandle *oh;
3093         int rc;
3094         ENTRY;
3095
3096         LASSERT(handle != NULL);
3097
3098         oh = container_of0(handle, struct osd_thandle, ot_super);
3099         LASSERT(oh->ot_handle == NULL);
3100
3101         /* EA object consumes more credits than regular object: osd_mk_index
3102          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3103          * leaf_node, could involves the block, block bitmap, groups, GDT
3104          * change for each block, so add 4 * 2 credits in that case. */
3105         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3106                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3107                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3108         /* Reuse idle OI block may cause additional one OI block
3109          * to be changed. */
3110         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3111                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3112
3113         if (!attr)
3114                 RETURN(0);
3115
3116         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3117                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3118                                    NULL, OSD_QID_INODE);
3119         if (rc != 0)
3120                 RETURN(rc);
3121
3122         /* will help to find FID->ino mapping at dt_insert() */
3123         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3124                                    osd_dt_obj(dt));
3125
3126         RETURN(rc);
3127 }
3128
3129 /**
3130  * Called to destroy on-disk representation of the object
3131  *
3132  * Concurrency: must be locked
3133  */
3134 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3135                                struct thandle *th)
3136 {
3137         struct osd_object  *obj = osd_dt_obj(dt);
3138         struct inode       *inode = obj->oo_inode;
3139         struct osd_thandle *oh;
3140         int                 rc;
3141         ENTRY;
3142
3143         if (inode == NULL)
3144                 RETURN(-ENOENT);
3145
3146         oh = container_of0(th, struct osd_thandle, ot_super);
3147         LASSERT(oh->ot_handle == NULL);
3148
3149         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3150                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3151         /* Recycle idle OI leaf may cause additional three OI blocks
3152          * to be changed. */
3153         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3154                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3155                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3156         /* one less inode */
3157         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3158                                    i_projid_read(inode), -1, oh, obj, NULL,
3159                                    OSD_QID_INODE);
3160         if (rc)
3161                 RETURN(rc);
3162         /* data to be truncated */
3163         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3164                                    i_projid_read(inode), 0, oh, obj, NULL,
3165                                    OSD_QID_BLK);
3166         if (rc)
3167                 RETURN(rc);
3168
3169         /* will help to find FID->ino when this object is being
3170          * added to PENDING/ */
3171         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3172
3173         RETURN(rc);
3174 }
3175
3176 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3177                        struct thandle *th)
3178 {
3179         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3180         struct osd_object *obj = osd_dt_obj(dt);
3181         struct inode *inode = obj->oo_inode;
3182         struct osd_device *osd = osd_obj2dev(obj);
3183         struct osd_thandle *oh;
3184         int result;
3185         ENTRY;
3186
3187         oh = container_of0(th, struct osd_thandle, ot_super);
3188         LASSERT(oh->ot_handle);
3189         LASSERT(inode);
3190         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3191
3192         if (unlikely(fid_is_acct(fid)))
3193                 RETURN(-EPERM);
3194
3195         if (S_ISDIR(inode->i_mode)) {
3196                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3197                         inode->i_nlink == 2);
3198                 /* it will check/delete the inode from remote parent,
3199                  * how to optimize it? unlink performance impaction XXX */
3200                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3201                 if (result != 0)
3202                         CERROR("%s: delete inode "DFID": rc = %d\n",
3203                                osd_name(osd), PFID(fid), result);
3204
3205                 spin_lock(&obj->oo_guard);
3206                 clear_nlink(inode);
3207                 spin_unlock(&obj->oo_guard);
3208                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3209         }
3210
3211         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3212
3213         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3214
3215         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3216                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3217                                        oh->ot_handle, OI_CHECK_FLD);
3218
3219         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3220         /* XXX: add to ext3 orphan list */
3221         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3222
3223         /* not needed in the cache anymore */
3224         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3225         obj->oo_destroyed = 1;
3226
3227         RETURN(0);
3228 }
3229
3230 /**
3231  * Put the fid into lustre_mdt_attrs, and then place the structure
3232  * inode's ea. This fid should not be altered during the life time
3233  * of the inode.
3234  *
3235  * \retval +ve, on success
3236  * \retval -ve, on error
3237  *
3238  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3239  */
3240 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3241                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3242 {
3243         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3244         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3245         int rc;
3246         ENTRY;
3247
3248         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3249                 RETURN(0);
3250
3251         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3252                 rc = -ENOMEM;
3253
3254         lustre_loa_init(loa, fid, compat, incompat);
3255         lustre_loa_swab(loa, false);
3256
3257         /* For the OST device with 256 bytes inode size by default,
3258          * the PFID EA will be stored together with LMA EA to avoid
3259          * performance trouble. Otherwise the PFID EA can be stored
3260          * independently. LU-8998 */
3261         if ((compat & LMAC_FID_ON_OST) &&
3262             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3263                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3264                                      sizeof(*loa), XATTR_CREATE);
3265         else
3266                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3267                                      sizeof(*lma), XATTR_CREATE);
3268         /* LMA may already exist, but we need to check that all the
3269          * desired compat/incompat flags have been added. */
3270         if (unlikely(rc == -EEXIST)) {
3271                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3272                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3273                 if (rc < 0)
3274                         RETURN(rc);
3275
3276                 if (rc < sizeof(*lma))
3277                         RETURN(-EINVAL);
3278
3279                 lustre_loa_swab(loa, true);
3280                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3281                     ((compat == 0 && incompat == 0) ||
3282                      (!(~lma->lma_compat & compat) &&
3283                       !(~lma->lma_incompat & incompat))))
3284                         RETURN(0);
3285
3286                 lma->lma_self_fid = *fid;
3287                 lma->lma_compat |= compat;
3288                 lma->lma_incompat |= incompat;
3289                 if (rc == sizeof(*lma)) {
3290                         lustre_lma_swab(lma);
3291                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3292                                              sizeof(*lma), XATTR_REPLACE);
3293                 } else {
3294                         lustre_loa_swab(loa, false);
3295                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3296                                              sizeof(*loa), XATTR_REPLACE);
3297                 }
3298         }
3299
3300         RETURN(rc);
3301 }
3302
3303 /**
3304  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3305  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3306  * To have compatilibility with 1.8 ldiskfs driver we need to have
3307  * magic number at start of fid data.
3308  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3309  * its inmemory API.
3310  */
3311 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3312                                          const struct lu_fid *fid)
3313 {
3314         if (!fid_is_namespace_visible(fid) ||
3315             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3316                 param->edp_magic = 0;
3317                 return;
3318         }
3319
3320         param->edp_magic = LDISKFS_LUFID_MAGIC;
3321         param->edp_len =  sizeof(struct lu_fid) + 1;
3322         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3323 }
3324
3325 /**
3326  * Try to read the fid from inode ea into dt_rec.
3327  *
3328  * \param fid object fid.
3329  *
3330  * \retval 0 on success
3331  */
3332 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3333                           __u32 ino, struct lu_fid *fid,
3334                           struct osd_inode_id *id)
3335 {
3336         struct osd_thread_info *info  = osd_oti_get(env);
3337         struct inode           *inode;
3338         ENTRY;
3339
3340         osd_id_gen(id, ino, OSD_OII_NOGEN);
3341         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3342         if (IS_ERR(inode))
3343                 RETURN(PTR_ERR(inode));
3344
3345         iput(inode);
3346         RETURN(0);
3347 }
3348
3349 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3350                                         struct inode *dir,
3351                                         struct inode *parent_dir,
3352                                         const struct lu_fid *dot_fid,
3353                                         const struct lu_fid *dot_dot_fid,
3354                                         struct osd_thandle *oth)
3355 {
3356         struct ldiskfs_dentry_param *dot_ldp;
3357         struct ldiskfs_dentry_param *dot_dot_ldp;
3358         __u32 saved_nlink = dir->i_nlink;
3359         int rc;
3360
3361         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3362         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3363
3364         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3365         dot_ldp->edp_magic = 0;
3366
3367         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3368                                     dir, dot_ldp, dot_dot_ldp);
3369         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3370          * the subseqent ref_add() will increase the dir->i_nlink
3371          * as 3. That is incorrect for new created directory.
3372          *
3373          * It looks like hack, because we want to make the OSD API
3374          * to be order-independent for new created directory object
3375          * between dt_insert(..) and ref_add() operations.
3376          *
3377          * Here, we only restore the in-RAM dir-inode's nlink attr,
3378          * becuase if the nlink attr is not 2, then there will be
3379          * ref_add() called following the dt_insert(..), such call
3380          * will make both the in-RAM and on-disk dir-inode's nlink
3381          * attr to be set as 2. LU-7447 */
3382         set_nlink(dir, saved_nlink);
3383         return rc;
3384 }
3385
3386 /**
3387  * Create an local agent inode for remote entry
3388  */
3389 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3390                                                   struct osd_device *osd,
3391                                                   struct osd_object *pobj,
3392                                                   const struct lu_fid *fid,
3393                                                   __u32 type,
3394                                                   struct thandle *th)
3395 {
3396         struct osd_thread_info  *info = osd_oti_get(env);
3397         struct inode            *local;
3398         struct osd_thandle      *oh;
3399         int                     rc;
3400         ENTRY;
3401
3402         LASSERT(th);
3403         oh = container_of(th, struct osd_thandle, ot_super);
3404         LASSERT(oh->ot_handle->h_transaction != NULL);
3405
3406         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3407         if (IS_ERR(local)) {
3408                 CERROR("%s: create local error %d\n", osd_name(osd),
3409                        (int)PTR_ERR(local));
3410                 RETURN(local);
3411         }
3412
3413         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3414          * correct gid on remote file, not agent here */
3415         local->i_gid = current_fsgid();
3416         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3417         unlock_new_inode(local);
3418
3419         /* Agent inode should not have project ID */
3420 #ifdef  HAVE_PROJECT_QUOTA
3421         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL) {
3422                 rc = __ldiskfs_ioctl_setproject(local, 0);
3423                 if (rc) {
3424                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3425                                "quota enforcement enabled on the ldiskfs "
3426                                "filesystem?\n", local->i_sb->s_id, rc);
3427                         RETURN(ERR_PTR(rc));
3428                 }
3429         }
3430 #endif
3431         /* Set special LMA flag for local agent inode */
3432         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3433         if (rc != 0) {
3434                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3435                        osd_name(osd), PFID(fid), rc);
3436                 RETURN(ERR_PTR(rc));
3437         }
3438
3439         if (!S_ISDIR(type))
3440                 RETURN(local);
3441
3442         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3443                                          lu_object_fid(&pobj->oo_dt.do_lu),
3444                                          fid, oh);
3445         if (rc != 0) {
3446                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3447                         osd_name(osd), PFID(fid), rc);
3448                 RETURN(ERR_PTR(rc));
3449         }
3450
3451         RETURN(local);
3452 }
3453
3454 /**
3455  * when direntry is deleted, we have to take care of possible agent inode
3456  * referenced by that. unfortunately we can't do this at that point:
3457  * iget() within a running transaction leads to deadlock and we better do
3458  * not call that every delete declaration to save performance. so we put
3459  * a potention agent inode on a list and process that once the transaction
3460  * is over. Notice it's not any worse in terms of real orphans as regular
3461  * object destroy doesn't put inodes on the on-disk orphan list. this should
3462  * be addressed separately
3463  */
3464 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3465                                             struct osd_thandle *oh,
3466                                             __u32 ino)
3467 {
3468         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3469         struct osd_obj_orphan *oor;
3470
3471         OBD_ALLOC_PTR(oor);
3472         if (oor == NULL)
3473                 return -ENOMEM;
3474
3475         oor->oor_ino = ino;
3476         oor->oor_env = (struct lu_env *)env;
3477         spin_lock(&osd->od_osfs_lock);
3478         list_add(&oor->oor_list, &osd->od_orphan_list);
3479         spin_unlock(&osd->od_osfs_lock);
3480
3481         oh->ot_remove_agents = 1;
3482
3483         return 0;
3484
3485 }
3486
3487 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3488                                                 struct osd_device *osd)
3489 {
3490         struct osd_thread_info *info = osd_oti_get(env);
3491         struct osd_obj_orphan *oor, *tmp;
3492         struct osd_inode_id id;
3493         struct list_head list;
3494         struct inode *inode;
3495         struct lu_fid fid;
3496         handle_t *jh;
3497         __u32 ino;
3498
3499         INIT_LIST_HEAD(&list);
3500
3501         spin_lock(&osd->od_osfs_lock);
3502         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3503                 if (oor->oor_env == env) {
3504                         list_del(&oor->oor_list);
3505                         list_add(&oor->oor_list, &list);
3506                 }
3507         }
3508         spin_unlock(&osd->od_osfs_lock);
3509
3510         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3511
3512                 ino = oor->oor_ino;
3513
3514                 list_del(&oor->oor_list);
3515                 OBD_FREE_PTR(oor);
3516
3517                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3518                 inode = osd_iget_fid(info, osd, &id, &fid);
3519                 if (IS_ERR(inode))
3520                         continue;
3521
3522                 if (!osd_remote_fid(env, osd, &fid)) {
3523                         iput(inode);
3524                         continue;
3525                 }
3526
3527                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3528                 clear_nlink(inode);
3529                 mark_inode_dirty(inode);
3530                 ldiskfs_journal_stop(jh);
3531                 iput(inode);
3532         }
3533
3534         return 0;
3535 }
3536
3537 /**
3538  * OSD layer object create function for OST objects (b=11826).
3539  *
3540  * The FID is inserted into inode xattr here.
3541  *
3542  * \retval   0, on success
3543  * \retval -ve, on error
3544  */
3545 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3546                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3547                       struct dt_object_format *dof, struct thandle *th)
3548 {
3549         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3550         struct osd_object       *obj    = osd_dt_obj(dt);
3551         struct osd_thread_info  *info   = osd_oti_get(env);
3552         int                      result, on_ost = 0;
3553
3554         ENTRY;
3555
3556         if (dt_object_exists(dt))
3557                 RETURN(-EEXIST);
3558
3559         LINVRNT(osd_invariant(obj));
3560         LASSERT(!dt_object_remote(dt));
3561         LASSERT(osd_is_write_locked(env, obj));
3562         LASSERT(th != NULL);
3563
3564         if (unlikely(fid_is_acct(fid)))
3565                 /* Quota files can't be created from the kernel any more,
3566                  * 'tune2fs -O quota' will take care of creating them */
3567                 RETURN(-EPERM);
3568
3569         result = __osd_create(info, obj, attr, hint, dof, th);
3570         if (result == 0) {
3571                 if (fid_is_idif(fid) &&
3572                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3573                         struct lu_fid *tfid = &info->oti_fid;
3574                         struct ost_id *oi   = &info->oti_ostid;
3575
3576                         fid_to_ostid(fid, oi);
3577                         ostid_to_fid(tfid, oi, 0);
3578                         on_ost = 1;
3579                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3580                                                 LMAC_FID_ON_OST, 0);
3581                 } else {
3582                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3583                                                fid, OI_CHECK_FLD);
3584                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3585                                                 on_ost ? LMAC_FID_ON_OST : 0,
3586                                                 0);
3587                 }
3588                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3589                         obj->oo_dt.do_body_ops = &osd_body_ops;
3590         }
3591
3592         if (result == 0)
3593                 result = __osd_oi_insert(env, obj, fid, th);
3594
3595         /* a small optimization - dt_insert() isn't usually applied
3596          * to OST objects, so we don't need to cache OI mapping for
3597          * OST objects */
3598         if (result == 0 && on_ost == 0) {
3599                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3600                 result = osd_idc_find_and_init(env, osd, obj);
3601                 LASSERT(result == 0);
3602         }
3603
3604         LASSERT(ergo(result == 0,
3605                      dt_object_exists(dt) && !dt_object_remote(dt)));
3606         LINVRNT(osd_invariant(obj));
3607         RETURN(result);
3608 }
3609
3610 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3611                                struct thandle *handle)
3612 {
3613         struct osd_thandle *oh;
3614
3615         /* it's possible that object doesn't exist yet */
3616         LASSERT(handle != NULL);
3617
3618         oh = container_of0(handle, struct osd_thandle, ot_super);
3619         LASSERT(oh->ot_handle == NULL);
3620
3621         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3622                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3623
3624         return 0;
3625 }
3626
3627 /*
3628  * Concurrency: @dt is write locked.
3629  */
3630 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3631                        struct thandle *th)
3632 {
3633         struct osd_object  *obj = osd_dt_obj(dt);
3634         struct inode       *inode = obj->oo_inode;
3635         struct osd_thandle *oh;
3636         int                 rc = 0;
3637
3638         if (!dt_object_exists(dt) || obj->oo_destroyed)
3639                 return -ENOENT;
3640
3641         LINVRNT(osd_invariant(obj));
3642         LASSERT(!dt_object_remote(dt));
3643         LASSERT(osd_is_write_locked(env, obj));
3644         LASSERT(th != NULL);
3645
3646         oh = container_of0(th, struct osd_thandle, ot_super);
3647         LASSERT(oh->ot_handle != NULL);
3648
3649         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3650
3651         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3652                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3653         /*
3654          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3655          * (65000) subdirectories by storing "1" in i_nlink if the link count
3656          * would otherwise overflow. Directory tranversal tools understand
3657          * that (st_nlink == 1) indicates that the filesystem dose not track
3658          * hard links count on the directory, and will not abort subdirectory
3659          * scanning early once (st_nlink - 2) subdirs have been found.
3660          *
3661          * This also has to properly handle the case of inodes with nlink == 0
3662          * in case they are being linked into the PENDING directory
3663          */
3664         spin_lock(&obj->oo_guard);
3665         if (unlikely(inode->i_nlink == 0))
3666                 /* inc_nlink from 0 may cause WARN_ON */
3667                 set_nlink(inode, 1);
3668         else {
3669                 ldiskfs_inc_count(oh->ot_handle, inode);
3670                 if (!S_ISDIR(inode->i_mode))
3671                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3672         }
3673         spin_unlock(&obj->oo_guard);
3674
3675         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3676         LINVRNT(osd_invariant(obj));
3677
3678         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3679
3680         return rc;
3681 }
3682
3683 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3684                                struct thandle *handle)
3685 {
3686         struct osd_thandle *oh;
3687
3688         if (!dt_object_exists(dt))
3689                 return -ENOENT;
3690
3691         LASSERT(!dt_object_remote(dt));
3692         LASSERT(handle != NULL);
3693
3694         oh = container_of0(handle, struct osd_thandle, ot_super);
3695         LASSERT(oh->ot_handle == NULL);
3696
3697         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3698                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3699
3700         return 0;
3701 }
3702
3703 /*
3704  * Concurrency: @dt is write locked.
3705  */
3706 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3707                        struct thandle *th)
3708 {
3709         struct osd_object       *obj = osd_dt_obj(dt);
3710         struct inode            *inode = obj->oo_inode;
3711         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3712         struct osd_thandle      *oh;
3713
3714         if (!dt_object_exists(dt))
3715                 return -ENOENT;
3716
3717         LINVRNT(osd_invariant(obj));
3718         LASSERT(!dt_object_remote(dt));
3719         LASSERT(osd_is_write_locked(env, obj));
3720         LASSERT(th != NULL);
3721
3722         oh = container_of0(th, struct osd_thandle, ot_super);
3723         LASSERT(oh->ot_handle != NULL);
3724
3725         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3726
3727         spin_lock(&obj->oo_guard);
3728         /* That can be result of upgrade from old Lustre version and
3729          * applied only to local files.  Just skip this ref_del call.
3730          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3731         if (inode->i_nlink == 0) {
3732                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3733                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3734                              ", maybe an upgraded file? (LU-3915)\n",
3735                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3736                 spin_unlock(&obj->oo_guard);
3737                 return 0;
3738         }
3739
3740         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3741                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3742
3743         ldiskfs_dec_count(oh->ot_handle, inode);
3744         spin_unlock(&obj->oo_guard);
3745
3746         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3747         LINVRNT(osd_invariant(obj));
3748
3749         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3750
3751         return 0;
3752 }
3753
3754 /*
3755  * Concurrency: @dt is read locked.
3756  */
3757 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3758                          struct lu_buf *buf, const char *name)
3759 {
3760         struct osd_object      *obj    = osd_dt_obj(dt);
3761         struct inode           *inode  = obj->oo_inode;
3762         struct osd_thread_info *info   = osd_oti_get(env);
3763         struct dentry          *dentry = &info->oti_obj_dentry;
3764         bool                    cache_xattr = false;
3765         int                     rc;
3766
3767         LASSERT(buf);
3768
3769         /* version get is not real XATTR but uses xattr API */
3770         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3771                 dt_obj_version_t *ver = buf->lb_buf;
3772
3773                 /* for version we are just using xattr API but change inode
3774                  * field instead */
3775                 if (buf->lb_len == 0)
3776                         return sizeof(dt_obj_version_t);
3777
3778                 if (buf->lb_len < sizeof(dt_obj_version_t))
3779                         return -ERANGE;
3780
3781                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3782                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3783
3784                 *ver = LDISKFS_I(inode)->i_fs_version;
3785
3786                 return sizeof(dt_obj_version_t);
3787         }
3788
3789         if (!dt_object_exists(dt))
3790                 return -ENOENT;
3791
3792         LASSERT(!dt_object_remote(dt));
3793         LASSERT(inode->i_op != NULL);
3794         LASSERT(inode->i_op->getxattr != NULL);
3795
3796         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3797             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3798                 cache_xattr = true;
3799
3800         if (cache_xattr) {
3801                 rc = osd_oxc_get(obj, name, buf);
3802                 if (rc != -ENOENT)
3803                         return rc;
3804         }
3805
3806         rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3807         if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3808                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3809                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3810                 struct filter_fid *ff;
3811                 struct ost_layout *ol;
3812
3813                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3814
3815                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3816                 if (rc)
3817                         return rc;
3818
3819                 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3820                         rc = -ENODATA;
3821                         goto cache;
3822                 }
3823
3824                 rc = sizeof(*ff);
3825                 if (buf->lb_len == 0 || !buf->lb_buf)
3826                         return rc;
3827
3828                 if (buf->lb_len < rc)
3829                         return -ERANGE;
3830
3831                 ff = buf->lb_buf;
3832                 ol = &ff->ff_layout;
3833                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3834                                                   PFID_STRIPE_IDX_BITS);
3835                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3836                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3837                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3838                 if (lma->lma_compat & LMAC_COMP_INFO) {
3839                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3840                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3841                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3842                 } else {
3843                         ol->ol_comp_start = 0;
3844                         ol->ol_comp_end = 0;
3845                         ol->ol_comp_id = 0;
3846                 }
3847         }
3848
3849 cache:
3850         if (cache_xattr) {
3851                 if (rc == -ENOENT || rc == -ENODATA)
3852                         osd_oxc_add(obj, name, NULL, 0);
3853                 else if (rc > 0 && buf->lb_buf != NULL)
3854                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3855         }
3856
3857         return rc;
3858 }
3859
3860 static int osd_declare_xattr_set(const struct lu_env *env,
3861                                  struct dt_object *dt,
3862                                  const struct lu_buf *buf, const char *name,
3863                                  int fl, struct thandle *handle)
3864 {
3865         struct osd_thandle *oh;
3866         int credits = 0;
3867         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3868
3869         LASSERT(handle != NULL);
3870
3871         oh = container_of0(handle, struct osd_thandle, ot_super);
3872         LASSERT(oh->ot_handle == NULL);
3873
3874         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3875                 /* For non-upgrading case, the LMA is set first and
3876                  * usually fit inode. But for upgrade case, the LMA
3877                  * may be in another separated EA block. */
3878                 if (dt_object_exists(dt)) {
3879                         if (fl == LU_XATTR_REPLACE)
3880                                 credits = 1;
3881                         else
3882                                 goto upgrade;
3883                 }
3884         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3885                 credits = 1;
3886         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3887                 /* We may need to delete the old PFID EA. */
3888                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3889                 if (fl == LU_XATTR_REPLACE)
3890                         credits += 1;
3891                 else
3892                         goto upgrade;
3893         } else {
3894
3895 upgrade:
3896                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3897
3898                 if (buf != NULL) {
3899                         ssize_t buflen;
3900
3901                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3902                                 /* learn xattr size from osd_xattr_get if
3903                                    attribute has not been read yet */
3904                                 buflen = __osd_xattr_get(
3905                                     osd_dt_obj(dt)->oo_inode,
3906                                     &osd_oti_get(env)->oti_obj_dentry,
3907                                     name, NULL, 0);
3908                                 if (buflen < 0)
3909                                         buflen = 0;
3910                         } else {
3911                                 buflen = buf->lb_len;
3912                         }
3913
3914                         if (buflen > sb->s_blocksize) {
3915                                 credits += osd_calc_bkmap_credits(
3916                                     sb, NULL, 0, -1,
3917                                     (buflen + sb->s_blocksize - 1) >>
3918                                     sb->s_blocksize_bits);
3919                         }
3920                 }
3921                 /*
3922                  * xattr set may involve inode quota change, reserve credits for
3923                  * dquot_initialize()
3924                  */
3925                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3926         }
3927
3928         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3929
3930         return 0;
3931 }
3932
3933 /*
3934  * Concurrency: @dt is write locked.
3935  */
3936 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3937                          const struct lu_buf *buf, const char *name, int fl,
3938                          struct thandle *handle)
3939 {
3940         struct osd_object *obj = osd_dt_obj(dt);
3941         struct inode *inode = obj->oo_inode;
3942         struct osd_thread_info *info = osd_oti_get(env);
3943         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3944         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3945         int fs_flags = 0;
3946         int len;
3947         int rc;
3948         ENTRY;
3949
3950         LASSERT(handle);
3951         LASSERT(buf);
3952
3953         /* version set is not real XATTR */
3954         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3955                 dt_obj_version_t *version = buf->lb_buf;
3956
3957                 /* for version we are just using xattr API but change inode
3958                  * field instead */
3959                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3960
3961                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3962                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3963
3964                 LDISKFS_I(inode)->i_fs_version = *version;
3965                 /* Version is set after all inode operations are finished,
3966                  * so we should mark it dirty here */
3967                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3968
3969                 RETURN(0);
3970         }
3971
3972         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3973                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3974
3975         len = buf->lb_len;
3976         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3977         if (fl & LU_XATTR_REPLACE)
3978                 fs_flags |= XATTR_REPLACE;
3979
3980         if (fl & LU_XATTR_CREATE)
3981                 fs_flags |= XATTR_CREATE;
3982
3983         /* For the OST device with 256 bytes inode size by default,
3984          * the PFID EA will be stored together with LMA EA to avoid
3985          * performance trouble. Otherwise the PFID EA can be stored
3986          * independently. LU-8998 */
3987         if (strcmp(name, XATTR_NAME_FID) == 0 &&
3988             LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
3989                 struct dentry *dentry = &info->oti_obj_dentry;
3990                 struct filter_fid *ff;
3991                 struct ost_layout *ol;
3992                 int fl;
3993
3994                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3995
3996                 ff = buf->lb_buf;
3997                 ol = &ff->ff_layout;
3998                 /* Old client does not send stripe information, store
3999                  * the PFID EA on disk directly. */
4000                 if (buf->lb_len == sizeof(struct lu_fid) ||
4001                     ol->ol_stripe_size == 0) {
4002                         len = sizeof(struct lu_fid);
4003                         goto set;
4004                 }
4005
4006                 if (buf->lb_len != sizeof(*ff))
4007                         RETURN(-EINVAL);
4008
4009                 rc = osd_get_lma(info, inode, dentry, loa);
4010                 if (unlikely(rc == -ENODATA)) {
4011                         /* Usually for upgarding from old device */
4012                         lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4013                                         LMAC_FID_ON_OST, 0);
4014                         fl = XATTR_CREATE;
4015                 } else if (rc) {
4016                         RETURN(rc);
4017                 } else {
4018                         fl = XATTR_REPLACE;
4019                 }
4020
4021                 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4022                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4023                                              PFID_STRIPE_IDX_BITS;
4024                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4025                 lma->lma_compat |= LMAC_STRIPE_INFO;
4026                 if (ol->ol_comp_id != 0) {
4027                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4028                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4029                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4030                         lma->lma_compat |= LMAC_COMP_INFO;
4031                 }
4032
4033                 lustre_loa_swab(loa, false);
4034
4035                 /* Remove old PFID EA entry firstly. */
4036                 ll_vfs_dq_init(inode);
4037                 rc = inode->i_op->removexattr(dentry, name);
4038                 if (rc && rc != -ENODATA)
4039                         RETURN(rc);
4040
4041                 /* Store the PFID EA inside the LMA EA. */
4042                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4043                                      sizeof(*loa), fl);
4044
4045                 RETURN(rc);
4046         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4047                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4048                 if (rc)
4049                         RETURN(rc);
4050
4051                 lma->lma_incompat |= LMAI_STRIPED;
4052                 lustre_lma_swab(lma);
4053                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4054                                      sizeof(*lma), XATTR_REPLACE);
4055                 if (rc != 0)
4056                         RETURN(rc);
4057         }
4058
4059 set:
4060         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4061         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4062
4063         if (rc == 0 &&
4064             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4065              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4066                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4067
4068         return rc;
4069 }
4070
4071 /*
4072  * Concurrency: @dt is read locked.
4073  */
4074 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4075                           const struct lu_buf *buf)
4076 {
4077         struct osd_object      *obj    = osd_dt_obj(dt);
4078         struct inode           *inode  = obj->oo_inode;
4079         struct osd_thread_info *info   = osd_oti_get(env);
4080         struct dentry          *dentry = &info->oti_obj_dentry;
4081
4082         if (!dt_object_exists(dt))
4083                 return -ENOENT;
4084
4085         LASSERT(!dt_object_remote(dt));
4086         LASSERT(inode->i_op != NULL);
4087         LASSERT(inode->i_op->listxattr != NULL);
4088
4089         dentry->d_inode = inode;
4090         dentry->d_sb = inode->i_sb;
4091         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4092 }
4093
4094 static int osd_declare_xattr_del(const struct lu_env *env,
4095                                  struct dt_object *dt, const char *name,
4096                                  struct thandle *handle)
4097 {
4098         struct osd_thandle *oh;
4099         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4100
4101         LASSERT(!dt_object_remote(dt));
4102         LASSERT(handle != NULL);
4103
4104         oh = container_of0(handle, struct osd_thandle, ot_super);
4105         LASSERT(oh->ot_handle == NULL);
4106
4107         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4108                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4109         /*
4110          * xattr del may involve inode quota change, reserve credits for
4111          * dquot_initialize()
4112          */
4113         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4114
4115         return 0;
4116 }
4117
4118 /*
4119  * Concurrency: @dt is write locked.
4120  */
4121 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4122                          const char *name, struct thandle *handle)
4123 {
4124         struct osd_object      *obj    = osd_dt_obj(dt);
4125         struct inode           *inode  = obj->oo_inode;
4126         struct osd_thread_info *info   = osd_oti_get(env);
4127         struct dentry          *dentry = &info->oti_obj_dentry;
4128         int                     rc;
4129
4130         if (!dt_object_exists(dt))
4131                 return -ENOENT;
4132
4133         LASSERT(!dt_object_remote(dt));
4134         LASSERT(inode->i_op != NULL);
4135         LASSERT(inode->i_op->removexattr != NULL);
4136         LASSERT(handle != NULL);
4137
4138         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4139
4140         ll_vfs_dq_init(inode);
4141         dentry->d_inode = inode;
4142         dentry->d_sb = inode->i_sb;
4143         rc = inode->i_op->removexattr(dentry, name);
4144         if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4145                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4146
4147                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4148
4149                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4150                                  &info->oti_ost_attrs);
4151                 if (!rc) {
4152                         if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4153                                 rc = -ENODATA;
4154                                 goto out;
4155                         }
4156
4157                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4158                         lustre_lma_swab(lma);
4159                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4160                                              sizeof(*lma), XATTR_REPLACE);
4161                 }
4162         }
4163
4164 out:
4165         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4166
4167         if (rc == 0 &&
4168             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4169              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4170                 osd_oxc_del(obj, name);
4171
4172         return rc;
4173 }
4174
4175 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4176                            __u64 start, __u64 end)
4177 {
4178         struct osd_object       *obj    = osd_dt_obj(dt);
4179         struct inode            *inode  = obj->oo_inode;
4180         struct osd_thread_info  *info   = osd_oti_get(env);
4181         struct dentry           *dentry = &info->oti_obj_dentry;
4182         struct file             *file   = &info->oti_file;
4183         int                     rc;
4184
4185         ENTRY;
4186
4187         dentry->d_inode = inode;
4188         dentry->d_sb = inode->i_sb;
4189         file->f_path.dentry = dentry;
4190         file->f_mapping = inode->i_mapping;
4191         file->f_op = inode->i_fop;
4192         set_file_inode(file, inode);
4193
4194         rc = ll_vfs_fsync_range(file, start, end, 0);
4195
4196         RETURN(rc);
4197 }
4198
4199 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4200 {
4201         return 0;
4202 }
4203
4204 /*
4205  * Index operations.
4206  */
4207
4208 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4209                            const struct dt_index_features *feat)
4210 {
4211         struct iam_descr *descr;
4212
4213         if (osd_object_is_root(o))
4214                 return feat == &dt_directory_features;
4215
4216         LASSERT(o->oo_dir != NULL);
4217
4218         descr = o->oo_dir->od_container.ic_descr;
4219         if (feat == &dt_directory_features) {
4220                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4221                         return 1;
4222                 else
4223                         return 0;
4224         } else {
4225                 return
4226                         feat->dif_keysize_min <= descr->id_key_size &&
4227                         descr->id_key_size <= feat->dif_keysize_max &&
4228                         feat->dif_recsize_min <= descr->id_rec_size &&
4229                         descr->id_rec_size <= feat->dif_recsize_max &&
4230                         !(feat->dif_flags & (DT_IND_VARKEY |
4231                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4232                         ergo(feat->dif_flags & DT_IND_UPDATE,
4233                              1 /* XXX check that object (and file system) is
4234                                 * writable */);
4235         }
4236 }
4237
4238 static int osd_iam_container_init(const struct lu_env *env,
4239                                   struct osd_object *obj,
4240                                   struct osd_directory *dir)
4241 {
4242         struct iam_container *bag = &dir->od_container;
4243         int result;
4244
4245         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4246         if (result != 0)
4247                 return result;
4248
4249         result = iam_container_setup(bag);
4250         if (result == 0)
4251                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4252         else
4253                 iam_container_fini(bag);
4254
4255         return result;
4256 }
4257
4258
4259 /*
4260  * Concurrency: no external locking is necessary.
4261  */
4262 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4263                          const struct dt_index_features *feat)
4264 {
4265         int                      result;
4266         int                      skip_iam = 0;
4267         struct osd_object       *obj = osd_dt_obj(dt);
4268
4269         LINVRNT(osd_invariant(obj));
4270
4271         if (osd_object_is_root(obj)) {
4272                 dt->do_index_ops = &osd_index_ea_ops;
4273                 result = 0;
4274         } else if (feat == &dt_directory_features) {
4275                 dt->do_index_ops = &osd_index_ea_ops;
4276                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4277                         result = 0;
4278                 else
4279                         result = -ENOTDIR;
4280                 skip_iam = 1;
4281         } else if (unlikely(feat == &dt_otable_features)) {
4282                 dt->do_index_ops = &osd_otable_ops;
4283                 return 0;
4284         } else if (unlikely(feat == &dt_acct_features)) {
4285                 dt->do_index_ops = &osd_acct_index_ops;
4286                 result = 0;
4287                 skip_iam = 1;
4288         } else if (!osd_has_index(obj)) {
4289                 struct osd_directory *dir;
4290
4291                 OBD_ALLOC_PTR(dir);
4292                 if (dir != NULL) {
4293
4294                         spin_lock(&obj->oo_guard);
4295                         if (obj->oo_dir == NULL)
4296                                 obj->oo_dir = dir;
4297                         else
4298                                 /*
4299                                  * Concurrent thread allocated container data.
4300                                  */
4301                                 OBD_FREE_PTR(dir);
4302                         spin_unlock(&obj->oo_guard);
4303                         /*
4304                          * Now, that we have container data, serialize its
4305                          * initialization.
4306                          */
4307                         down_write(&obj->oo_ext_idx_sem);
4308                         /*
4309                          * recheck under lock.
4310                          */
4311                         if (!osd_has_index(obj))
4312                                 result = osd_iam_container_init(env, obj,
4313                                                                 obj->oo_dir);
4314                         else
4315                                 result = 0;
4316                         up_write(&obj->oo_ext_idx_sem);
4317                 } else {
4318                         result = -ENOMEM;
4319                 }
4320         } else {
4321                 result = 0;
4322         }
4323
4324         if (result == 0 && skip_iam == 0) {
4325                 if (!osd_iam_index_probe(env, obj, feat))
4326                         result = -ENOTDIR;
4327         }
4328         LINVRNT(osd_invariant(obj));
4329
4330         return result;
4331 }
4332
4333 static int osd_otable_it_attr_get(const struct lu_env *env,
4334                                  struct dt_object *dt,
4335                                  struct lu_attr *attr)
4336 {
4337         attr->la_valid = 0;
4338         return 0;
4339 }
4340
4341 static const struct dt_object_operations osd_obj_ops = {
4342         .do_read_lock           = osd_read_lock,
4343         .do_write_lock          = osd_write_lock,
4344         .do_read_unlock         = osd_read_unlock,
4345         .do_write_unlock        = osd_write_unlock,
4346         .do_write_locked        = osd_write_locked,
4347         .do_attr_get            = osd_attr_get,
4348         .do_declare_attr_set    = osd_declare_attr_set,
4349         .do_attr_set            = osd_attr_set,
4350         .do_ah_init             = osd_ah_init,
4351         .do_declare_create      = osd_declare_create,
4352         .do_create              = osd_create,
4353         .do_declare_destroy     = osd_declare_destroy,
4354         .do_destroy             = osd_destroy,
4355         .do_index_try           = osd_index_try,
4356         .do_declare_ref_add     = osd_declare_ref_add,
4357         .do_ref_add             = osd_ref_add,
4358         .do_declare_ref_del     = osd_declare_ref_del,
4359         .do_ref_del             = osd_ref_del,
4360         .do_xattr_get           = osd_xattr_get,
4361         .do_declare_xattr_set   = osd_declare_xattr_set,
4362         .do_xattr_set           = osd_xattr_set,
4363         .do_declare_xattr_del   = osd_declare_xattr_del,
4364         .do_xattr_del           = osd_xattr_del,
4365         .do_xattr_list          = osd_xattr_list,
4366         .do_object_sync         = osd_object_sync,
4367         .do_invalidate          = osd_invalidate,
4368 };
4369
4370 static const struct dt_object_operations osd_obj_otable_it_ops = {
4371         .do_attr_get    = osd_otable_it_attr_get,
4372         .do_index_try   = osd_index_try,
4373 };
4374
4375 static int osd_index_declare_iam_delete(const struct lu_env *env,
4376                                         struct dt_object *dt,
4377                                         const struct dt_key *key,
4378                                         struct thandle *handle)
4379 {
4380         struct osd_thandle    *oh;
4381
4382         oh = container_of0(handle, struct osd_thandle, ot_super);
4383         LASSERT(oh->ot_handle == NULL);
4384
4385         /* Recycle  may cause additional three blocks to be changed. */
4386         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4387                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4388
4389         return 0;
4390 }
4391
4392 /**
4393  *      delete a (key, value) pair from index \a dt specified by \a key
4394  *
4395  *      \param  dt      osd index object
4396  *      \param  key     key for index
4397  *      \param  rec     record reference
4398  *      \param  handle  transaction handler
4399  *
4400  *      \retval  0  success
4401  *      \retval -ve   failure
4402  */
4403 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4404                                 const struct dt_key *key,
4405                                 struct thandle *handle)
4406 {
4407         struct osd_thread_info *oti = osd_oti_get(env);
4408         struct osd_object      *obj = osd_dt_obj(dt);
4409         struct osd_thandle     *oh;
4410         struct iam_path_descr  *ipd;
4411         struct iam_container   *bag = &obj->oo_dir->od_container;
4412         int                     rc;
4413         ENTRY;
4414
4415         if (!dt_object_exists(dt))
4416                 RETURN(-ENOENT);
4417
4418         LINVRNT(osd_invariant(obj));
4419         LASSERT(!dt_object_remote(dt));
4420         LASSERT(bag->ic_object == obj->oo_inode);
4421         LASSERT(handle != NULL);
4422
4423         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4424
4425         ipd = osd_idx_ipd_get(env, bag);
4426         if (unlikely(ipd == NULL))
4427                 RETURN(-ENOMEM);
4428
4429         oh = container_of0(handle, struct osd_thandle, ot_super);
4430         LASSERT(oh->ot_handle != NULL);
4431         LASSERT(oh->ot_handle->h_transaction != NULL);
4432
4433         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4434                 /* swab quota uid/gid provided by caller */
4435                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4436                 key = (const struct dt_key *)&oti->oti_quota_id;
4437         }
4438
4439         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4440         osd_ipd_put(env, bag, ipd);
4441         LINVRNT(osd_invariant(obj));
4442         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4443         RETURN(rc);
4444 }
4445
4446 static int osd_index_declare_ea_delete(const struct lu_env *env,
4447                                        struct dt_object *dt,
4448                                        const struct dt_key *key,
4449                                        struct thandle *handle)
4450 {
4451         struct osd_thandle *oh;
4452         struct inode       *inode;
4453         int                 rc, credits;
4454         ENTRY;
4455
4456         LASSERT(!dt_object_remote(dt));
4457         LASSERT(handle != NULL);
4458
4459         oh = container_of0(handle, struct osd_thandle, ot_super);
4460         LASSERT(oh->ot_handle == NULL);
4461
4462         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4463         if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4464                 /* '..' to a remote object has a local representative */
4465                 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4466                 /* to reset LMAI_REMOTE_PARENT */
4467                 credits += 1;
4468         }
4469         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4470
4471         inode = osd_dt_obj(dt)->oo_inode;
4472         if (inode == NULL)
4473                 RETURN(-ENOENT);
4474
4475         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4476                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4477                                    NULL, OSD_QID_BLK);
4478         RETURN(rc);
4479 }
4480
4481 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4482                                           struct dt_rec *fid)
4483 {
4484         struct osd_fid_pack *rec;
4485         int                  rc = -ENODATA;
4486
4487         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4488                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4489                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4490                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4491                         rc = -EINVAL;
4492         }
4493         return rc;
4494 }
4495
4496 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4497                           const struct lu_fid *fid)
4498 {
4499         struct seq_server_site  *ss = osd_seq_site(osd);
4500         ENTRY;
4501
4502         /* FID seqs not in FLDB, must be local seq */
4503         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4504                 RETURN(0);
4505
4506         /* If FLD is not being initialized yet, it only happens during the
4507          * initialization, likely during mgs initialization, and we assume
4508          * this is local FID. */
4509         if (ss == NULL || ss->ss_server_fld == NULL)
4510                 RETURN(0);
4511
4512         /* Only check the local FLDB here */
4513         if (osd_seq_exists(env, osd, fid_seq(fid)))
4514                 RETURN(0);
4515
4516         RETURN(1);
4517 }
4518
4519 /**
4520  * Index delete function for interoperability mode (b11826).
4521  * It will remove the directory entry added by osd_index_ea_insert().
4522  * This entry is needed to maintain name->fid mapping.
4523  *
4524  * \param key,  key i.e. file entry to be deleted
4525  *
4526  * \retval   0, on success
4527  * \retval -ve, on error
4528  */
4529 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4530                                const struct dt_key *key, struct thandle *handle)
4531 {
4532         struct osd_object          *obj = osd_dt_obj(dt);
4533         struct inode               *dir = obj->oo_inode;
4534         struct dentry              *dentry;
4535         struct osd_thandle         *oh;
4536         struct ldiskfs_dir_entry_2 *de = NULL;
4537         struct buffer_head         *bh;
4538         struct htree_lock          *hlock = NULL;
4539         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
4540         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4541         int                        rc;
4542         ENTRY;
4543
4544         if (!dt_object_exists(dt))
4545                 RETURN(-ENOENT);
4546
4547         LINVRNT(osd_invariant(obj));
4548         LASSERT(!dt_object_remote(dt));
4549         LASSERT(handle != NULL);
4550
4551         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4552
4553         oh = container_of(handle, struct osd_thandle, ot_super);
4554         LASSERT(oh->ot_handle != NULL);
4555         LASSERT(oh->ot_handle->h_transaction != NULL);
4556
4557         ll_vfs_dq_init(dir);
4558         dentry = osd_child_dentry_get(env, obj,
4559                                       (char *)key, strlen((char *)key));
4560
4561         if (obj->oo_hl_head != NULL) {
4562                 hlock = osd_oti_get(env)->oti_hlock;
4563                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4564                                    dir, LDISKFS_HLOCK_DEL);
4565         } else {
4566                 down_write(&obj->oo_ext_idx_sem);
4567         }
4568
4569         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4570         if (!IS_ERR(bh)) {
4571                 /* If this is not the ".." entry, it might be a remote DNE
4572                  * entry and  we need to check if the FID is for a remote
4573                  * MDT.  If the FID is  not in the directory entry (e.g.
4574                  * upgraded 1.8 filesystem without dirdata enabled) then
4575                  * we need to get the FID from the LMA. For a remote directory
4576                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4577                  * case.
4578                  *
4579                  * Delete the entry before the agent inode in order to
4580                  * simplify error handling.  At worst an error after deleting
4581                  * the entry first might leak the agent inode afterward. The
4582                  * reverse would need filesystem abort in case of error deleting
4583                  * the entry after the agent had been removed, or leave a
4584                  * dangling entry pointing at a random inode. */
4585                 if (strcmp((char *)key, dotdot) != 0) {
4586                         LASSERT(de != NULL);
4587                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4588                         if (rc == -ENODATA) {
4589                                 /* can't get FID, postpone to the end of the
4590                                  * transaction when iget() is safe */
4591                                 osd_schedule_agent_inode_removal(env, oh,
4592                                                 le32_to_cpu(de->inode));
4593                         } else if (rc == 0 &&
4594                                    unlikely(osd_remote_fid(env, osd, fid))) {
4595                                 osd_schedule_agent_inode_removal(env, oh,
4596                                                 le32_to_cpu(de->inode));
4597                         }
4598                 }
4599                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4600                 brelse(bh);
4601         } else {
4602                 rc = PTR_ERR(bh);
4603         }
4604         if (hlock != NULL)
4605                 ldiskfs_htree_unlock(hlock);
4606         else
4607                 up_write(&obj->oo_ext_idx_sem);
4608
4609         if (rc != 0)
4610                 GOTO(out, rc);
4611
4612         /* For inode on the remote MDT, .. will point to
4613          * /Agent directory, Check whether it needs to delete
4614          * from agent directory */
4615         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4616                 int ret;
4617
4618                 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4619                                                     obj, oh);
4620                 if (ret != 0)
4621                         /* Sigh, the entry has been deleted, and
4622                          * it is not easy to revert it back, so
4623                          * let's keep this error private, and let
4624                          * LFSCK fix it. XXX */
4625                         CERROR("%s: delete remote parent "DFID": rc = %d\n",
4626                                osd_name(osd), PFID(fid), ret);
4627         }
4628 out:
4629         LASSERT(osd_invariant(obj));
4630         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4631         RETURN(rc);
4632 }
4633
4634 /**
4635  *      Lookup index for \a key and copy record to \a rec.
4636  *
4637  *      \param  dt      osd index object
4638  *      \param  key     key for index
4639  *      \param  rec     record reference
4640  *
4641  *      \retval  +ve  success : exact mach
4642  *      \retval  0    return record with key not greater than \a key
4643  *      \retval -ve   failure
4644  */
4645 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4646                                 struct dt_rec *rec, const struct dt_key *key)
4647 {
4648         struct osd_object      *obj = osd_dt_obj(dt);
4649         struct iam_path_descr  *ipd;
4650         struct iam_container   *bag = &obj->oo_dir->od_container;
4651         struct osd_thread_info *oti = osd_oti_get(env);
4652         struct iam_iterator    *it = &oti->oti_idx_it;
4653         struct iam_rec         *iam_rec;
4654         int                     rc;
4655         ENTRY;
4656
4657         if (!dt_object_exists(dt))
4658                 RETURN(-ENOENT);
4659
4660         LASSERT(osd_invariant(obj));
4661         LASSERT(!dt_object_remote(dt));
4662         LASSERT(bag->ic_object == obj->oo_inode);
4663
4664         ipd = osd_idx_ipd_get(env, bag);
4665         if (IS_ERR(ipd))
4666                 RETURN(-ENOMEM);
4667
4668         /* got ipd now we can start iterator. */
4669         iam_it_init(it, bag, 0, ipd);
4670
4671         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4672                 /* swab quota uid/gid provided by caller */
4673                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4674                 key = (const struct dt_key *)&oti->oti_quota_id;
4675         }
4676
4677         rc = iam_it_get(it, (struct iam_key *)key);
4678         if (rc >= 0) {
4679                 if (S_ISDIR(obj->oo_inode->i_mode))
4680                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4681                 else
4682                         iam_rec = (struct iam_rec *) rec;
4683
4684                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4685
4686                 if (S_ISDIR(obj->oo_inode->i_mode))
4687                         osd_fid_unpack((struct lu_fid *) rec,
4688                                        (struct osd_fid_pack *)iam_rec);
4689                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4690                         osd_quota_unpack(obj, rec);
4691         }
4692
4693         iam_it_put(it);
4694         iam_it_fini(it);
4695         osd_ipd_put(env, bag, ipd);
4696
4697         LINVRNT(osd_invariant(obj));
4698
4699         RETURN(rc);
4700 }
4701
4702 static int osd_index_declare_iam_insert(const struct lu_env *env,
4703                                         struct dt_object *dt,
4704                                         const struct dt_rec *rec,
4705                                         const struct dt_key *key,
4706                                         struct thandle *handle)
4707 {
4708         struct osd_thandle *oh;
4709
4710         LASSERT(handle != NULL);
4711
4712         oh = container_of0(handle, struct osd_thandle, ot_super);
4713         LASSERT(oh->ot_handle == NULL);
4714
4715         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4716                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4717
4718         return 0;
4719 }
4720
4721 /**
4722  *      Inserts (key, value) pair in \a dt index object.
4723  *
4724  *      \param  dt      osd index object
4725  *      \param  key     key for index
4726  *      \param  rec     record reference
4727  *      \param  th      transaction handler
4728  *
4729  *      \retval  0  success
4730  *      \retval -ve failure
4731  */
4732 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4733                                 const struct dt_rec *rec,
4734                                 const struct dt_key *key, struct thandle *th,
4735                                 int ignore_quota)
4736 {
4737         struct osd_object     *obj = osd_dt_obj(dt);
4738         struct iam_path_descr *ipd;
4739         struct osd_thandle    *oh;
4740         struct iam_container  *bag;
4741         struct osd_thread_info *oti = osd_oti_get(env);
4742         struct iam_rec         *iam_rec;
4743         int                     rc;
4744         ENTRY;
4745
4746         if (!dt_object_exists(dt))
4747                 RETURN(-ENOENT);
4748
4749         LINVRNT(osd_invariant(obj));
4750         LASSERT(!dt_object_remote(dt));
4751
4752         bag = &obj->oo_dir->od_container;
4753         LASSERT(bag->ic_object == obj->oo_inode);
4754         LASSERT(th != NULL);
4755
4756         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4757
4758         ipd = osd_idx_ipd_get(env, bag);
4759         if (unlikely(ipd == NULL))
4760                 RETURN(-ENOMEM);
4761
4762         oh = container_of0(th, struct osd_thandle, ot_super);
4763         LASSERT(oh->ot_handle != NULL);
4764         LASSERT(oh->ot_handle->h_transaction != NULL);
4765         if (S_ISDIR(obj->oo_inode->i_mode)) {
4766                 iam_rec = (struct iam_rec *)oti->oti_ldp;
4767                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4768         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4769                 /* pack quota uid/gid */
4770                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4771                 key = (const struct dt_key *)&oti->oti_quota_id;
4772                 /* pack quota record */
4773                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4774                 iam_rec = (struct iam_rec *)rec;
4775         } else {
4776                 iam_rec = (struct iam_rec *)rec;
4777         }
4778
4779         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4780                         iam_rec, ipd);
4781         osd_ipd_put(env, bag, ipd);
4782         LINVRNT(osd_invariant(obj));
4783         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4784         RETURN(rc);
4785 }
4786
4787 /**
4788  * Calls ldiskfs_add_entry() to add directory entry
4789  * into the directory. This is required for
4790  * interoperability mode (b11826)
4791  *
4792  * \retval   0, on success
4793  * \retval -ve, on error
4794  */
4795 static int __osd_ea_add_rec(struct osd_thread_info *info,
4796                             struct osd_object *pobj, struct inode  *cinode,
4797                             const char *name, const struct lu_fid *fid,
4798                             struct htree_lock *hlock, struct thandle *th)
4799 {
4800         struct ldiskfs_dentry_param *ldp;
4801         struct dentry               *child;
4802         struct osd_thandle          *oth;
4803         int                          rc;
4804
4805         oth = container_of(th, struct osd_thandle, ot_super);
4806         LASSERT(oth->ot_handle != NULL);
4807         LASSERT(oth->ot_handle->h_transaction != NULL);
4808         LASSERT(pobj->oo_inode);
4809
4810         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4811         if (unlikely(pobj->oo_inode ==
4812                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4813                 ldp->edp_magic = 0;
4814         else
4815                 osd_get_ldiskfs_dirent_param(ldp, fid);
4816         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4817         child->d_fsdata = (void *)ldp;
4818         ll_vfs_dq_init(pobj->oo_inode);
4819         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4820                                    child, cinode, hlock);
4821         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4822                 struct ldiskfs_dir_entry_2      *de;
4823                 struct buffer_head              *bh;
4824                 int                              rc1;
4825
4826                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4827                                             NULL, hlock);
4828                 if (!IS_ERR(bh)) {
4829                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4830                                                                bh);
4831                         if (rc1 == 0) {
4832                                 if (S_ISDIR(cinode->i_mode))
4833                                         de->file_type = LDISKFS_DIRENT_LUFID |
4834                                                         LDISKFS_FT_REG_FILE;
4835                                 else
4836                                         de->file_type = LDISKFS_DIRENT_LUFID |
4837                                                         LDISKFS_FT_DIR;
4838                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4839                                                               NULL, bh);
4840                         }
4841                         brelse(bh);
4842                 }
4843         }
4844
4845         RETURN(rc);
4846 }
4847
4848 /**
4849  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4850  * into the directory.Also sets flags into osd object to
4851  * indicate dot and dotdot are created. This is required for
4852  * interoperability mode (b11826)
4853  *
4854  * \param dir   directory for dot and dotdot fixup.
4855  * \param obj   child object for linking
4856  *
4857  * \retval   0, on success
4858  * \retval -ve, on error
4859  */
4860 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4861                               struct osd_object *dir,
4862                               struct inode *parent_dir, const char *name,
4863                               const struct lu_fid *dot_fid,
4864                               const struct lu_fid *dot_dot_fid,
4865                               struct thandle *th)
4866 {
4867         struct inode                *inode = dir->oo_inode;
4868         struct osd_thandle          *oth;
4869         int result = 0;
4870
4871         oth = container_of(th, struct osd_thandle, ot_super);
4872         LASSERT(oth->ot_handle->h_transaction != NULL);
4873         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4874
4875         if (strcmp(name, dot) == 0) {
4876                 if (dir->oo_compat_dot_created) {
4877                         result = -EEXIST;
4878                 } else {
4879                         LASSERT(inode->i_ino == parent_dir->i_ino);
4880                         dir->oo_compat_dot_created = 1;
4881                         result = 0;
4882                 }
4883         } else if (strcmp(name, dotdot) == 0) {
4884                 if (!dir->oo_compat_dot_created)
4885                         return -EINVAL;
4886                 /* in case of rename, dotdot is already created */
4887                 if (dir->oo_compat_dotdot_created) {
4888                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4889                                                 dot_dot_fid, NULL, th);
4890                 }
4891
4892                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4893                         struct lu_fid tfid = *dot_dot_fid;
4894
4895                         tfid.f_oid--;
4896                         result = osd_add_dot_dotdot_internal(info,
4897                                         dir->oo_inode, parent_dir, dot_fid,
4898                                         &tfid, oth);
4899                 } else {
4900                         result = osd_add_dot_dotdot_internal(info,
4901                                         dir->oo_inode, parent_dir, dot_fid,
4902                                         dot_dot_fid, oth);
4903                 }
4904
4905                 if (result == 0)
4906                         dir->oo_compat_dotdot_created = 1;
4907         }
4908
4909         return result;
4910 }
4911
4912
4913 /**
4914  * It will call the appropriate osd_add* function and return the
4915  * value, return by respective functions.
4916  */
4917 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4918                           struct inode *cinode, const char *name,
4919                           const struct lu_fid *fid, struct thandle *th)
4920 {
4921         struct osd_thread_info *info   = osd_oti_get(env);
4922         struct htree_lock      *hlock;
4923         int                     rc;
4924
4925         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4926
4927         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4928                                                    name[2] =='\0'))) {
4929                 if (hlock != NULL) {
4930                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4931                                            pobj->oo_inode, 0);
4932                 } else {
4933                         down_write(&pobj->oo_ext_idx_sem);
4934                 }
4935
4936                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4937                                         lu_object_fid(&pobj->oo_dt.do_lu),
4938                                         fid, th);
4939         } else {
4940                 if (hlock != NULL) {
4941                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4942                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4943                 } else {
4944                         down_write(&pobj->oo_ext_idx_sem);
4945                 }
4946
4947                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4948                         struct lu_fid *tfid = &info->oti_fid;
4949
4950                         *tfid = *fid;
4951                         tfid->f_ver = ~0;
4952                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4953                                               tfid, hlock, th);
4954                 } else {
4955                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4956                                               hlock, th);
4957                 }
4958         }
4959         if (hlock != NULL)
4960                 ldiskfs_htree_unlock(hlock);
4961         else
4962                 up_write(&pobj->oo_ext_idx_sem);
4963
4964         return rc;
4965 }
4966
4967 static int
4968 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4969                       struct osd_idmap_cache *oic)
4970 {
4971         struct osd_scrub *scrub = &dev->od_scrub;
4972         struct lu_fid *fid = &oic->oic_fid;
4973         struct osd_inode_id *id = &oic->oic_lid;
4974         struct inode *inode = NULL;
4975         int once  = 0;
4976         bool insert;
4977         int rc;
4978         ENTRY;
4979
4980         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4981                 RETURN(0);
4982
4983         if (scrub->os_pos_current > id->oii_ino)
4984                 RETURN(0);
4985
4986 again:
4987         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4988         if (rc == -ENOENT) {
4989                 __u32 gen = id->oii_gen;
4990
4991                 insert = true;
4992                 if (inode != NULL)
4993                         goto trigger;
4994
4995                 inode = osd_iget(oti, dev, id);
4996                 /* The inode has been removed (by race maybe). */
4997                 if (IS_ERR(inode)) {
4998                         rc = PTR_ERR(inode);
4999
5000                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5001                 }
5002
5003                 /* The OI mapping is lost. */
5004                 if (gen != OSD_OII_NOGEN)
5005                         goto trigger;
5006
5007                 iput(inode);
5008                 /* The inode may has been reused by others, we do not know,
5009                  * leave it to be handled by subsequent osd_fid_lookup(). */
5010                 RETURN(0);
5011         } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
5012                 RETURN(rc);
5013         } else {
5014                 insert = false;
5015         }
5016
5017 trigger:
5018         if (thread_is_running(&scrub->os_thread)) {
5019                 if (inode == NULL) {
5020                         inode = osd_iget(oti, dev, id);
5021                         /* The inode has been removed (by race maybe). */
5022                         if (IS_ERR(inode)) {
5023                                 rc = PTR_ERR(inode);
5024
5025                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5026                         }
5027                 }
5028
5029                 rc = osd_oii_insert(dev, oic, insert);
5030                 /* There is race condition between osd_oi_lookup and OI scrub.
5031                  * The OI scrub finished just after osd_oi_lookup() failure.
5032                  * Under such case, it is unnecessary to trigger OI scrub again,
5033                  * but try to call osd_oi_lookup() again. */
5034                 if (unlikely(rc == -EAGAIN))
5035                         goto again;
5036
5037                 if (!S_ISDIR(inode->i_mode))
5038                         rc = 0;
5039                 else
5040                         rc = osd_check_lmv(oti, dev, inode, oic);
5041
5042                 iput(inode);
5043                 RETURN(rc);
5044         }
5045
5046         if (!dev->od_noscrub && ++once == 1) {
5047                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5048                                      SS_CLEAR_FAILOUT);
5049                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5050                        "%.16s: trigger partial OI scrub for RPC inconsistency "
5051                        "checking FID "DFID": rc = %d\n",
5052                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
5053                        PFID(fid), rc);
5054                 if (rc == 0 || rc == -EALREADY)
5055                         goto again;
5056         }
5057
5058         if (inode != NULL)
5059                 iput(inode);
5060
5061         RETURN(rc);
5062 }
5063
5064 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5065                                struct osd_device *dev,
5066                                struct osd_idmap_cache *oic,
5067                                struct lu_fid *fid, __u32 ino)
5068 {
5069         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5070         struct inode *inode;
5071         int rc;
5072
5073         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5074         inode = osd_iget(oti, dev, &oic->oic_lid);
5075         if (IS_ERR(inode)) {
5076                 fid_zero(&oic->oic_fid);
5077                 return PTR_ERR(inode);
5078         }
5079
5080         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5081         iput(inode);
5082         if (rc != 0)
5083                 fid_zero(&oic->oic_fid);
5084         else
5085                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5086         return rc;
5087 }
5088
5089 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5090                       struct osd_inode_id *id, const struct lu_fid *fid)
5091 {
5092         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5093                id->oii_ino, id->oii_gen, info);
5094         info->oti_cache.oic_lid = *id;
5095         info->oti_cache.oic_fid = *fid;
5096         info->oti_cache.oic_dev = osd;
5097 }
5098
5099 /**
5100  * Get parent FID from the linkEA.
5101  *
5102  * For a directory which parent resides on remote MDT, to satisfy the
5103  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5104  * the other hand, to make the lookup(..) on the directory can return
5105  * the real parent FID, we append the real parent FID after its ".."
5106  * name entry in the /REMOTE_PARENT_DIR.
5107  *
5108  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5109  * backup. So after the restore, we cannot get the right parent FID from
5110  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5111  * we have stored the real parent FID in the directory object's linkEA,
5112  * we can parse the linkEA for the real parent FID.
5113  *
5114  * \param[in] env       pointer to the thread context
5115  * \param[in] obj       pointer to the object to be handled
5116  * \param[out]fid       pointer to the buffer to hold the parent FID
5117  *
5118  * \retval              0 for getting the real parent FID successfully
5119  * \retval              negative error number on failure
5120  */
5121 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5122                                     struct osd_object *obj,
5123                                     struct lu_fid *fid)
5124 {
5125         struct osd_thread_info  *oti    = osd_oti_get(env);
5126         struct lu_buf           *buf    = &oti->oti_big_buf;
5127         struct dentry           *dentry = &oti->oti_obj_dentry;
5128         struct inode            *inode  = obj->oo_inode;
5129         struct linkea_data       ldata  = { NULL };
5130         int                      rc;
5131         ENTRY;
5132
5133         fid_zero(fid);
5134         if (!S_ISDIR(inode->i_mode))
5135                 RETURN(-EIO);
5136
5137 again:
5138         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5139                              buf->lb_buf, buf->lb_len);
5140         if (rc == -ERANGE) {
5141                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5142                                      NULL, 0);
5143                 if (rc > 0) {
5144                         lu_buf_realloc(buf, rc);
5145                         if (buf->lb_buf == NULL)
5146                                 RETURN(-ENOMEM);
5147
5148                         goto again;
5149                 }
5150         }
5151
5152         if (unlikely(rc == 0))
5153                 RETURN(-ENODATA);
5154
5155         if (rc < 0)
5156                 RETURN(rc);
5157
5158         if (unlikely(buf->lb_buf == NULL)) {
5159                 lu_buf_realloc(buf, rc);
5160                 if (buf->lb_buf == NULL)
5161                         RETURN(-ENOMEM);
5162
5163                 goto again;
5164         }
5165
5166         ldata.ld_buf = buf;
5167         rc = linkea_init_with_rec(&ldata);
5168         if (!rc) {
5169                 linkea_first_entry(&ldata);
5170                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5171         }
5172
5173         RETURN(rc);
5174 }
5175
5176 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5177                                     struct inode *inode,
5178                                     const struct lu_fid *pfid,
5179                                     const char *name, const int namelen)
5180 {
5181         struct osd_thread_info *oti = osd_oti_get(env);
5182         struct lu_buf *buf = &oti->oti_big_buf;
5183         struct dentry *dentry = &oti->oti_obj_dentry;
5184         struct linkea_data ldata = { NULL };
5185         struct lu_name cname = { .ln_name = name,
5186                                  .ln_namelen = namelen };
5187         int rc;
5188         ENTRY;
5189
5190 again:
5191         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5192                              buf->lb_buf, buf->lb_len);
5193         if (rc == -ERANGE)
5194                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5195
5196         if (rc < 0)
5197                 RETURN(rc);
5198
5199         if (unlikely(rc == 0))
5200                 RETURN(-ENODATA);
5201
5202         if (buf->lb_len < rc) {
5203                 lu_buf_realloc(buf, rc);
5204                 if (buf->lb_buf == NULL)
5205                         RETURN(-ENOMEM);
5206
5207                 goto again;
5208         }
5209
5210         ldata.ld_buf = buf;
5211         rc = linkea_init_with_rec(&ldata);
5212         if (!rc)
5213                 rc = linkea_links_find(&ldata, &cname, pfid);
5214
5215         RETURN(rc);
5216 }
5217
5218 /**
5219  * Calls ->lookup() to find dentry. From dentry get inode and
5220  * read inode's ea to get fid. This is required for  interoperability
5221  * mode (b11826)
5222  *
5223  * \retval   0, on success
5224  * \retval -ve, on error
5225  */
5226 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5227                              struct dt_rec *rec, const struct dt_key *key)
5228 {
5229         struct inode                    *dir    = obj->oo_inode;
5230         struct dentry                   *dentry;
5231         struct ldiskfs_dir_entry_2      *de;
5232         struct buffer_head              *bh;
5233         struct lu_fid                   *fid = (struct lu_fid *) rec;
5234         struct htree_lock               *hlock = NULL;
5235         int                             ino;
5236         int                             rc;
5237         ENTRY;
5238
5239         LASSERT(dir->i_op != NULL);
5240         LASSERT(dir->i_op->lookup != NULL);
5241
5242         dentry = osd_child_dentry_get(env, obj,
5243                                       (char *)key, strlen((char *)key));
5244
5245         if (obj->oo_hl_head != NULL) {
5246                 hlock = osd_oti_get(env)->oti_hlock;
5247                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5248                                    dir, LDISKFS_HLOCK_LOOKUP);
5249         } else {
5250                 down_read(&obj->oo_ext_idx_sem);
5251         }
5252
5253         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5254         if (!IS_ERR(bh)) {
5255                 struct osd_thread_info *oti = osd_oti_get(env);
5256                 struct osd_inode_id *id = &oti->oti_id;
5257                 struct osd_idmap_cache *oic = &oti->oti_cache;
5258                 struct osd_device *dev = osd_obj2dev(obj);
5259
5260                 ino = le32_to_cpu(de->inode);
5261                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5262                         brelse(bh);
5263                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5264                         GOTO(out, rc);
5265                 }
5266
5267                 rc = osd_get_fid_from_dentry(de, rec);
5268
5269                 /* done with de, release bh */
5270                 brelse(bh);
5271                 if (rc != 0) {
5272                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
5273                                 const char *name = (const char *)key;
5274
5275                                 /* If the parent is on remote MDT, and there
5276                                  * is no FID-in-dirent, then we have to get
5277                                  * the parent FID from the linkEA.  */
5278                                 if (likely(strlen(name) == 2 &&
5279                                            name[0] == '.' && name[1] == '.'))
5280                                         rc = osd_get_pfid_from_linkea(env, obj,
5281                                                                       fid);
5282                         } else {
5283                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5284                         }
5285                 } else {
5286                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5287                 }
5288
5289                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5290                         fid_zero(&oic->oic_fid);
5291
5292                         GOTO(out, rc);
5293                 }
5294
5295                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5296                 rc = osd_consistency_check(oti, dev, oic);
5297                 if (rc != 0)
5298                         fid_zero(&oic->oic_fid);
5299         } else {
5300                 rc = PTR_ERR(bh);
5301         }
5302
5303         GOTO(out, rc);
5304
5305 out:
5306         if (hlock != NULL)
5307                 ldiskfs_htree_unlock(hlock);
5308         else
5309                 up_read(&obj->oo_ext_idx_sem);
5310         return rc;
5311 }
5312
5313 static int osd_index_declare_ea_insert(const struct lu_env *env,
5314                                        struct dt_object *dt,
5315                                        const struct dt_rec *rec,
5316                                        const struct dt_key *key,
5317                                        struct thandle *handle)
5318 {
5319         struct osd_thandle      *oh;
5320         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5321         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5322         const struct lu_fid     *fid    = rec1->rec_fid;
5323         int                      credits, rc = 0;
5324         struct osd_idmap_cache  *idc;
5325         ENTRY;
5326
5327         LASSERT(!dt_object_remote(dt));
5328         LASSERT(handle != NULL);
5329         LASSERT(fid != NULL);
5330         LASSERT(rec1->rec_type != 0);
5331
5332         oh = container_of0(handle, struct osd_thandle, ot_super);
5333         LASSERT(oh->ot_handle == NULL);
5334
5335         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5336
5337         /* we can't call iget() while a transactions is running
5338          * (this can lead to a deadlock), but we need to know
5339          * inum and object type. so we find this information at
5340          * declaration and cache in per-thread info */
5341         idc = osd_idc_find_or_init(env, osd, fid);
5342         if (IS_ERR(idc))
5343                 RETURN(PTR_ERR(idc));
5344         if (idc->oic_remote) {
5345                 /* a reference to remote inode is represented by an
5346                  * agent inode which we have to create */
5347                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5348                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5349         }
5350
5351         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5352
5353         if (osd_dt_obj(dt)->oo_inode != NULL) {
5354                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5355
5356                 /* We ignore block quota on meta pool (MDTs), so needn't
5357                  * calculate how many blocks will be consumed by this index
5358                  * insert */
5359                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5360                                            i_gid_read(inode),
5361                                            i_projid_read(inode), 0,
5362                                            oh, osd_dt_obj(dt), NULL,
5363                                            OSD_QID_BLK);
5364         }
5365
5366         RETURN(rc);
5367 }
5368
5369 /**
5370  * Index add function for interoperability mode (b11826).
5371  * It will add the directory entry.This entry is needed to
5372  * maintain name->fid mapping.
5373  *
5374  * \param key it is key i.e. file entry to be inserted
5375  * \param rec it is value of given key i.e. fid
5376  *
5377  * \retval   0, on success
5378  * \retval -ve, on error
5379  */
5380 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5381                                const struct dt_rec *rec,
5382                                const struct dt_key *key, struct thandle *th,
5383                                int ignore_quota)
5384 {
5385         struct osd_object       *obj = osd_dt_obj(dt);
5386         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5387         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5388         const struct lu_fid     *fid    = rec1->rec_fid;
5389         const char              *name = (const char *)key;
5390         struct osd_thread_info  *oti   = osd_oti_get(env);
5391         struct inode            *child_inode = NULL;
5392         struct osd_idmap_cache  *idc;
5393         int                     rc;
5394         ENTRY;
5395
5396         if (!dt_object_exists(dt))
5397                 RETURN(-ENOENT);
5398
5399         LASSERT(osd_invariant(obj));
5400         LASSERT(!dt_object_remote(dt));
5401         LASSERT(th != NULL);
5402
5403         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5404
5405         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5406
5407         idc = osd_idc_find(env, osd, fid);
5408         if (unlikely(idc == NULL)) {
5409                 /* this dt_insert() wasn't declared properly, so
5410                  * FID is missing in OI cache. we better do not
5411                  * lookup FID in FLDB/OI and don't risk to deadlock,
5412                  * but in some special cases (lfsck testing, etc)
5413                  * it's much simpler than fixing a caller */
5414                 CERROR("%s: "DFID" wasn't declared for insert\n",
5415                        osd_name(osd), PFID(fid));
5416                 dump_stack();
5417                 idc = osd_idc_find_or_init(env, osd, fid);
5418                 if (IS_ERR(idc))
5419                         RETURN(PTR_ERR(idc));
5420         }
5421
5422         if (idc->oic_remote) {
5423                 /* Insert remote entry */
5424                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5425                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
5426                         struct osd_thandle      *oh;
5427
5428                         /* If parent on remote MDT, we need put this object
5429                          * under AGENT */
5430                         oh = container_of(th, typeof(*oh), ot_super);
5431                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
5432                         if (rc != 0) {
5433                                 CERROR("%s: add "DFID" error: rc = %d\n",
5434                                        osd_name(osd),
5435                                        PFID(lu_object_fid(&dt->do_lu)), rc);
5436                                 RETURN(rc);
5437                         }
5438
5439                         child_inode = igrab(omm->omm_remote_parent->d_inode);
5440                 } else {
5441                         child_inode = osd_create_local_agent_inode(env, osd,
5442                                         obj, fid, rec1->rec_type & S_IFMT, th);
5443                         if (IS_ERR(child_inode))
5444                                 RETURN(PTR_ERR(child_inode));
5445                 }
5446         } else {
5447                 /* Insert local entry */
5448                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5449                         /* for a reason OI cache wasn't filled properly */
5450                         CERROR("%s: OIC for "DFID" isn't filled\n",
5451                                osd_name(osd), PFID(fid));
5452                         RETURN(-EINVAL);
5453                 }
5454                 child_inode = oti->oti_inode;
5455                 if (unlikely(child_inode == NULL)) {
5456                         struct ldiskfs_inode_info *lii;
5457                         OBD_ALLOC_PTR(lii);
5458                         if (lii == NULL)
5459                                 RETURN(-ENOMEM);
5460                         child_inode = oti->oti_inode = &lii->vfs_inode;
5461                 }
5462                 child_inode->i_sb = osd_sb(osd);
5463                 child_inode->i_ino = idc->oic_lid.oii_ino;
5464                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5465         }
5466
5467         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5468
5469         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5470                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5471
5472         if (child_inode && child_inode != oti->oti_inode)
5473                 iput(child_inode);
5474         LASSERT(osd_invariant(obj));
5475         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5476         RETURN(rc);
5477 }
5478
5479 /**
5480  *  Initialize osd Iterator for given osd index object.
5481  *
5482  *  \param  dt      osd index object
5483  */
5484
5485 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5486                                      struct dt_object *dt,
5487                                      __u32 unused)
5488 {
5489         struct osd_it_iam      *it;
5490         struct osd_object      *obj = osd_dt_obj(dt);
5491         struct lu_object       *lo  = &dt->do_lu;
5492         struct iam_path_descr  *ipd;
5493         struct iam_container   *bag = &obj->oo_dir->od_container;
5494
5495         if (!dt_object_exists(dt))
5496                 return ERR_PTR(-ENOENT);
5497
5498         OBD_ALLOC_PTR(it);
5499         if (it == NULL)
5500                 return ERR_PTR(-ENOMEM);
5501
5502         ipd = osd_it_ipd_get(env, bag);
5503         if (likely(ipd != NULL)) {
5504                 it->oi_obj = obj;
5505                 it->oi_ipd = ipd;
5506                 lu_object_get(lo);
5507                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5508                 return (struct dt_it *)it;
5509         } else {
5510                 OBD_FREE_PTR(it);
5511                 return ERR_PTR(-ENOMEM);
5512         }
5513 }
5514
5515 /**
5516  * free given Iterator.
5517  */
5518 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5519 {
5520         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5521         struct osd_object       *obj = it->oi_obj;
5522
5523         iam_it_fini(&it->oi_it);
5524         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5525         osd_object_put(env, obj);
5526         OBD_FREE_PTR(it);
5527 }
5528
5529 /**
5530  *  Move Iterator to record specified by \a key
5531  *
5532  *  \param  di      osd iterator
5533  *  \param  key     key for index
5534  *
5535  *  \retval +ve  di points to record with least key not larger than key
5536  *  \retval  0   di points to exact matched key
5537  *  \retval -ve  failure
5538  */
5539
5540 static int osd_it_iam_get(const struct lu_env *env,
5541                           struct dt_it *di, const struct dt_key *key)
5542 {
5543         struct osd_thread_info  *oti = osd_oti_get(env);
5544         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5545
5546         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5547                 /* swab quota uid/gid */
5548                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5549                 key = (struct dt_key *)&oti->oti_quota_id;
5550         }
5551
5552         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5553 }
5554
5555 /**
5556  *  Release Iterator
5557  *
5558  *  \param  di      osd iterator
5559  */
5560 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5561 {
5562         struct osd_it_iam *it = (struct osd_it_iam *)di;
5563
5564         iam_it_put(&it->oi_it);
5565 }
5566
5567 /**
5568  *  Move iterator by one record
5569  *
5570  *  \param  di      osd iterator
5571  *
5572  *  \retval +1   end of container reached
5573  *  \retval  0   success
5574  *  \retval -ve  failure
5575  */
5576
5577 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5578 {
5579         struct osd_it_iam *it = (struct osd_it_iam *)di;
5580
5581         return iam_it_next(&it->oi_it);
5582 }
5583
5584 /**
5585  * Return pointer to the key under iterator.
5586  */
5587
5588 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5589                                  const struct dt_it *di)
5590 {
5591         struct osd_thread_info *oti = osd_oti_get(env);
5592         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5593         struct osd_object      *obj = it->oi_obj;
5594         struct dt_key          *key;
5595
5596         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5597
5598         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5599                 /* swab quota uid/gid */
5600                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5601                 key = (struct dt_key *)&oti->oti_quota_id;
5602         }
5603
5604         return key;
5605 }
5606
5607 /**
5608  * Return size of key under iterator (in bytes)
5609  */
5610
5611 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5612 {
5613         struct osd_it_iam *it = (struct osd_it_iam *)di;
5614
5615         return iam_it_key_size(&it->oi_it);
5616 }
5617
5618 static inline void
5619 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5620 {
5621         /* check if file type is required */
5622         if (ent->lde_attrs & LUDA_TYPE) {
5623                 struct luda_type *lt;
5624                 int align = sizeof(*lt) - 1;
5625
5626                 len = (len + align) & ~align;
5627                 lt = (struct luda_type *)(ent->lde_name + len);
5628                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5629         }
5630
5631         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5632 }
5633
5634 /**
5635  * build lu direct from backend fs dirent.
5636  */
5637
5638 static inline void
5639 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5640                    char *name, __u16 namelen, __u16 type, __u32 attr)
5641 {
5642         ent->lde_attrs = attr | LUDA_FID;
5643         fid_cpu_to_le(&ent->lde_fid, fid);
5644
5645         ent->lde_hash = cpu_to_le64(offset);
5646         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5647
5648         strncpy(ent->lde_name, name, namelen);
5649         ent->lde_name[namelen] = '\0';
5650         ent->lde_namelen = cpu_to_le16(namelen);
5651
5652         /* append lustre attributes */
5653         osd_it_append_attrs(ent, namelen, type);
5654 }
5655
5656 /**
5657  * Return pointer to the record under iterator.
5658  */
5659 static int osd_it_iam_rec(const struct lu_env *env,
5660                           const struct dt_it *di,
5661                           struct dt_rec *dtrec, __u32 attr)
5662 {
5663         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5664         struct osd_thread_info *info = osd_oti_get(env);
5665         ENTRY;
5666
5667         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5668                 const struct osd_fid_pack *rec;
5669                 struct lu_fid             *fid = &info->oti_fid;
5670                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5671                 char                      *name;
5672                 int                        namelen;
5673                 __u64                      hash;
5674                 int                        rc;
5675
5676                 name = (char *)iam_it_key_get(&it->oi_it);
5677                 if (IS_ERR(name))
5678                         RETURN(PTR_ERR(name));
5679
5680                 namelen = iam_it_key_size(&it->oi_it);
5681
5682                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5683                 if (IS_ERR(rec))
5684                         RETURN(PTR_ERR(rec));
5685
5686                 rc = osd_fid_unpack(fid, rec);
5687                 if (rc)
5688                         RETURN(rc);
5689
5690                 hash = iam_it_store(&it->oi_it);
5691
5692                 /* IAM does not store object type in IAM index (dir) */
5693                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5694                                    0, LUDA_FID);
5695         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5696                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5697                            (struct iam_rec *)dtrec);
5698                 osd_quota_unpack(it->oi_obj, dtrec);
5699         } else {
5700                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5701                            (struct iam_rec *)dtrec);
5702         }
5703
5704         RETURN(0);
5705 }
5706
5707 /**
5708  * Returns cookie for current Iterator position.
5709  */
5710 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5711 {
5712         struct osd_it_iam *it = (struct osd_it_iam *)di;
5713
5714         return iam_it_store(&it->oi_it);
5715 }
5716
5717 /**
5718  * Restore iterator from cookie.
5719  *
5720  * \param  di      osd iterator
5721  * \param  hash    Iterator location cookie
5722  *
5723  * \retval +ve  di points to record with least key not larger than key.
5724  * \retval  0   di points to exact matched key
5725  * \retval -ve  failure
5726  */
5727
5728 static int osd_it_iam_load(const struct lu_env *env,
5729                            const struct dt_it *di, __u64 hash)
5730 {
5731         struct osd_it_iam *it = (struct osd_it_iam *)di;
5732
5733         return iam_it_load(&it->oi_it, hash);
5734 }
5735
5736 static const struct dt_index_operations osd_index_iam_ops = {
5737         .dio_lookup         = osd_index_iam_lookup,
5738         .dio_declare_insert = osd_index_declare_iam_insert,
5739         .dio_insert         = osd_index_iam_insert,
5740         .dio_declare_delete = osd_index_declare_iam_delete,
5741         .dio_delete         = osd_index_iam_delete,
5742         .dio_it     = {
5743                 .init     = osd_it_iam_init,
5744                 .fini     = osd_it_iam_fini,
5745                 .get      = osd_it_iam_get,
5746                 .put      = osd_it_iam_put,
5747                 .next     = osd_it_iam_next,
5748                 .key      = osd_it_iam_key,
5749                 .key_size = osd_it_iam_key_size,
5750                 .rec      = osd_it_iam_rec,
5751                 .store    = osd_it_iam_store,
5752                 .load     = osd_it_iam_load
5753         }
5754 };
5755
5756
5757 /**
5758  * Creates or initializes iterator context.
5759  *
5760  * \retval struct osd_it_ea, iterator structure on success
5761  *
5762  */
5763 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5764                                     struct dt_object *dt,
5765                                     __u32 attr)
5766 {
5767         struct osd_object       *obj  = osd_dt_obj(dt);
5768         struct osd_thread_info  *info = osd_oti_get(env);
5769         struct osd_it_ea        *oie;
5770         struct file             *file;
5771         struct lu_object        *lo   = &dt->do_lu;
5772         struct dentry           *obj_dentry;
5773         ENTRY;
5774
5775         if (!dt_object_exists(dt) || obj->oo_destroyed)
5776                 RETURN(ERR_PTR(-ENOENT));
5777
5778         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5779         if (oie == NULL)
5780                 RETURN(ERR_PTR(-ENOMEM));
5781         obj_dentry = &oie->oie_dentry;
5782
5783         obj_dentry->d_inode = obj->oo_inode;
5784         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5785         obj_dentry->d_name.hash = 0;
5786
5787         oie->oie_rd_dirent       = 0;
5788         oie->oie_it_dirent       = 0;
5789         oie->oie_dirent          = NULL;
5790         if (unlikely(!info->oti_it_ea_buf_used)) {
5791                 oie->oie_buf = info->oti_it_ea_buf;
5792                 info->oti_it_ea_buf_used = 1;
5793         } else {
5794                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5795                 if (oie->oie_buf == NULL)
5796                         RETURN(ERR_PTR(-ENOMEM));
5797         }
5798         oie->oie_obj             = obj;
5799
5800         file = &oie->oie_file;
5801
5802         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5803         if (attr & LUDA_64BITHASH)
5804                 file->f_mode    = FMODE_64BITHASH;
5805         else
5806                 file->f_mode    = FMODE_32BITHASH;
5807         file->f_path.dentry     = obj_dentry;
5808         file->f_mapping         = obj->oo_inode->i_mapping;
5809         file->f_op              = obj->oo_inode->i_fop;
5810         set_file_inode(file, obj->oo_inode);
5811
5812         lu_object_get(lo);
5813         RETURN((struct dt_it *) oie);
5814 }
5815
5816 /**
5817  * Destroy or finishes iterator context.
5818  *
5819  * \param di iterator structure to be destroyed
5820  */
5821 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5822 {
5823         struct osd_thread_info  *info = osd_oti_get(env);
5824         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
5825         struct osd_object       *obj    = oie->oie_obj;
5826         struct inode            *inode  = obj->oo_inode;
5827
5828         ENTRY;
5829         oie->oie_file.f_op->release(inode, &oie->oie_file);
5830         osd_object_put(env, obj);
5831         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5832                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5833         else
5834                 info->oti_it_ea_buf_used = 0;
5835         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5836         EXIT;
5837 }
5838
5839 /**
5840  * It position the iterator at given key, so that next lookup continues from
5841  * that key Or it is similar to dio_it->load() but based on a key,
5842  * rather than file position.
5843  *
5844  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5845  * to the beginning.
5846  *
5847  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5848  */
5849 static int osd_it_ea_get(const struct lu_env *env,
5850                          struct dt_it *di, const struct dt_key *key)
5851 {
5852         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
5853
5854         ENTRY;
5855         LASSERT(((const char *)key)[0] == '\0');
5856         it->oie_file.f_pos      = 0;
5857         it->oie_rd_dirent       = 0;
5858         it->oie_it_dirent       = 0;
5859         it->oie_dirent          = NULL;
5860
5861         RETURN(+1);
5862 }
5863
5864 /**
5865  * Does nothing
5866  */
5867 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5868 {
5869 }
5870
5871 struct osd_filldir_cbs {
5872 #ifdef HAVE_DIR_CONTEXT
5873         struct dir_context ctx;
5874 #endif
5875         struct osd_it_ea  *it;
5876 };
5877 /**
5878  * It is called internally by ->readdir(). It fills the
5879  * iterator's in-memory data structure with required
5880  * information i.e. name, namelen, rec_size etc.
5881  *
5882  * \param buf in which information to be filled in.
5883  * \param name name of the file in given dir
5884  *
5885  * \retval 0 on success
5886  * \retval 1 on buffer full
5887  */
5888 #ifdef HAVE_FILLDIR_USE_CTX
5889 static int osd_ldiskfs_filldir(struct dir_context *buf,
5890 #else
5891 static int osd_ldiskfs_filldir(void *buf,
5892 #endif
5893                                const char *name, int namelen,
5894                                loff_t offset, __u64 ino, unsigned d_type)
5895 {
5896         struct osd_it_ea        *it   =
5897                 ((struct osd_filldir_cbs *)buf)->it;
5898         struct osd_object       *obj  = it->oie_obj;
5899         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5900         struct lu_fid           *fid  = &ent->oied_fid;
5901         struct osd_fid_pack     *rec;
5902         ENTRY;
5903
5904         /* this should never happen */
5905         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5906                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5907                 RETURN(-EIO);
5908         }
5909
5910         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5911             OSD_IT_EA_BUFSIZE)
5912                 RETURN(1);
5913
5914         /* "." is just the object itself. */
5915         if (namelen == 1 && name[0] == '.') {
5916                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5917         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5918                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5919                 if (osd_fid_unpack(fid, rec) != 0)
5920                         fid_zero(fid);
5921         } else {
5922                 fid_zero(fid);
5923         }
5924         d_type &= ~LDISKFS_DIRENT_LUFID;
5925
5926         /* NOT export local root. */
5927         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5928                 ino = obj->oo_inode->i_ino;
5929                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5930         }
5931
5932         ent->oied_ino     = ino;
5933         ent->oied_off     = offset;
5934         ent->oied_namelen = namelen;
5935         ent->oied_type    = d_type;
5936
5937         memcpy(ent->oied_name, name, namelen);
5938
5939         it->oie_rd_dirent++;
5940         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5941         RETURN(0);
5942 }
5943
5944 /**
5945  * Calls ->readdir() to load a directory entry at a time
5946  * and stored it in iterator's in-memory data structure.
5947  *
5948  * \param di iterator's in memory structure
5949  *
5950  * \retval   0 on success
5951  * \retval -ve on error
5952  * \retval +1 reach the end of entry
5953  */
5954 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5955                                const struct dt_it *di)
5956 {
5957         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5958         struct osd_object  *obj   = it->oie_obj;
5959         struct inode       *inode = obj->oo_inode;
5960         struct htree_lock  *hlock = NULL;
5961         struct file        *filp  = &it->oie_file;
5962         int                 rc = 0;
5963         struct osd_filldir_cbs buf = {
5964 #ifdef HAVE_DIR_CONTEXT
5965                 .ctx.actor = osd_ldiskfs_filldir,
5966 #endif
5967                 .it = it
5968         };
5969
5970         ENTRY;
5971         it->oie_dirent = it->oie_buf;
5972         it->oie_rd_dirent = 0;
5973
5974         if (obj->oo_hl_head != NULL) {
5975                 hlock = osd_oti_get(env)->oti_hlock;
5976                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5977                                    inode, LDISKFS_HLOCK_READDIR);
5978         } else {
5979                 down_read(&obj->oo_ext_idx_sem);
5980         }
5981
5982 #ifdef HAVE_DIR_CONTEXT
5983         buf.ctx.pos = filp->f_pos;
5984         rc = inode->i_fop->iterate(filp, &buf.ctx);
5985         filp->f_pos = buf.ctx.pos;
5986 #else
5987         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5988 #endif
5989
5990         if (hlock != NULL)
5991                 ldiskfs_htree_unlock(hlock);
5992         else
5993                 up_read(&obj->oo_ext_idx_sem);
5994
5995         if (it->oie_rd_dirent == 0) {
5996                 /*If it does not get any dirent, it means it has been reached
5997                  *to the end of the dir */
5998                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5999                 if (rc == 0)
6000                         rc = 1;
6001         } else {
6002                 it->oie_dirent = it->oie_buf;
6003                 it->oie_it_dirent = 1;
6004         }
6005
6006         RETURN(rc);
6007 }
6008
6009 /**
6010  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6011  * to load a directory entry at a time and stored it in
6012  * iterator's in-memory data structure.
6013  *
6014  * \param di iterator's in memory structure
6015  *
6016  * \retval +ve iterator reached to end
6017  * \retval   0 iterator not reached to end
6018  * \retval -ve on error
6019  */
6020 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6021 {
6022         struct osd_it_ea *it = (struct osd_it_ea *)di;
6023         int rc;
6024
6025         ENTRY;
6026
6027         if (it->oie_it_dirent < it->oie_rd_dirent) {
6028                 it->oie_dirent =
6029                         (void *) it->oie_dirent +
6030                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6031                                        it->oie_dirent->oied_namelen);
6032                 it->oie_it_dirent++;
6033                 RETURN(0);
6034         } else {
6035                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6036                         rc = +1;
6037                 else
6038                         rc = osd_ldiskfs_it_fill(env, di);
6039         }
6040
6041         RETURN(rc);
6042 }
6043
6044 /**
6045  * Returns the key at current position from iterator's in memory structure.
6046  *
6047  * \param di iterator's in memory structure
6048  *
6049  * \retval key i.e. struct dt_key on success
6050  */
6051 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6052                                     const struct dt_it *di)
6053 {
6054         struct osd_it_ea *it = (struct osd_it_ea *)di;
6055
6056         return (struct dt_key *)it->oie_dirent->oied_name;
6057 }
6058
6059 /**
6060  * Returns the key's size at current position from iterator's in memory structure.
6061  *
6062  * \param di iterator's in memory structure
6063  *
6064  * \retval key_size i.e. struct dt_key on success
6065  */
6066 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6067 {
6068         struct osd_it_ea *it = (struct osd_it_ea *)di;
6069
6070         return it->oie_dirent->oied_namelen;
6071 }
6072
6073 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6074 {
6075         if (LDISKFS_DIR_REC_LEN(de) >=
6076             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6077                 return true;
6078
6079         return false;
6080 }
6081
6082 static inline bool
6083 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6084                      unsigned blocksize, bool dotdot)
6085 {
6086         if (dotdot)
6087                 return osd_dotdot_has_space(de);
6088
6089         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6090             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6091                 return true;
6092
6093         return false;
6094 }
6095
6096 static int
6097 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6098                     handle_t *jh, struct dentry *dentry,
6099                     const struct lu_fid *fid, struct buffer_head *bh,
6100                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6101                     bool dotdot)
6102 {
6103         struct inode                *dir        = dentry->d_parent->d_inode;
6104         struct inode                *inode      = dentry->d_inode;
6105         struct osd_fid_pack         *rec;
6106         struct ldiskfs_dentry_param *ldp;
6107         int                          namelen    = dentry->d_name.len;
6108         int                          rc;
6109         struct osd_thread_info     *info        = osd_oti_get(env);
6110         ENTRY;
6111
6112         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6113                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6114                 RETURN(0);
6115
6116         /* There is enough space to hold the FID-in-dirent. */
6117         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6118                 rc = ldiskfs_journal_get_write_access(jh, bh);
6119                 if (rc != 0)
6120                         RETURN(rc);
6121
6122                 de->name[namelen] = 0;
6123                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6124                 rec->fp_len = sizeof(struct lu_fid) + 1;
6125                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6126                 de->file_type |= LDISKFS_DIRENT_LUFID;
6127                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6128
6129                 RETURN(rc);
6130         }
6131
6132         LASSERT(!dotdot);
6133
6134         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6135         if (rc != 0)
6136                 RETURN(rc);
6137
6138         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6139         osd_get_ldiskfs_dirent_param(ldp, fid);
6140         dentry->d_fsdata = (void *)ldp;
6141         ll_vfs_dq_init(dir);
6142         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6143         /* It is too bad, we cannot reinsert the name entry back.
6144          * That means we lose it! */
6145         if (rc != 0)
6146                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6147                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6148                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6149                        dir->i_ino, dir->i_generation, namelen,
6150                        dentry->d_name.name, PFID(fid), rc);
6151
6152         RETURN(rc);
6153 }
6154
6155 static int
6156 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6157                         struct osd_it_ea *it, struct lu_fid *fid,
6158                         struct osd_inode_id *id, __u32 *attr)
6159 {
6160         struct osd_thread_info     *info        = osd_oti_get(env);
6161         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6162         struct osd_device          *dev         = osd_obj2dev(obj);
6163         struct super_block         *sb          = osd_sb(dev);
6164         const char                 *devname     = osd_name(dev);
6165         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6166         struct inode               *dir         = obj->oo_inode;
6167         struct htree_lock          *hlock       = NULL;
6168         struct buffer_head         *bh          = NULL;
6169         handle_t                   *jh          = NULL;
6170         struct ldiskfs_dir_entry_2 *de;
6171         struct dentry              *dentry;
6172         struct inode               *inode;
6173         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6174         int                         credits;
6175         int                         rc;
6176         bool                        dotdot      = false;
6177         bool                        dirty       = false;
6178         ENTRY;
6179
6180         if (ent->oied_name[0] == '.') {
6181                 if (ent->oied_namelen == 1)
6182                         RETURN(0);
6183
6184                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6185                         dotdot = true;
6186         }
6187
6188         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6189         inode = osd_iget(info, dev, id);
6190         if (IS_ERR(inode)) {
6191                 rc = PTR_ERR(inode);
6192                 if (rc == -ENOENT || rc == -ESTALE) {
6193                         /* Maybe dangling name entry, or
6194                          * corrupted directory entry. */
6195                         *attr |= LUDA_UNKNOWN;
6196                         rc = 0;
6197                 } else {
6198                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6199                                "check_repair, dir = %lu/%u, name = %.*s, "
6200                                "ino = %llu, rc = %d\n",
6201                                devname, dir->i_ino, dir->i_generation,
6202                                ent->oied_namelen, ent->oied_name,
6203                                ent->oied_ino, rc);
6204                 }
6205
6206                 RETURN(rc);
6207         }
6208
6209         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6210                                            ent->oied_namelen);
6211         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6212         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6213                 lma = NULL;
6214         else if (rc != 0)
6215                 GOTO(out, rc);
6216
6217         /* We need to ensure that the name entry is still valid.
6218          * Because it may be removed or renamed by other already.
6219          *
6220          * The unlink or rename operation will start journal before PDO lock,
6221          * so to avoid deadlock, here we need to start journal handle before
6222          * related PDO lock also. But because we do not know whether there
6223          * will be something to be repaired before PDO lock, we just start
6224          * journal without conditions.
6225          *
6226          * We may need to remove the name entry firstly, then insert back.
6227          * One credit is for user quota file update.
6228          * One credit is for group quota file update.
6229          * Two credits are for dirty inode. */
6230         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6231                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6232
6233         if (dev->od_dirent_journal != 0) {
6234
6235 again:
6236                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6237                 if (IS_ERR(jh)) {
6238                         rc = PTR_ERR(jh);
6239                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6240                                "check_repair, dir = %lu/%u, credits = %d, "
6241                                "name = %.*s, ino = %llu: rc = %d\n",
6242                                devname, dir->i_ino, dir->i_generation, credits,
6243                                ent->oied_namelen, ent->oied_name,
6244                                ent->oied_ino, rc);
6245
6246                         GOTO(out_inode, rc);
6247                 }
6248
6249                 if (obj->oo_hl_head != NULL) {
6250                         hlock = osd_oti_get(env)->oti_hlock;
6251                         /* "0" means exclusive lock for the whole directory.
6252                          * We need to prevent others access such name entry
6253                          * during the delete + insert. Neither HLOCK_ADD nor
6254                          * HLOCK_DEL cannot guarantee the atomicity. */
6255                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6256                 } else {
6257                         down_write(&obj->oo_ext_idx_sem);
6258                 }
6259         } else {
6260                 if (obj->oo_hl_head != NULL) {
6261                         hlock = osd_oti_get(env)->oti_hlock;
6262                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6263                                            LDISKFS_HLOCK_LOOKUP);
6264                 } else {
6265                         down_read(&obj->oo_ext_idx_sem);
6266                 }
6267         }
6268
6269         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6270         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6271                 *attr |= LUDA_IGNORE;
6272
6273                 GOTO(out, rc = 0);
6274         }
6275
6276         /* For dotdot entry, if there is not enough space to hold the
6277          * FID-in-dirent, just keep them there. It only happens when the
6278          * device upgraded from 1.8 or restored from MDT file-level backup.
6279          * For the whole directory, only dotdot entry have no FID-in-dirent
6280          * and needs to get FID from LMA when readdir, it will not affect the
6281          * performance much. */
6282         if (dotdot && !osd_dotdot_has_space(de)) {
6283                 *attr |= LUDA_UNKNOWN;
6284
6285                 GOTO(out, rc = 0);
6286         }
6287
6288         if (lma != NULL) {
6289                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6290                         GOTO(out, rc = 0);
6291
6292                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6293                         struct lu_fid *tfid = &lma->lma_self_fid;
6294
6295                         if (likely(dotdot &&
6296                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6297                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6298                                 /* It must be REMOTE_PARENT_DIR and as the
6299                                  * 'dotdot' entry of remote directory */
6300                                 *attr |= LUDA_IGNORE;
6301                         } else {
6302                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6303                                        "parent directory, but got %.*s under "
6304                                        "dir = %lu/%u with the FID "DFID"\n",
6305                                        devname, ent->oied_namelen,
6306                                        ent->oied_name, dir->i_ino,
6307                                        dir->i_generation, PFID(tfid));
6308
6309                                 *attr |= LUDA_UNKNOWN;
6310                         }
6311
6312                         GOTO(out, rc = 0);
6313                 }
6314         }
6315
6316         if (!fid_is_zero(fid)) {
6317                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6318                                               ent->oied_namelen);
6319                 if (rc == -ENOENT ||
6320                     (rc == -ENODATA &&
6321                      !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6322                         /* linkEA does not recognize the dirent entry,
6323                          * it may because the dirent entry corruption
6324                          * and points to other's inode. */
6325                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6326                                "recognize the dirent, dir = %lu/%u, "
6327                                " name = %.*s, ino = %llu, "
6328                                DFID": rc = %d\n", devname, dir->i_ino,
6329                                dir->i_generation, ent->oied_namelen,
6330                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6331                         *attr |= LUDA_UNKNOWN;
6332
6333                         GOTO(out, rc = 0);
6334                 }
6335
6336                 if (rc && rc != -ENODATA) {
6337                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6338                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6339                                DFID": rc = %d\n", devname, dir->i_ino,
6340                                dir->i_generation, ent->oied_namelen,
6341                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6342                         *attr |= LUDA_UNKNOWN;
6343
6344                         GOTO(out, rc = 0);
6345                 }
6346         }
6347
6348         if (lma != NULL) {
6349                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6350                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6351                 if (*attr & LUDA_VERIFY_DRYRUN) {
6352                         *fid = lma->lma_self_fid;
6353                         *attr |= LUDA_REPAIR;
6354
6355                         GOTO(out, rc = 0);
6356                 }
6357
6358                 if (jh == NULL) {
6359                         brelse(bh);
6360                         dev->od_dirent_journal = 1;
6361                         if (hlock != NULL) {
6362                                 ldiskfs_htree_unlock(hlock);
6363                                 hlock = NULL;
6364                         } else {
6365                                 up_read(&obj->oo_ext_idx_sem);
6366                         }
6367
6368                         goto again;
6369                 }
6370
6371                 *fid = lma->lma_self_fid;
6372                 dirty = true;
6373                 /* Update or append the FID-in-dirent. */
6374                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6375                                          bh, de, hlock, dotdot);
6376                 if (rc == 0)
6377                         *attr |= LUDA_REPAIR;
6378                 else
6379                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6380                                "the dirent, dir = %lu/%u, name = %.*s, "
6381                                "ino = %llu, "DFID": rc = %d\n",
6382                                devname, dir->i_ino, dir->i_generation,
6383                                ent->oied_namelen, ent->oied_name,
6384                                ent->oied_ino, PFID(fid), rc);
6385         } else {
6386                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6387                 if (*attr & LUDA_VERIFY_DRYRUN) {
6388                         if (fid_is_sane(fid)) {
6389                                 *attr |= LUDA_REPAIR;
6390                         } else if (dev->od_index == 0) {
6391                                 lu_igif_build(fid, inode->i_ino,
6392                                               inode->i_generation);
6393                                 *attr |= LUDA_UPGRADE;
6394                         }
6395
6396                         GOTO(out, rc = 0);
6397                 }
6398
6399                 if (jh == NULL) {
6400                         brelse(bh);
6401                         dev->od_dirent_journal = 1;
6402                         if (hlock != NULL) {
6403                                 ldiskfs_htree_unlock(hlock);
6404                                 hlock = NULL;
6405                         } else {
6406                                 up_read(&obj->oo_ext_idx_sem);
6407                         }
6408
6409                         goto again;
6410                 }
6411
6412                 dirty = true;
6413                 if (unlikely(fid_is_sane(fid))) {
6414                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6415                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6416                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6417                         if (rc == 0)
6418                                 *attr |= LUDA_REPAIR;
6419                         else
6420                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6421                                        "update dirent, dir = %lu/%u, "
6422                                        "name = %.*s, ino = %llu, "
6423                                        DFID": rc = %d\n",
6424                                        devname, dir->i_ino, dir->i_generation,
6425                                        ent->oied_namelen, ent->oied_name,
6426                                        ent->oied_ino, PFID(fid), rc);
6427                 } else if (dev->od_index == 0) {
6428                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6429                         /* It is probably IGIF object. Only aappend the
6430                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6431                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6432                                                  bh, de, hlock, dotdot);
6433                         if (rc == 0)
6434                                 *attr |= LUDA_UPGRADE;
6435                         else
6436                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6437                                        "after the dirent, dir = %lu/%u, "
6438                                        "name = %.*s, ino = %llu, "
6439                                        DFID": rc = %d\n",
6440                                        devname, dir->i_ino, dir->i_generation,
6441                                        ent->oied_namelen, ent->oied_name,
6442                                        ent->oied_ino, PFID(fid), rc);
6443                 }
6444         }
6445
6446         GOTO(out, rc);
6447
6448 out:
6449         if (!IS_ERR(bh))
6450                 brelse(bh);
6451         if (hlock != NULL) {
6452                 ldiskfs_htree_unlock(hlock);
6453         } else {
6454                 if (dev->od_dirent_journal != 0)
6455                         up_write(&obj->oo_ext_idx_sem);
6456                 else
6457                         up_read(&obj->oo_ext_idx_sem);
6458         }
6459
6460         if (jh != NULL)
6461                 ldiskfs_journal_stop(jh);
6462
6463 out_inode:
6464         iput(inode);
6465         if (rc >= 0 && !dirty)
6466                 dev->od_dirent_journal = 0;
6467
6468         return rc;
6469 }
6470
6471 /**
6472  * Returns the value at current position from iterator's in memory structure.
6473  *
6474  * \param di struct osd_it_ea, iterator's in memory structure
6475  * \param attr attr requested for dirent.
6476  * \param lde lustre dirent
6477  *
6478  * \retval   0 no error and \param lde has correct lustre dirent.
6479  * \retval -ve on error
6480  */
6481 static inline int osd_it_ea_rec(const struct lu_env *env,
6482                                 const struct dt_it *di,
6483                                 struct dt_rec *dtrec, __u32 attr)
6484 {
6485         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6486         struct osd_object      *obj   = it->oie_obj;
6487         struct osd_device      *dev   = osd_obj2dev(obj);
6488         struct osd_thread_info *oti   = osd_oti_get(env);
6489         struct osd_inode_id    *id    = &oti->oti_id;
6490         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6491         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6492         __u32                   ino   = it->oie_dirent->oied_ino;
6493         int                     rc    = 0;
6494         ENTRY;
6495
6496         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6497
6498         if (attr & LUDA_VERIFY) {
6499                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6500                         attr |= LUDA_IGNORE;
6501                         /* If the parent is on remote MDT, and there
6502                          * is no FID-in-dirent, then we have to get
6503                          * the parent FID from the linkEA.  */
6504                         if (!fid_is_sane(fid) &&
6505                             it->oie_dirent->oied_namelen == 2 &&
6506                             it->oie_dirent->oied_name[0] == '.' &&
6507                             it->oie_dirent->oied_name[1] == '.')
6508                                 osd_get_pfid_from_linkea(env, obj, fid);
6509                 } else {
6510                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6511                                                      &attr);
6512                 }
6513
6514                 if (!fid_is_sane(fid))
6515                         attr |= LUDA_UNKNOWN;
6516         } else {
6517                 attr &= ~LU_DIRENT_ATTRS_MASK;
6518                 if (!fid_is_sane(fid)) {
6519                         bool is_dotdot = false;
6520                         if (it->oie_dirent->oied_namelen == 2 &&
6521                             it->oie_dirent->oied_name[0] == '.' &&
6522                             it->oie_dirent->oied_name[1] == '.')
6523                                 is_dotdot = true;
6524                         /* If the parent is on remote MDT, and there
6525                          * is no FID-in-dirent, then we have to get
6526                          * the parent FID from the linkEA.  */
6527                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6528                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6529                         } else {
6530                                 if (is_dotdot == false &&
6531                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6532                                         RETURN(-ENOENT);
6533
6534                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6535                         }
6536                 } else {
6537                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6538                 }
6539         }
6540
6541         /* Pack the entry anyway, at least the offset is right. */
6542         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6543                            it->oie_dirent->oied_name,
6544                            it->oie_dirent->oied_namelen,
6545                            it->oie_dirent->oied_type, attr);
6546
6547         if (rc < 0)
6548                 RETURN(rc);
6549
6550         if (osd_remote_fid(env, dev, fid))
6551                 RETURN(0);
6552
6553         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6554                 osd_add_oi_cache(oti, dev, id, fid);
6555
6556         RETURN(rc > 0 ? 0 : rc);
6557 }
6558
6559 /**
6560  * Returns the record size size at current position.
6561  *
6562  * This function will return record(lu_dirent) size in bytes.
6563  *
6564  * \param[in] env       execution environment
6565  * \param[in] di        iterator's in memory structure
6566  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6567  *                      calculate the lu_dirent size.
6568  *
6569  * \retval      record size(in bytes & in memory) of the current lu_dirent
6570  *              entry.
6571  */
6572 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6573                               __u32 attr)
6574 {
6575         struct osd_it_ea *it = (struct osd_it_ea *)di;
6576
6577         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6578 }
6579
6580 /**
6581  * Returns a cookie for current position of the iterator head, so that
6582  * user can use this cookie to load/start the iterator next time.
6583  *
6584  * \param di iterator's in memory structure
6585  *
6586  * \retval cookie for current position, on success
6587  */
6588 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6589 {
6590         struct osd_it_ea *it = (struct osd_it_ea *)di;
6591
6592         return it->oie_dirent->oied_off;
6593 }
6594
6595 /**
6596  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6597  * to load a directory entry at a time and stored it i inn,
6598  * in iterator's in-memory data structure.
6599  *
6600  * \param di struct osd_it_ea, iterator's in memory structure
6601  *
6602  * \retval +ve on success
6603  * \retval -ve on error
6604  */
6605 static int osd_it_ea_load(const struct lu_env *env,
6606                           const struct dt_it *di, __u64 hash)
6607 {
6608         struct osd_it_ea *it = (struct osd_it_ea *)di;
6609         int rc;
6610
6611         ENTRY;
6612         it->oie_file.f_pos = hash;
6613
6614         rc =  osd_ldiskfs_it_fill(env, di);
6615         if (rc > 0)
6616                 rc = -ENODATA;
6617
6618         if (rc == 0)
6619                 rc = +1;
6620
6621         RETURN(rc);
6622 }
6623
6624 /**
6625  * Index lookup function for interoperability mode (b11826).
6626  *
6627  * \param key,  key i.e. file name to be searched
6628  *
6629  * \retval +ve, on success
6630  * \retval -ve, on error
6631  */
6632 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6633                                struct dt_rec *rec, const struct dt_key *key)
6634 {
6635         struct osd_object *obj = osd_dt_obj(dt);
6636         int rc = 0;
6637
6638         ENTRY;
6639
6640         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6641         LINVRNT(osd_invariant(obj));
6642
6643         rc = osd_ea_lookup_rec(env, obj, rec, key);
6644         if (rc == 0)
6645                 rc = +1;
6646         RETURN(rc);
6647 }
6648
6649 /**
6650  * Index and Iterator operations for interoperability
6651  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6652  */
6653 static const struct dt_index_operations osd_index_ea_ops = {
6654         .dio_lookup         = osd_index_ea_lookup,
6655         .dio_declare_insert = osd_index_declare_ea_insert,
6656         .dio_insert         = osd_index_ea_insert,
6657         .dio_declare_delete = osd_index_declare_ea_delete,
6658         .dio_delete         = osd_index_ea_delete,
6659         .dio_it     = {
6660                 .init     = osd_it_ea_init,
6661                 .fini     = osd_it_ea_fini,
6662                 .get      = osd_it_ea_get,
6663                 .put      = osd_it_ea_put,
6664                 .next     = osd_it_ea_next,
6665                 .key      = osd_it_ea_key,
6666                 .key_size = osd_it_ea_key_size,
6667                 .rec      = osd_it_ea_rec,
6668                 .rec_size = osd_it_ea_rec_size,
6669                 .store    = osd_it_ea_store,
6670                 .load     = osd_it_ea_load
6671         }
6672 };
6673
6674 static void *osd_key_init(const struct lu_context *ctx,
6675                           struct lu_context_key *key)
6676 {
6677         struct osd_thread_info *info;
6678
6679         OBD_ALLOC_PTR(info);
6680         if (info == NULL)
6681                 return ERR_PTR(-ENOMEM);
6682
6683         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6684         if (info->oti_it_ea_buf == NULL)
6685                 goto out_free_info;
6686
6687         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6688
6689         info->oti_hlock = ldiskfs_htree_lock_alloc();
6690         if (info->oti_hlock == NULL)
6691                 goto out_free_ea;
6692
6693         return info;
6694
6695  out_free_ea:
6696         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6697  out_free_info:
6698         OBD_FREE_PTR(info);
6699         return ERR_PTR(-ENOMEM);
6700 }
6701
6702 static void osd_key_fini(const struct lu_context *ctx,
6703                          struct lu_context_key *key, void* data)
6704 {
6705         struct osd_thread_info *info = data;
6706         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6707         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6708
6709         if (info->oti_inode != NULL)
6710                 OBD_FREE_PTR(lli);
6711         if (info->oti_hlock != NULL)
6712                 ldiskfs_htree_lock_free(info->oti_hlock);
6713         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6714         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6715         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6716         lu_buf_free(&info->oti_big_buf);
6717         if (idc != NULL) {
6718                 LASSERT(info->oti_ins_cache_size > 0);
6719                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6720                 info->oti_ins_cache = NULL;
6721                 info->oti_ins_cache_size = 0;
6722         }
6723         OBD_FREE_PTR(info);
6724 }
6725
6726 static void osd_key_exit(const struct lu_context *ctx,
6727                          struct lu_context_key *key, void *data)
6728 {
6729         struct osd_thread_info *info = data;
6730
6731         LASSERT(info->oti_r_locks == 0);
6732         LASSERT(info->oti_w_locks == 0);
6733         LASSERT(info->oti_txns    == 0);
6734 }
6735
6736 /* type constructor/destructor: osd_type_init, osd_type_fini */
6737 LU_TYPE_INIT_FINI(osd, &osd_key);
6738
6739 struct lu_context_key osd_key = {
6740         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6741         .lct_init = osd_key_init,
6742         .lct_fini = osd_key_fini,
6743         .lct_exit = osd_key_exit
6744 };
6745
6746
6747 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6748                            const char *name, struct lu_device *next)
6749 {
6750         struct osd_device *osd = osd_dev(d);
6751
6752         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6753             >= sizeof(osd->od_svname))
6754                 return -E2BIG;
6755         return osd_procfs_init(osd, name);
6756 }
6757
6758 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6759 {
6760         struct seq_server_site  *ss = osd_seq_site(osd);
6761         int                     rc;
6762         ENTRY;
6763
6764         if (osd->od_is_ost || osd->od_cl_seq != NULL)
6765                 RETURN(0);
6766
6767         if (unlikely(ss == NULL))
6768                 RETURN(-ENODEV);
6769
6770         OBD_ALLOC_PTR(osd->od_cl_seq);
6771         if (osd->od_cl_seq == NULL)
6772                 RETURN(-ENOMEM);
6773
6774         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6775                              osd->od_svname, ss->ss_server_seq);
6776         if (rc != 0) {
6777                 OBD_FREE_PTR(osd->od_cl_seq);
6778                 osd->od_cl_seq = NULL;
6779                 RETURN(rc);
6780         }
6781
6782         if (ss->ss_node_id == 0) {
6783                 /* If the OSD on the sequence controller(MDT0), then allocate
6784                  * sequence here, otherwise allocate sequence after connected
6785                  * to MDT0 (see mdt_register_lwp_callback()). */
6786                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6787                                    &osd->od_cl_seq->lcs_space, env);
6788         }
6789
6790         RETURN(rc);
6791 }
6792
6793 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6794 {
6795         if (osd->od_cl_seq == NULL)
6796                 return;
6797
6798         seq_client_fini(osd->od_cl_seq);
6799         OBD_FREE_PTR(osd->od_cl_seq);
6800         osd->od_cl_seq = NULL;
6801 }
6802
6803 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6804 {
6805         ENTRY;
6806
6807         /* shutdown quota slave instance associated with the device */
6808         if (o->od_quota_slave != NULL) {
6809                 qsd_fini(env, o->od_quota_slave);
6810                 o->od_quota_slave = NULL;
6811         }
6812
6813         osd_fid_fini(env, o);
6814
6815         RETURN(0);
6816 }
6817
6818 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6819 {
6820         ENTRY;
6821
6822         if (o->od_mnt != NULL) {
6823                 shrink_dcache_sb(osd_sb(o));
6824                 osd_sync(env, &o->od_dt_dev);
6825
6826                 mntput(o->od_mnt);
6827                 o->od_mnt = NULL;
6828         }
6829
6830         EXIT;
6831 }
6832
6833 static int osd_mount(const struct lu_env *env,
6834                      struct osd_device *o, struct lustre_cfg *cfg)
6835 {
6836         const char              *name  = lustre_cfg_string(cfg, 0);
6837         const char              *dev  = lustre_cfg_string(cfg, 1);
6838         const char              *opts;
6839         unsigned long            page, s_flags, lmd_flags = 0;
6840         struct page             *__page;
6841         struct file_system_type *type;
6842         char                    *options = NULL;
6843         char                    *str;
6844         struct osd_thread_info  *info = osd_oti_get(env);
6845         struct lu_fid           *fid = &info->oti_fid;
6846         struct inode            *inode;
6847         int                      rc = 0, force_over_512tb = 0;
6848         ENTRY;
6849
6850         if (o->od_mnt != NULL)
6851                 RETURN(0);
6852
6853         if (strlen(dev) >= sizeof(o->od_mntdev))
6854                 RETURN(-E2BIG);
6855         strcpy(o->od_mntdev, dev);
6856
6857         str = lustre_cfg_string(cfg, 2);
6858         s_flags = simple_strtoul(str, NULL, 0);
6859         str = strstr(str, ":");
6860         if (str)
6861                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6862         opts = lustre_cfg_string(cfg, 3);
6863 #ifdef __BIG_ENDIAN
6864         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6865                 CERROR("%s: device %s extents feature is not guaranteed to "
6866                        "work on big-endian systems. Use \"bigendian_extents\" "
6867                        "mount option to override.\n", name, dev);
6868                 RETURN(-EINVAL);
6869         }
6870 #endif
6871 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6872         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6873                 CWARN("force_over_128tb option is deprecated. "
6874                       "Filesystems less than 512TB can be created without any "
6875                       "force options. Use force_over_512tb option for "
6876                       "filesystems greater than 512TB.\n");
6877         }
6878 #endif
6879 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6880         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6881                 CWARN("force_over_256tb option is deprecated. "
6882                       "Filesystems less than 512TB can be created without any "
6883                       "force options. Use force_over_512tb option for "
6884                       "filesystems greater than 512TB.\n");
6885         }
6886 #endif
6887
6888         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6889                 force_over_512tb = 1;
6890
6891         __page = alloc_page(GFP_KERNEL);
6892         if (__page == NULL)
6893                 GOTO(out, rc = -ENOMEM);
6894         page = (unsigned long)page_address(__page);
6895         options = (char *)page;
6896         *options = '\0';
6897         if (opts != NULL) {
6898                 /* strip out the options for back compatiblity */
6899                 static char *sout[] = {
6900                         "mballoc",
6901                         "iopen",
6902                         "noiopen",
6903                         "iopen_nopriv",
6904                         "extents",
6905                         "noextents",
6906                         /* strip out option we processed in osd */
6907                         "bigendian_extents",
6908                         "force_over_128tb",
6909                         "force_over_256tb",
6910                         "force_over_512tb",
6911                         NULL
6912                 };
6913                 strcat(options, opts);
6914                 for (rc = 0, str = options; sout[rc]; ) {
6915                         char *op = strstr(str, sout[rc]);
6916                         if (op == NULL) {
6917                                 rc++;
6918                                 str = options;
6919                                 continue;
6920                         }
6921                         if (op == options || *(op - 1) == ',') {
6922                                 str = op + strlen(sout[rc]);
6923                                 if (*str == ',' || *str == '\0') {
6924                                         *str == ',' ? str++ : str;
6925                                         memmove(op, str, strlen(str) + 1);
6926                                 }
6927                         }
6928                         for (str = op; *str != ',' && *str != '\0'; str++)
6929                                 ;
6930                 }
6931         } else {
6932                 strncat(options, "user_xattr,acl", 14);
6933         }
6934
6935         /* Glom up mount options */
6936         if (*options != '\0')
6937                 strcat(options, ",");
6938         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6939
6940         type = get_fs_type("ldiskfs");
6941         if (!type) {
6942                 CERROR("%s: cannot find ldiskfs module\n", name);
6943                 GOTO(out, rc = -ENODEV);
6944         }
6945
6946         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6947         module_put(type->owner);
6948
6949         if (IS_ERR(o->od_mnt)) {
6950                 rc = PTR_ERR(o->od_mnt);
6951                 o->od_mnt = NULL;
6952                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6953                 GOTO(out, rc);
6954         }
6955
6956         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
6957                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
6958                                  force_over_512tb == 0) {
6959                 CERROR("%s: device %s LDISKFS does not support filesystems "
6960                        "greater than 512TB and can cause data corruption. "
6961                        "Use \"force_over_512tb\" mount option to override.\n",
6962                        name, dev);
6963                 GOTO(out_mnt, rc = -EINVAL);
6964         }
6965
6966         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
6967 #ifdef HAVE_DEV_SET_RDONLY
6968                 dev_set_rdonly(osd_sb(o)->s_bdev);
6969                 o->od_dt_dev.dd_rdonly = 1;
6970                 LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name);
6971 #else
6972                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
6973                               name);
6974
6975                 GOTO(out_mnt, rc = -EOPNOTSUPP);
6976 #endif
6977         } else {
6978 #ifdef HAVE_DEV_SET_RDONLY
6979                 if (dev_check_rdonly(osd_sb(o)->s_bdev)) {
6980                         CERROR("%s: underlying device %s is marked as "
6981                                "read-only. Setup failed\n", name, dev);
6982
6983                         GOTO(out_mnt, rc = -EROFS);
6984                 }
6985 #endif
6986         }
6987
6988         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6989                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6990                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6991                 GOTO(out_mnt, rc = -EINVAL);
6992         }
6993
6994 #ifdef LDISKFS_MOUNT_DIRDATA
6995         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6996                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6997                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6998         else if (!o->od_is_ost)
6999                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7000                       "enabling the dirdata feature. If you do not want to "
7001                       "downgrade to Lustre-1.x again, you can enable it via "
7002                       "'tune2fs -O dirdata device'\n", name, dev);
7003 #endif
7004         inode = osd_sb(o)->s_root->d_inode;
7005         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7006         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7007         if (rc != 0) {
7008                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7009                 GOTO(out_mnt, rc);
7010         }
7011
7012         if (lmd_flags & LMD_FLG_NOSCRUB)
7013                 o->od_noscrub = 1;
7014
7015         GOTO(out, rc = 0);
7016
7017 out_mnt:
7018         mntput(o->od_mnt);
7019         o->od_mnt = NULL;
7020
7021 out:
7022         if (__page)
7023                 __free_page(__page);
7024
7025         return rc;
7026 }
7027
7028 static struct lu_device *osd_device_fini(const struct lu_env *env,
7029                                          struct lu_device *d)
7030 {
7031         struct osd_device *o = osd_dev(d);
7032         ENTRY;
7033
7034         osd_shutdown(env, o);
7035         osd_procfs_fini(o);
7036         osd_scrub_cleanup(env, o);
7037         osd_obj_map_fini(o);
7038         osd_umount(env, o);
7039
7040         RETURN(NULL);
7041 }
7042
7043 static int osd_device_init0(const struct lu_env *env,
7044                             struct osd_device *o,
7045                             struct lustre_cfg *cfg)
7046 {
7047         struct lu_device        *l = osd2lu_dev(o);
7048         struct osd_thread_info *info;
7049         int                     rc;
7050         int                     cplen = 0;
7051
7052         /* if the module was re-loaded, env can loose its keys */
7053         rc = lu_env_refill((struct lu_env *) env);
7054         if (rc)
7055                 GOTO(out, rc);
7056         info = osd_oti_get(env);
7057         LASSERT(info);
7058
7059         l->ld_ops = &osd_lu_ops;
7060         o->od_dt_dev.dd_ops = &osd_dt_ops;
7061
7062         spin_lock_init(&o->od_osfs_lock);
7063         mutex_init(&o->od_otable_mutex);
7064         INIT_LIST_HEAD(&o->od_orphan_list);
7065
7066         o->od_read_cache = 1;
7067         o->od_writethrough_cache = 1;
7068         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7069
7070         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7071                         sizeof(o->od_svname));
7072         if (cplen >= sizeof(o->od_svname)) {
7073                 rc = -E2BIG;
7074                 GOTO(out, rc);
7075         }
7076
7077         o->od_index = -1; /* -1 means index is invalid */
7078         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7079         if (rc == LDD_F_SV_TYPE_OST)
7080                 o->od_is_ost = 1;
7081
7082         o->od_full_scrub_ratio = OFSR_DEFAULT;
7083         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7084         rc = osd_mount(env, o, cfg);
7085         if (rc != 0)
7086                 GOTO(out, rc);
7087
7088         rc = osd_obj_map_init(env, o);
7089         if (rc != 0)
7090                 GOTO(out_mnt, rc);
7091
7092         rc = lu_site_init(&o->od_site, l);
7093         if (rc != 0)
7094                 GOTO(out_compat, rc);
7095         o->od_site.ls_bottom_dev = l;
7096
7097         rc = lu_site_init_finish(&o->od_site);
7098         if (rc != 0)
7099                 GOTO(out_site, rc);
7100
7101         INIT_LIST_HEAD(&o->od_ios_list);
7102         /* setup scrub, including OI files initialization */
7103         rc = osd_scrub_setup(env, o);
7104         if (rc < 0)
7105                 GOTO(out_site, rc);
7106
7107         rc = osd_procfs_init(o, o->od_svname);
7108         if (rc != 0) {
7109                 CERROR("%s: can't initialize procfs: rc = %d\n",
7110                        o->od_svname, rc);
7111                 GOTO(out_scrub, rc);
7112         }
7113
7114         LASSERT(l->ld_site->ls_linkage.next != NULL);
7115         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7116
7117         /* initialize quota slave instance */
7118         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7119                                      o->od_proc_entry);
7120         if (IS_ERR(o->od_quota_slave)) {
7121                 rc = PTR_ERR(o->od_quota_slave);
7122                 o->od_quota_slave = NULL;
7123                 GOTO(out_procfs, rc);
7124         }
7125
7126         RETURN(0);
7127
7128 out_procfs:
7129         osd_procfs_fini(o);
7130 out_scrub:
7131         osd_scrub_cleanup(env, o);
7132 out_site:
7133         lu_site_fini(&o->od_site);
7134 out_compat:
7135         osd_obj_map_fini(o);
7136 out_mnt:
7137         osd_umount(env, o);
7138 out:
7139         return rc;
7140 }
7141
7142 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7143                                           struct lu_device_type *t,
7144                                           struct lustre_cfg *cfg)
7145 {
7146         struct osd_device *o;
7147         int                rc;
7148
7149         OBD_ALLOC_PTR(o);
7150         if (o == NULL)
7151                 return ERR_PTR(-ENOMEM);
7152
7153         rc = dt_device_init(&o->od_dt_dev, t);
7154         if (rc == 0) {
7155                 /* Because the ctx might be revived in dt_device_init,
7156                  * refill the env here */
7157                 lu_env_refill((struct lu_env *)env);
7158                 rc = osd_device_init0(env, o, cfg);
7159                 if (rc)
7160                         dt_device_fini(&o->od_dt_dev);
7161         }
7162
7163         if (unlikely(rc != 0))
7164                 OBD_FREE_PTR(o);
7165
7166         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7167 }
7168
7169 static struct lu_device *osd_device_free(const struct lu_env *env,
7170                                          struct lu_device *d)
7171 {
7172         struct osd_device *o = osd_dev(d);
7173         ENTRY;
7174
7175         /* XXX: make osd top device in order to release reference */
7176         d->ld_site->ls_top_dev = d;
7177         lu_site_purge(env, d->ld_site, -1);
7178         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7179                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7180                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7181         }
7182         lu_site_fini(&o->od_site);
7183         dt_device_fini(&o->od_dt_dev);
7184         OBD_FREE_PTR(o);
7185         RETURN(NULL);
7186 }
7187
7188 static int osd_process_config(const struct lu_env *env,
7189                               struct lu_device *d, struct lustre_cfg *cfg)
7190 {
7191         struct osd_device               *o = osd_dev(d);
7192         int                             rc;
7193         ENTRY;
7194
7195         switch (cfg->lcfg_command) {
7196         case LCFG_SETUP:
7197                 rc = osd_mount(env, o, cfg);
7198                 break;
7199         case LCFG_CLEANUP:
7200                 lu_dev_del_linkage(d->ld_site, d);
7201                 rc = osd_shutdown(env, o);
7202                 break;
7203         case LCFG_PARAM:
7204                 LASSERT(&o->od_dt_dev);
7205                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7206                                               cfg, &o->od_dt_dev);
7207                 if (rc > 0 || rc == -ENOSYS)
7208                         rc = class_process_proc_param(PARAM_OST,
7209                                                       lprocfs_osd_obd_vars,
7210                                                       cfg, &o->od_dt_dev);
7211                 break;
7212         default:
7213                 rc = -ENOSYS;
7214         }
7215
7216         RETURN(rc);
7217 }
7218
7219 static int osd_recovery_complete(const struct lu_env *env,
7220                                  struct lu_device *d)
7221 {
7222         struct osd_device       *osd = osd_dev(d);
7223         int                      rc = 0;
7224         ENTRY;
7225
7226         if (osd->od_quota_slave == NULL)
7227                 RETURN(0);
7228
7229         /* start qsd instance on recovery completion, this notifies the quota
7230          * slave code that we are about to process new requests now */
7231         rc = qsd_start(env, osd->od_quota_slave);
7232         RETURN(rc);
7233 }
7234
7235 /*
7236  * we use exports to track all osd users
7237  */
7238 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7239                            struct obd_device *obd, struct obd_uuid *cluuid,
7240                            struct obd_connect_data *data, void *localdata)
7241 {
7242         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7243         struct lustre_handle  conn;
7244         int                   rc;
7245         ENTRY;
7246
7247         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7248
7249         rc = class_connect(&conn, obd, cluuid);
7250         if (rc)
7251                 RETURN(rc);
7252
7253         *exp = class_conn2export(&conn);
7254
7255         spin_lock(&osd->od_osfs_lock);
7256         osd->od_connects++;
7257         spin_unlock(&osd->od_osfs_lock);
7258
7259         RETURN(0);
7260 }
7261
7262 /*
7263  * once last export (we don't count self-export) disappeared
7264  * osd can be released
7265  */
7266 static int osd_obd_disconnect(struct obd_export *exp)
7267 {
7268         struct obd_device *obd = exp->exp_obd;
7269         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7270         int                rc, release = 0;
7271         ENTRY;
7272
7273         /* Only disconnect the underlying layers on the final disconnect. */
7274         spin_lock(&osd->od_osfs_lock);
7275         osd->od_connects--;
7276         if (osd->od_connects == 0)
7277                 release = 1;
7278         spin_unlock(&osd->od_osfs_lock);
7279
7280         rc = class_disconnect(exp); /* bz 9811 */
7281
7282         if (rc == 0 && release)
7283                 class_manual_cleanup(obd);
7284         RETURN(rc);
7285 }
7286
7287 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7288                        struct lu_device *dev)
7289 {
7290         struct osd_device       *osd    = osd_dev(dev);
7291         struct lr_server_data   *lsd    =
7292                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7293         int                      result = 0;
7294         ENTRY;
7295
7296         if (osd->od_quota_slave != NULL) {
7297                 /* set up quota slave objects */
7298                 result = qsd_prepare(env, osd->od_quota_slave);
7299                 if (result != 0)
7300                         RETURN(result);
7301         }
7302
7303         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7304 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7305                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7306                         osd->od_index_in_idif = 1;
7307                 } else {
7308                         osd->od_index_in_idif = 0;
7309                         result = osd_register_proc_index_in_idif(osd);
7310                         if (result != 0)
7311                                 RETURN(result);
7312                 }
7313 #else
7314                 osd->od_index_in_idif = 1;
7315 #endif
7316         }
7317
7318         result = osd_fid_init(env, osd);
7319
7320         RETURN(result);
7321 }
7322
7323 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7324                          struct lu_fid *fid, struct md_op_data *op_data)
7325 {
7326         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7327
7328         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7329 }
7330
7331 static const struct lu_object_operations osd_lu_obj_ops = {
7332         .loo_object_init      = osd_object_init,
7333         .loo_object_delete    = osd_object_delete,
7334         .loo_object_release   = osd_object_release,
7335         .loo_object_free      = osd_object_free,
7336         .loo_object_print     = osd_object_print,
7337         .loo_object_invariant = osd_object_invariant
7338 };
7339
7340 const struct lu_device_operations osd_lu_ops = {
7341         .ldo_object_alloc      = osd_object_alloc,
7342         .ldo_process_config    = osd_process_config,
7343         .ldo_recovery_complete = osd_recovery_complete,
7344         .ldo_prepare           = osd_prepare,
7345 };
7346
7347 static const struct lu_device_type_operations osd_device_type_ops = {
7348         .ldto_init = osd_type_init,
7349         .ldto_fini = osd_type_fini,
7350
7351         .ldto_start = osd_type_start,
7352         .ldto_stop  = osd_type_stop,
7353
7354         .ldto_device_alloc = osd_device_alloc,
7355         .ldto_device_free  = osd_device_free,
7356
7357         .ldto_device_init    = osd_device_init,
7358         .ldto_device_fini    = osd_device_fini
7359 };
7360
7361 static struct lu_device_type osd_device_type = {
7362         .ldt_tags     = LU_DEVICE_DT,
7363         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7364         .ldt_ops      = &osd_device_type_ops,
7365         .ldt_ctx_tags = LCT_LOCAL,
7366 };
7367
7368 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7369 {
7370         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7371         struct super_block *sb = osd_sb(osd);
7372
7373         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7374 }
7375
7376 /*
7377  * lprocfs legacy support.
7378  */
7379 static struct obd_ops osd_obd_device_ops = {
7380         .o_owner = THIS_MODULE,
7381         .o_connect      = osd_obd_connect,
7382         .o_disconnect   = osd_obd_disconnect,
7383         .o_fid_alloc    = osd_fid_alloc,
7384         .o_health_check = osd_health_check,
7385 };
7386
7387 static int __init osd_init(void)
7388 {
7389         int rc;
7390
7391         LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7392 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7393         /* please, try to keep osd_thread_info smaller than a page */
7394         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7395 #endif
7396
7397         osd_oi_mod_init();
7398
7399         rc = lu_kmem_init(ldiskfs_caches);
7400         if (rc)
7401                 return rc;
7402
7403         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7404                                  lprocfs_osd_module_vars,
7405                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7406         if (rc)
7407                 lu_kmem_fini(ldiskfs_caches);
7408         return rc;
7409 }
7410
7411 static void __exit osd_exit(void)
7412 {
7413         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7414         lu_kmem_fini(ldiskfs_caches);
7415 }
7416
7417 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7418 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7419 MODULE_VERSION(LUSTRE_VERSION_STRING);
7420 MODULE_LICENSE("GPL");
7421
7422 module_init(osd_init);
7423 module_exit(osd_exit);