lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157         return oti->oti_w_locks > 0 && o->oo_owner == env;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't access mutable data
 162  */
 163 static int osd_root_get(const struct lu_env *env,
 164                         struct dt_device *dev, struct lu_fid *f)
 165 {
 166         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 167         return 0;
 168 }
 169
 170 /*
 171  * the following set of functions are used to maintain per-thread
 172  * cache of FID->ino mapping. this mechanism is needed to resolve
 173  * FID to inode at dt_insert() which in turn stores ino in the
 174  * directory entries to keep ldiskfs compatible with ext[34].
 175  * due to locking-originated restrictions we can't lookup ino
 176  * using LU cache (deadlock is possible). lookup using OI is quite
 177  * expensive. so instead we maintain this cache and methods like
 178  * dt_create() fill it. so in the majority of cases dt_insert() is
 179  * able to find needed mapping in lockless manner.
 180  */
 181 static struct osd_idmap_cache *
 182 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 183              const struct lu_fid *fid)
 184 {
 185         struct osd_thread_info  *oti   = osd_oti_get(env);
 186         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
 187         int i;
 188         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 189                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 190                         continue;
 191                 if (idc[i].oic_dev != osd)
 192                         continue;
 193
 194                 return idc + i;
 195         }
 196
 197         return NULL;
 198 }
 199
 200 static struct osd_idmap_cache *
 201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 202             const struct lu_fid *fid)
 203 {
 204         struct osd_thread_info  *oti   = osd_oti_get(env);
 205         struct osd_idmap_cache  *idc;
 206         int i;
 207
 208         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 209                 i = oti->oti_ins_cache_size * 2;
 210                 if (i == 0)
 211                         i = OSD_INS_CACHE_SIZE;
 212                 OBD_ALLOC(idc, sizeof(*idc) * i);
 213                 if (idc == NULL)
 214                         return ERR_PTR(-ENOMEM);
 215                 if (oti->oti_ins_cache != NULL) {
 216                         memcpy(idc, oti->oti_ins_cache,
 217                                oti->oti_ins_cache_used * sizeof(*idc));
 218                         OBD_FREE(oti->oti_ins_cache,
 219                                  oti->oti_ins_cache_used * sizeof(*idc));
 220                 }
 221                 oti->oti_ins_cache = idc;
 222                 oti->oti_ins_cache_size = i;
 223         }
 224
 225         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 226         idc->oic_fid = *fid;
 227         idc->oic_dev = osd;
 228         idc->oic_lid.oii_ino = 0;
 229         idc->oic_lid.oii_gen = 0;
 230         idc->oic_remote = 0;
 231
 232         return idc;
 233 }
 234
 235 /*
 236  * lookup mapping for the given fid in the cache, initialize a
 237  * new one if not found. the initialization checks whether the
 238  * object is local or remote. for local objects, OI is used to
 239  * learn ino/generation. the function is used when the caller
 240  * has no information about the object, e.g. at dt_insert().
 241  */
 242 static struct osd_idmap_cache *
 243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 244                      const struct lu_fid *fid)
 245 {
 246         struct osd_idmap_cache *idc;
 247         int rc;
 248
 249         idc = osd_idc_find(env, osd, fid);
 250         LASSERT(!IS_ERR(idc));
 251         if (idc != NULL)
 252                 return idc;
 253
 254         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 255                osd->od_svname, PFID(fid));
 256
 257         /* new mapping is needed */
 258         idc = osd_idc_add(env, osd, fid);
 259         if (IS_ERR(idc)) {
 260                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 261                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 262                 return idc;
 263         }
 264
 265         /* initialize it */
 266         rc = osd_remote_fid(env, osd, fid);
 267         if (unlikely(rc < 0))
 268                 return ERR_PTR(rc);
 269
 270         if (rc == 0) {
 271                 /* the object is local, lookup in OI */
 272                 /* XXX: probably cheaper to lookup in LU first? */
 273                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 274                                    &idc->oic_lid, 0);
 275                 if (unlikely(rc < 0)) {
 276                         CERROR("can't lookup: rc = %d\n", rc);
 277                         return ERR_PTR(rc);
 278                 }
 279         } else {
 280                 /* the object is remote */
 281                 idc->oic_remote = 1;
 282         }
 283
 284         return idc;
 285 }
 286
 287 /*
 288  * lookup mapping for given FID and fill it from the given object.
 289  * the object is lolcal by definition.
 290  */
 291 static int osd_idc_find_and_init(const struct lu_env *env,
 292                                  struct osd_device *osd,
 293                                  struct osd_object *obj)
 294 {
 295         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 296         struct osd_idmap_cache  *idc;
 297
 298         idc = osd_idc_find(env, osd, fid);
 299         LASSERT(!IS_ERR(idc));
 300         if (idc != NULL) {
 301                 if (obj->oo_inode == NULL)
 302                         return 0;
 303                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 304                         LASSERT(idc->oic_lid.oii_ino == 0);
 305                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 306                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 307                 }
 308                 return 0;
 309         }
 310
 311         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 312                osd->od_svname, PFID(fid));
 313
 314         /* new mapping is needed */
 315         idc = osd_idc_add(env, osd, fid);
 316         if (IS_ERR(idc)) {
 317                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 318                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 319                 return PTR_ERR(idc);
 320         }
 321
 322         if (obj->oo_inode != NULL) {
 323                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 324                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 325         }
 326         return 0;
 327 }
 328
 329 /*
 330  * OSD object methods.
 331  */
 332
 333 /*
 334  * Concurrency: no concurrent access is possible that early in object
 335  * life-cycle.
 336  */
 337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 338                                           const struct lu_object_header *hdr,
 339                                           struct lu_device *d)
 340 {
 341         struct osd_object *mo;
 342
 343         OBD_ALLOC_PTR(mo);
 344         if (mo != NULL) {
 345                 struct lu_object *l;
 346                 struct lu_object_header *h;
 347                 struct osd_device *o = osd_dev(d);
 348
 349                 l = &mo->oo_dt.do_lu;
 350                 if (unlikely(o->od_in_init)) {
 351                         OBD_ALLOC_PTR(h);
 352                         if (!h) {
 353                                 OBD_FREE_PTR(mo);
 354                                 return NULL;
 355                         }
 356
 357                         lu_object_header_init(h);
 358                         lu_object_init(l, h, d);
 359                         lu_object_add_top(h, l);
 360                         mo->oo_header = h;
 361                 } else {
 362                         dt_object_init(&mo->oo_dt, NULL, d);
 363                         mo->oo_header = NULL;
 364                 }
 365
 366                 mo->oo_dt.do_ops = &osd_obj_ops;
 367                 l->lo_ops = &osd_lu_obj_ops;
 368                 init_rwsem(&mo->oo_sem);
 369                 init_rwsem(&mo->oo_ext_idx_sem);
 370                 spin_lock_init(&mo->oo_guard);
 371                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 372                 return l;
 373         } else {
 374                 return NULL;
 375         }
 376 }
 377
 378 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 379                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 380 {
 381         int rc;
 382
 383         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 384                              (void *)loa, sizeof(*loa));
 385         if (rc > 0) {
 386                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 387
 388                 if (rc < sizeof(*lma))
 389                         return -EINVAL;
 390
 391                 rc = 0;
 392                 lustre_loa_swab(loa, true);
 393                 /* Check LMA compatibility */
 394                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 395                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 396                               "for fid = "DFID", ino = %lu\n",
 397                               osd_ino2name(inode),
 398                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 399                               PFID(&lma->lma_self_fid), inode->i_ino);
 400                         rc = -EOPNOTSUPP;
 401                 }
 402         } else if (rc == 0) {
 403                 rc = -ENODATA;
 404         }
 405
 406         return rc;
 407 }
 408
 409 /*
 410  * retrieve object from backend ext fs.
 411  **/
 412 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 413                        struct osd_inode_id *id)
 414 {
 415         int rc;
 416         struct inode *inode = NULL;
 417
 418         /* if we look for an inode withing a running
 419          * transaction, then we risk to deadlock */
 420         /* osd_dirent_check_repair() breaks this */
 421         /*LASSERT(current->journal_info == NULL);*/
 422
 423         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 424         if (IS_ERR(inode)) {
 425                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 426                        id->oii_ino, PTR_ERR(inode));
 427         } else if (id->oii_gen != OSD_OII_NOGEN &&
 428                    inode->i_generation != id->oii_gen) {
 429                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 430                        "i_generation = %u\n",
 431                        id->oii_ino, id->oii_gen, inode->i_generation);
 432                 iput(inode);
 433                 inode = ERR_PTR(-ESTALE);
 434         } else if (inode->i_nlink == 0) {
 435                 /* due to parallel readdir and unlink,
 436                 * we can have dead inode here. */
 437                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 438                 iput(inode);
 439                 inode = ERR_PTR(-ESTALE);
 440         } else if (is_bad_inode(inode)) {
 441                 CWARN("%s: bad inode: ino = %u\n",
 442                 osd_dev2name(dev), id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ENOENT);
 445         } else if ((rc = osd_attach_jinode(inode))) {
 446                 iput(inode);
 447                 inode = ERR_PTR(rc);
 448         } else {
 449                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 450                 if (id->oii_gen == OSD_OII_NOGEN)
 451                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 452
 453                 /* Do not update file c/mtime in ldiskfs.
 454                  * NB: we don't have any lock to protect this because we don't
 455                  * have reference on osd_object now, but contention with
 456                  * another lookup + attr_set can't happen in the tiny window
 457                  * between if (...) and set S_NOCMTIME. */
 458                 if (!(inode->i_flags & S_NOCMTIME))
 459                         inode->i_flags |= S_NOCMTIME;
 460         }
 461         return inode;
 462 }
 463
 464 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 465                           handle_t *handle, struct dentry *child,
 466                           struct inode *inode, struct htree_lock *hlock)
 467 {
 468         int rc, rc2;
 469
 470         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 471         if (rc == -ENOBUFS || rc == -ENOSPC) {
 472                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 473                 struct inode *parent = child->d_parent->d_inode;
 474                 struct lu_fid *fid = NULL;
 475
 476                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 477                 if (!rc2) {
 478                         fid = &loa->loa_lma.lma_self_fid;
 479                 } else if (rc2 == -ENODATA) {
 480                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 481                                 fid = &info->oti_fid3;
 482                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 483                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 484                                 fid = &info->oti_fid3;
 485                                 lu_igif_build(fid, parent->i_ino,
 486                                               parent->i_generation);
 487                         }
 488                 }
 489
 490                 if (fid != NULL)
 491                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 492                               "maximum entry limit\n",
 493                               osd_name(osd), parent->i_ino, PFID(fid),
 494                               rc == -ENOSPC ? "has reached" : "is approaching");
 495                 else
 496                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 497                               "maximum entry limit\n",
 498                               osd_name(osd), parent->i_ino,
 499                               rc == -ENOSPC ? "has reached" : "is approaching");
 500
 501                 /* ignore such error now */
 502                 if (rc == -ENOBUFS)
 503                         rc = 0;
 504         }
 505
 506         return rc;
 507 }
 508
 509
 510 struct inode *
 511 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 512              struct osd_inode_id *id, struct lu_fid *fid)
 513 {
 514         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 515         struct inode *inode;
 516         int rc;
 517
 518         inode = osd_iget(info, dev, id);
 519         if (IS_ERR(inode))
 520                 return inode;
 521
 522         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 523         if (!rc) {
 524                 *fid = loa->loa_lma.lma_self_fid;
 525         } else if (rc == -ENODATA) {
 526                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 527                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 528                 else
 529                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 530         } else {
 531                 iput(inode);
 532                 inode = ERR_PTR(rc);
 533         }
 534         return inode;
 535 }
 536
 537 static struct inode *osd_iget_check(struct osd_thread_info *info,
 538                                     struct osd_device *dev,
 539                                     const struct lu_fid *fid,
 540                                     struct osd_inode_id *id,
 541                                     bool trusted)
 542 {
 543         struct inode *inode;
 544         int rc = 0;
 545         ENTRY;
 546
 547         /* The cached OI mapping is trustable. If we cannot locate the inode
 548          * via the cached OI mapping, then return the failure to the caller
 549          * directly without further OI checking. */
 550
 551 again:
 552         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 553         if (IS_ERR(inode)) {
 554                 rc = PTR_ERR(inode);
 555                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 556                         goto check_oi;
 557
 558                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 559                        PFID(fid), id->oii_ino, rc);
 560                 GOTO(put, rc);
 561         }
 562
 563         if (is_bad_inode(inode)) {
 564                 rc = -ENOENT;
 565                 if (!trusted)
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 569                        PFID(fid), id->oii_ino);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (id->oii_gen != OSD_OII_NOGEN &&
 574             inode->i_generation != id->oii_gen) {
 575                 rc = -ESTALE;
 576                 if (!trusted)
 577                         goto check_oi;
 578
 579                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 580                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 581                        id->oii_ino, id->oii_gen, inode->i_generation);
 582                 GOTO(put, rc);
 583         }
 584
 585         if (inode->i_nlink == 0) {
 586                 rc = -ENOENT;
 587                 if (!trusted)
 588                         goto check_oi;
 589
 590                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 591                        PFID(fid), id->oii_ino);
 592                 GOTO(put, rc);
 593         }
 594
 595         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 596
 597 check_oi:
 598         if (rc != 0) {
 599                 __u32 saved_ino = id->oii_ino;
 600                 __u32 saved_gen = id->oii_gen;
 601
 602                 LASSERT(!trusted);
 603                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 604
 605                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 606                 /* XXX: There are four possible cases:
 607                  *      1. rc = 0.
 608                  *         Backup/restore caused the OI invalid.
 609                  *      2. rc = 0.
 610                  *         Someone unlinked the object but NOT removed
 611                  *         the OI mapping, such as mount target device
 612                  *         as ldiskfs, and modify something directly.
 613                  *      3. rc = -ENOENT.
 614                  *         Someone just removed the object between the
 615                  *         former oi_lookup and the iget. It is normal.
 616                  *      4. Other failure cases.
 617                  *
 618                  *      Generally, when the device is mounted, it will
 619                  *      auto check whether the system is restored from
 620                  *      file-level backup or not. We trust such detect
 621                  *      to distinguish the 1st case from the 2nd case:
 622                  *      if the OI files are consistent but may contain
 623                  *      stale OI mappings because of case 2, if iget()
 624                  *      returns -ENOENT or -ESTALE, then it should be
 625                  *      the case 2. */
 626                 if (rc != 0)
 627                         /* If the OI mapping was in OI file before the
 628                          * osd_iget_check(), but now, it is disappear,
 629                          * then it must be removed by race. That is a
 630                          * normal race case. */
 631                         GOTO(put, rc);
 632
 633                 /* It is the OI scrub updated the OI mapping by race.
 634                  * The new OI mapping must be valid. */
 635                 if (saved_ino != id->oii_ino ||
 636                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 637                         if (!IS_ERR(inode))
 638                                 iput(inode);
 639
 640                         trusted = true;
 641                         goto again;
 642                 }
 643
 644                 if (IS_ERR(inode)) {
 645                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 646                             SF_INCONSISTENT)
 647                                 /* It still can be the case 2, but we cannot
 648                                  * distinguish it from the case 1. So return
 649                                  * -EREMCHG to block current operation until
 650                                  *  OI scrub rebuilt the OI mappings. */
 651                                 rc = -EREMCHG;
 652                         else
 653                                 rc = -ENOENT;
 654
 655                         GOTO(put, rc);
 656                 }
 657
 658                 if (inode->i_generation == id->oii_gen)
 659                         rc = -ENOENT;
 660                 else
 661                         rc = -EREMCHG;
 662         } else {
 663                 if (id->oii_gen == OSD_OII_NOGEN)
 664                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 665
 666                 /* Do not update file c/mtime in ldiskfs.
 667                  * NB: we don't have any lock to protect this because we don't
 668                  * have reference on osd_object now, but contention with
 669                  * another lookup + attr_set can't happen in the tiny window
 670                  * between if (...) and set S_NOCMTIME. */
 671                 if (!(inode->i_flags & S_NOCMTIME))
 672                         inode->i_flags |= S_NOCMTIME;
 673         }
 674
 675         GOTO(put, rc);
 676
 677 put:
 678         if (rc != 0) {
 679                 if (!IS_ERR(inode))
 680                         iput(inode);
 681
 682                 inode = ERR_PTR(rc);
 683         }
 684
 685         return inode;
 686 }
 687
 688 /**
 689  * \retval +v: new filter_fid, does not contain self-fid
 690  * \retval 0:  filter_fid_old, contains self-fid
 691  * \retval -v: other failure cases
 692  */
 693 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 694                  struct dentry *dentry, struct lu_fid *fid)
 695 {
 696         struct filter_fid_old   *ff     = &info->oti_ff;
 697         struct ost_id           *ostid  = &info->oti_ostid;
 698         int                      rc;
 699
 700         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 701         if (rc == sizeof(*ff)) {
 702                 rc = 0;
 703                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 704                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 705                 /*
 706                  * XXX: use 0 as the index for compatibility, the caller will
 707                  *      handle index related issues when necessary.
 708                  */
 709                 if (!rc)
 710                         ostid_to_fid(fid, ostid, 0);
 711         } else if (rc == sizeof(struct filter_fid)) {
 712                 rc = 1;
 713         } else if (rc >= 0) {
 714                 rc = -EINVAL;
 715         }
 716
 717         return rc;
 718 }
 719
 720 static int osd_lma_self_repair(struct osd_thread_info *info,
 721                                struct osd_device *osd, struct inode *inode,
 722                                const struct lu_fid *fid, __u32 compat)
 723 {
 724         handle_t *jh;
 725         int       rc;
 726
 727         LASSERT(current->journal_info == NULL);
 728
 729         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 730                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 731         if (IS_ERR(jh)) {
 732                 rc = PTR_ERR(jh);
 733                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 734                       osd_name(osd), rc);
 735                 return rc;
 736         }
 737
 738         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 739         if (rc != 0)
 740                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 741                       osd_name(osd), rc);
 742         ldiskfs_journal_stop(jh);
 743         return rc;
 744 }
 745
 746 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 747 {
 748         struct osd_thread_info  *info   = osd_oti_get(env);
 749         struct osd_device       *osd    = osd_obj2dev(obj);
 750         struct lustre_ost_attrs *loa    = &info->oti_ost_attrs;
 751         struct lustre_mdt_attrs *lma    = &loa->loa_lma;
 752         struct inode            *inode  = obj->oo_inode;
 753         struct dentry           *dentry = &info->oti_obj_dentry;
 754         struct lu_fid           *fid    = NULL;
 755         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 756         int                      rc;
 757         ENTRY;
 758
 759         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 760                              (void *)loa, sizeof(*loa));
 761         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 762                 fid = &lma->lma_self_fid;
 763                 rc = osd_get_idif(info, inode, dentry, fid);
 764                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 765                         /* For the given OST-object, if it has neither LMA nor
 766                          * FID in XATTR_NAME_FID, then the given FID (which is
 767                          * contained in the @obj, from client RPC for locating
 768                          * the OST-object) is trusted. We use it to generate
 769                          * the LMA. */
 770                         osd_lma_self_repair(info, osd, inode, rfid,
 771                                             LMAC_FID_ON_OST);
 772                         RETURN(0);
 773                 }
 774         }
 775
 776         if (rc < 0)
 777                 RETURN(rc);
 778
 779         if (rc > 0) {
 780                 rc = 0;
 781                 lustre_lma_swab(lma);
 782                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 783                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 784                               S_ISREG(inode->i_mode)))) {
 785                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 786                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 787                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 788                               PFID(rfid), inode->i_ino);
 789                         rc = -EOPNOTSUPP;
 790                 } else {
 791                         fid = &lma->lma_self_fid;
 792                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 793                             osd->od_is_ost)
 794                                 obj->oo_pfid_in_lma = 1;
 795                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 796                             !osd->od_is_ost)
 797                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 798                 }
 799         }
 800
 801         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 802                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 803                         struct ost_id   *oi   = &info->oti_ostid;
 804                         struct lu_fid   *fid1 = &info->oti_fid3;
 805                         __u32            idx  = fid_idif_ost_idx(rfid);
 806
 807                         /* For old IDIF, the OST index is not part of the IDIF,
 808                          * Means that different OSTs may have the same IDIFs.
 809                          * Under such case, we need to make some compatible
 810                          * check to make sure to trigger OI scrub properly. */
 811                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 812                                 /* Given @rfid is new, LMA is old. */
 813                                 fid_to_ostid(fid, oi);
 814                                 ostid_to_fid(fid1, oi, idx);
 815                                 if (lu_fid_eq(fid1, rfid)) {
 816                                         if (osd->od_index_in_idif)
 817                                                 osd_lma_self_repair(info, osd,
 818                                                         inode, rfid,
 819                                                         LMAC_FID_ON_OST);
 820                                         RETURN(0);
 821                                 }
 822                         }
 823                 }
 824
 825                 rc = -EREMCHG;
 826         }
 827
 828         RETURN(rc);
 829 }
 830
 831 struct osd_check_lmv_buf {
 832 #ifdef HAVE_DIR_CONTEXT
 833         /* please keep it as first member */
 834         struct dir_context       ctx;
 835 #endif
 836         struct osd_thread_info  *oclb_info;
 837         struct osd_device       *oclb_dev;
 838         struct osd_idmap_cache  *oclb_oic;
 839         int                      oclb_items;
 840         bool                     oclb_found;
 841 };
 842
 843 /**
 844  * It is called internally by ->readdir() to filter out the
 845  * local slave object's FID of the striped directory.
 846  *
 847  * \retval      1 found the local slave's FID
 848  * \retval      0 continue to check next item
 849  * \retval      -ve for failure
 850  */
 851 #ifdef HAVE_FILLDIR_USE_CTX
 852 static int osd_stripe_dir_filldir(struct dir_context *buf,
 853 #else
 854 static int osd_stripe_dir_filldir(void *buf,
 855 #endif
 856                                   const char *name, int namelen,
 857                                   loff_t offset, __u64 ino, unsigned d_type)
 858 {
 859         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 860         struct osd_thread_info *oti = oclb->oclb_info;
 861         struct lu_fid *fid = &oti->oti_fid3;
 862         struct osd_inode_id *id = &oti->oti_id3;
 863         struct osd_device *dev = oclb->oclb_dev;
 864         struct osd_idmap_cache *oic = oclb->oclb_oic;
 865         struct inode *inode;
 866
 867         oclb->oclb_items++;
 868
 869         if (name[0] == '.')
 870                 return 0;
 871
 872         fid_zero(fid);
 873         sscanf(name + 1, SFID, RFID(fid));
 874         if (!fid_is_sane(fid))
 875                 return 0;
 876
 877         if (osd_remote_fid(oti->oti_env, dev, fid))
 878                 return 0;
 879
 880         osd_id_gen(id, ino, OSD_OII_NOGEN);
 881         inode = osd_iget(oti, dev, id);
 882         if (IS_ERR(inode))
 883                 return PTR_ERR(inode);
 884
 885         iput(inode);
 886         osd_add_oi_cache(oti, dev, id, fid);
 887         oic->oic_fid = *fid;
 888         oic->oic_lid = *id;
 889         oic->oic_dev = dev;
 890         osd_oii_insert(dev, oic, true);
 891         oclb->oclb_found = true;
 892
 893         return 1;
 894 }
 895
 896 /* When lookup item under striped directory, we need to locate the master
 897  * MDT-object of the striped directory firstly, then the client will send
 898  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 899  * and the item's name. If the system is restored from MDT file level backup,
 900  * then before the OI scrub completely built the OI files, the OI mappings of
 901  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 902  * not a problem for the master MDT-object. Because when locate the master
 903  * MDT-object, we will do name based lookup (for the striped directory itself)
 904  * firstly, during such process we can setup the correct OI mapping for the
 905  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 906  * the client will not trigger name based lookup on the MDT to locate the slave
 907  * MDT-object before locating item under the striped directory, then when
 908  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 909  * is invalid and does not know what the right OI mapping is, then the MDT has
 910  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 911  * the OI file, related OI mapping is unknown yet, please try again later. And
 912  * then client will re-try the RPC again and again until related OI mapping has
 913  * been updated. That is quite inefficient.
 914  *
 915  * To resolve above trouble, we will handle it as the following two cases:
 916  *
 917  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 918  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 919  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 920  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 921  *    directly. Please check osd_fid_lookup().
 922  *
 923  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 924  *    Under such case, during lookup the master MDT-object, we will lookup the
 925  *    slave MDT-object via readdir against the master MDT-object, because the
 926  *    slave MDT-objects information are stored as sub-directories with the name
 927  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 928  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 929  *    the correct OI mapping for the slave MDT-object. */
 930 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 931                          struct inode *inode, struct osd_idmap_cache *oic)
 932 {
 933         struct lu_buf *buf = &oti->oti_big_buf;
 934         struct dentry *dentry = &oti->oti_obj_dentry;
 935         struct file *filp = &oti->oti_file;
 936         const struct file_operations *fops;
 937         struct lmv_mds_md_v1 *lmv1;
 938         struct osd_check_lmv_buf oclb = {
 939 #ifdef HAVE_DIR_CONTEXT
 940                 .ctx.actor = osd_stripe_dir_filldir,
 941 #endif
 942                 .oclb_info = oti,
 943                 .oclb_dev = dev,
 944                 .oclb_oic = oic,
 945                 .oclb_found = false,
 946         };
 947         int rc = 0;
 948         ENTRY;
 949
 950 again:
 951         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 952                              buf->lb_len);
 953         if (rc == -ERANGE) {
 954                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 955                 if (rc > 0) {
 956                         lu_buf_realloc(buf, rc);
 957                         if (buf->lb_buf == NULL)
 958                                 GOTO(out, rc = -ENOMEM);
 959
 960                         goto again;
 961                 }
 962         }
 963
 964         if (unlikely(rc == 0 || rc == -ENODATA))
 965                 GOTO(out, rc = 0);
 966
 967         if (rc < 0)
 968                 GOTO(out, rc);
 969
 970         if (unlikely(buf->lb_buf == NULL)) {
 971                 lu_buf_realloc(buf, rc);
 972                 if (buf->lb_buf == NULL)
 973                         GOTO(out, rc = -ENOMEM);
 974
 975                 goto again;
 976         }
 977
 978         lmv1 = buf->lb_buf;
 979         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 980                 GOTO(out, rc = 0);
 981
 982         fops = inode->i_fop;
 983         dentry->d_inode = inode;
 984         dentry->d_sb = inode->i_sb;
 985         filp->f_pos = 0;
 986         filp->f_path.dentry = dentry;
 987         filp->f_mode = FMODE_64BITHASH;
 988         filp->f_mapping = inode->i_mapping;
 989         filp->f_op = fops;
 990         filp->private_data = NULL;
 991         set_file_inode(filp, inode);
 992
 993         do {
 994                 oclb.oclb_items = 0;
 995 #ifdef HAVE_DIR_CONTEXT
 996                 oclb.ctx.pos = filp->f_pos;
 997                 rc = fops->iterate(filp, &oclb.ctx);
 998                 filp->f_pos = oclb.ctx.pos;
 999 #else
1000                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1001 #endif
1002         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1003                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1004         fops->release(inode, filp);
1005
1006 out:
1007         if (rc < 0)
1008                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1009                        DFID": rc = %d\n", osd_ino2name(inode),
1010                        inode->i_ino, inode->i_generation,
1011                        PFID(&oic->oic_fid), rc);
1012         else
1013                 rc = 0;
1014
1015         RETURN(rc);
1016 }
1017
1018 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1019                           const struct lu_fid *fid,
1020                           const struct lu_object_conf *conf)
1021 {
1022         struct osd_thread_info *info;
1023         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1024         struct osd_device *dev;
1025         struct osd_idmap_cache *oic;
1026         struct osd_inode_id *id;
1027         struct inode *inode = NULL;
1028         struct lustre_scrub *scrub;
1029         struct scrub_file *sf;
1030         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1031         __u32 saved_ino;
1032         __u32 saved_gen;
1033         int result = 0;
1034         int rc1 = 0;
1035         bool remote = false;
1036         bool trusted = true;
1037         bool updated = false;
1038         bool checked = false;
1039         ENTRY;
1040
1041         LINVRNT(osd_invariant(obj));
1042         LASSERT(obj->oo_inode == NULL);
1043         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1044
1045         dev = osd_dev(ldev);
1046         scrub = &dev->od_scrub.os_scrub;
1047         sf = &scrub->os_file;
1048         info = osd_oti_get(env);
1049         LASSERT(info);
1050         oic = &info->oti_cache;
1051
1052         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1053                 RETURN(-ENOENT);
1054
1055         /* For the object is created as locking anchor, or for the object to
1056          * be created on disk. No need to osd_oi_lookup() at here because FID
1057          * shouldn't never be re-used, if it's really a duplicate FID from
1058          * unexpected reason, we should be able to detect it later by calling
1059          * do_create->osd_oi_insert(). */
1060         if (conf && conf->loc_flags & LOC_F_NEW)
1061                 GOTO(out, result = 0);
1062
1063         /* Search order: 1. per-thread cache. */
1064         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1065                 id = &oic->oic_lid;
1066                 goto iget;
1067         }
1068
1069         id = &info->oti_id;
1070         if (!list_empty(&scrub->os_inconsistent_items)) {
1071                 /* Search order: 2. OI scrub pending list. */
1072                 result = osd_oii_lookup(dev, fid, id);
1073                 if (!result)
1074                         goto iget;
1075         }
1076
1077         /* The OI mapping in the OI file can be updated by the OI scrub
1078          * when we locate the inode via FID. So it may be not trustable. */
1079         trusted = false;
1080
1081         /* Search order: 3. OI files. */
1082         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1083         if (result == -ENOENT) {
1084                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1085                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1086                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1087                                       sf->sf_oi_bitmap))
1088                         GOTO(out, result = 0);
1089
1090                 goto trigger;
1091         }
1092
1093         if (result)
1094                 GOTO(out, result);
1095
1096 iget:
1097         obj->oo_inode = NULL;
1098         /* for later passes through checks, not true on first pass */
1099         if (!IS_ERR_OR_NULL(inode))
1100                 iput(inode);
1101
1102         inode = osd_iget_check(info, dev, fid, id, trusted);
1103         if (!IS_ERR(inode)) {
1104                 obj->oo_inode = inode;
1105                 result = 0;
1106                 if (remote)
1107                         goto trigger;
1108
1109                 goto check_lma;
1110         }
1111
1112         result = PTR_ERR(inode);
1113         if (result == -ENOENT || result == -ESTALE)
1114                 GOTO(out, result = 0);
1115
1116         if (result != -EREMCHG)
1117                 GOTO(out, result);
1118
1119 trigger:
1120         /* We still have chance to get the valid inode: for the
1121          * object which is referenced by remote name entry, the
1122          * object on the local MDT will be linked under the dir
1123          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1124          *
1125          * We do not know whether the object for the given FID
1126          * is referenced by some remote name entry or not, and
1127          * especially for DNE II, a multiple-linked object may
1128          * have many name entries reside on many MDTs.
1129          *
1130          * To simplify the operation, OSD will not distinguish
1131          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1132          * only happened for the RPC from other MDT during the
1133          * OI scrub, or for the client side RPC with FID only,
1134          * such as FID to path, or from old connected client. */
1135         if (!remote) {
1136                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1137                 if (!rc1) {
1138                         remote = true;
1139                         trusted = true;
1140                         flags |= SS_AUTO_PARTIAL;
1141                         flags &= ~SS_AUTO_FULL;
1142                         goto iget;
1143                 }
1144         }
1145
1146         if (thread_is_running(&scrub->os_thread)) {
1147                 if (scrub->os_partial_scan && !scrub->os_in_join)
1148                         goto join;
1149
1150                 if (IS_ERR_OR_NULL(inode) || result)
1151                         GOTO(out, result = -EINPROGRESS);
1152
1153                 LASSERT(remote);
1154                 LASSERT(obj->oo_inode == inode);
1155
1156                 osd_add_oi_cache(info, dev, id, fid);
1157                 osd_oii_insert(dev, oic, true);
1158                 goto found;
1159         }
1160
1161         if (dev->od_auto_scrub_interval == AS_NEVER) {
1162                 if (!remote)
1163                         GOTO(out, result = -EREMCHG);
1164
1165                 LASSERT(!result);
1166                 LASSERT(obj->oo_inode == inode);
1167
1168                 osd_add_oi_cache(info, dev, id, fid);
1169                 goto found;
1170         }
1171
1172 join:
1173         rc1 = osd_scrub_start(env, dev, flags);
1174         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1175                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1176         if (rc1 && rc1 != -EALREADY)
1177                 GOTO(out, result = -EREMCHG);
1178
1179         if (IS_ERR_OR_NULL(inode) || result)
1180                 GOTO(out, result = -EINPROGRESS);
1181
1182         LASSERT(remote);
1183         LASSERT(obj->oo_inode == inode);
1184
1185         osd_add_oi_cache(info, dev, id, fid);
1186         osd_oii_insert(dev, oic, true);
1187         goto found;
1188
1189 check_lma:
1190         checked = true;
1191         if (unlikely(obj->oo_header))
1192                 goto found;
1193
1194         result = osd_check_lma(env, obj);
1195         if (!result)
1196                 goto found;
1197
1198         LASSERTF(id->oii_ino == inode->i_ino &&
1199                  id->oii_gen == inode->i_generation,
1200                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1201                  PFID(fid), id->oii_ino, id->oii_gen,
1202                  inode->i_ino, inode->i_generation);
1203
1204         saved_ino = inode->i_ino;
1205         saved_gen = inode->i_generation;
1206
1207         if (unlikely(result == -ENODATA)) {
1208                 /* If the OI scrub updated the OI mapping by race, it
1209                  * must be valid. Trust the inode that has no LMA EA. */
1210                 if (updated)
1211                         goto found;
1212
1213                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1214                 if (!result) {
1215                         /* The OI mapping is still there, the inode is still
1216                          * valid. It is just becaues the inode has no LMA EA. */
1217                         if (saved_ino == id->oii_ino &&
1218                             saved_gen == id->oii_gen)
1219                                 goto found;
1220
1221                         /* It is the OI scrub updated the OI mapping by race.
1222                          * The new OI mapping must be valid. */
1223                         trusted = true;
1224                         updated = true;
1225                         goto iget;
1226                 }
1227
1228                 /* "result == -ENOENT" means that the OI mappinghas been
1229                  * removed by race, so the inode belongs to other object.
1230                  *
1231                  * Others error can be returned  directly. */
1232                 if (result == -ENOENT) {
1233                         LASSERT(trusted);
1234
1235                         obj->oo_inode = NULL;
1236                         result = 0;
1237                 }
1238         }
1239
1240         if (result != -EREMCHG)
1241                 GOTO(out, result);
1242
1243         LASSERT(!updated);
1244
1245         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1246         /* "result == -ENOENT" means the cached OI mapping has been removed
1247          * from the OI file by race, above inode belongs to other object. */
1248         if (result == -ENOENT) {
1249                 LASSERT(trusted);
1250
1251                 obj->oo_inode = NULL;
1252                 GOTO(out, result = 0);
1253         }
1254
1255         if (result)
1256                 GOTO(out, result);
1257
1258         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1259                 result = -EREMCHG;
1260                 goto trigger;
1261         }
1262
1263         /* It is the OI scrub updated the OI mapping by race.
1264          * The new OI mapping must be valid. */
1265         trusted = true;
1266         updated = true;
1267         goto iget;
1268
1269 found:
1270         if (!checked) {
1271                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1272                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1273
1274                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1275                 if (!result) {
1276                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1277                             dev->od_is_ost)
1278                                 obj->oo_pfid_in_lma = 1;
1279                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1280                             !dev->od_is_ost)
1281                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1282                 } else if (result != -ENODATA) {
1283                         GOTO(out, result);
1284                 }
1285         }
1286
1287         obj->oo_compat_dot_created = 1;
1288         obj->oo_compat_dotdot_created = 1;
1289
1290         if (S_ISDIR(inode->i_mode) &&
1291             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1292                 osd_check_lmv(info, dev, inode, oic);
1293
1294         result = osd_attach_jinode(inode);
1295         if (result)
1296                 GOTO(out, result);
1297
1298         if (!ldiskfs_pdo)
1299                 GOTO(out, result = 0);
1300
1301         LASSERT(!obj->oo_hl_head);
1302         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1303
1304         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1305
1306 out:
1307         if (result || !obj->oo_inode) {
1308                 if (!IS_ERR_OR_NULL(inode))
1309                         iput(inode);
1310
1311                 obj->oo_inode = NULL;
1312                 if (trusted)
1313                         fid_zero(&oic->oic_fid);
1314         }
1315
1316         LINVRNT(osd_invariant(obj));
1317         return result;
1318 }
1319
1320 /*
1321  * Concurrency: shouldn't matter.
1322  */
1323 static void osd_object_init0(struct osd_object *obj)
1324 {
1325         LASSERT(obj->oo_inode != NULL);
1326         obj->oo_dt.do_body_ops = &osd_body_ops;
1327         obj->oo_dt.do_lu.lo_header->loh_attr |=
1328                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1329 }
1330
1331 /*
1332  * Concurrency: no concurrent access is possible that early in object
1333  * life-cycle.
1334  */
1335 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1336                            const struct lu_object_conf *conf)
1337 {
1338         struct osd_object *obj = osd_obj(l);
1339         int result;
1340
1341         LINVRNT(osd_invariant(obj));
1342
1343         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1344                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1345                 l->lo_header->loh_attr |= LOHA_EXISTS;
1346                 return 0;
1347         }
1348
1349         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1350         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1351         if (result == 0 && obj->oo_inode != NULL) {
1352                 struct osd_thread_info *oti = osd_oti_get(env);
1353                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1354
1355                 osd_object_init0(obj);
1356                 if (unlikely(obj->oo_header))
1357                         return 0;
1358
1359                 result = osd_get_lma(oti, obj->oo_inode,
1360                                      &oti->oti_obj_dentry, loa);
1361                 if (!result) {
1362                         /* Convert LMAI flags to lustre LMA flags
1363                          * and cache it to oo_lma_flags */
1364                         obj->oo_lma_flags =
1365                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1366                 } else if (result == -ENODATA) {
1367                         result = 0;
1368                 }
1369         }
1370
1371         LINVRNT(osd_invariant(obj));
1372         return result;
1373 }
1374
1375 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1376  * The left part is for value, binary mode. */
1377 struct osd_xattr_entry {
1378         struct list_head        oxe_list;
1379         size_t                  oxe_len;
1380         size_t                  oxe_namelen;
1381         bool                    oxe_exist;
1382         struct rcu_head         oxe_rcu;
1383         char                    oxe_buf[0];
1384 };
1385
1386 static int osd_oxc_get(struct osd_object *obj, const char *name,
1387                        struct lu_buf *buf)
1388 {
1389         struct osd_xattr_entry *tmp;
1390         struct osd_xattr_entry *oxe = NULL;
1391         size_t namelen = strlen(name);
1392         int rc;
1393         ENTRY;
1394
1395         rcu_read_lock();
1396         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1397                 if (namelen == tmp->oxe_namelen &&
1398                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1399                         oxe = tmp;
1400                         break;
1401                 }
1402         }
1403
1404         if (oxe == NULL)
1405                 GOTO(out, rc = -ENOENT);
1406
1407         if (!oxe->oxe_exist)
1408                 GOTO(out, rc = -ENODATA);
1409
1410         /* vallen */
1411         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1412         LASSERT(rc > 0);
1413
1414         if (buf->lb_buf == NULL)
1415                 GOTO(out, rc);
1416
1417         if (buf->lb_len < rc)
1418                 GOTO(out, rc = -ERANGE);
1419
1420         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1421         EXIT;
1422 out:
1423         rcu_read_unlock();
1424
1425         return rc;
1426 }
1427
1428 static void osd_oxc_free(struct rcu_head *head)
1429 {
1430         struct osd_xattr_entry *oxe;
1431
1432         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1433         OBD_FREE(oxe, oxe->oxe_len);
1434 }
1435
1436 static void osd_oxc_add(struct osd_object *obj, const char *name,
1437                         const char *buf, int buflen)
1438 {
1439         struct osd_xattr_entry *oxe;
1440         struct osd_xattr_entry *old = NULL;
1441         struct osd_xattr_entry *tmp;
1442         size_t namelen = strlen(name);
1443         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1444
1445         OBD_ALLOC(oxe, len);
1446         if (oxe == NULL)
1447                 return;
1448
1449         INIT_LIST_HEAD(&oxe->oxe_list);
1450         oxe->oxe_len = len;
1451         oxe->oxe_namelen = namelen;
1452         memcpy(oxe->oxe_buf, name, namelen);
1453         if (buflen > 0) {
1454                 LASSERT(buf != NULL);
1455                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1456                 oxe->oxe_exist = true;
1457         } else {
1458                 oxe->oxe_exist = false;
1459         }
1460
1461         /* this should be rarely called, just remove old and add new */
1462         spin_lock(&obj->oo_guard);
1463         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1464                 if (namelen == tmp->oxe_namelen &&
1465                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1466                         old = tmp;
1467                         break;
1468                 }
1469         }
1470         if (old != NULL) {
1471                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1472                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1473         } else {
1474                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1475         }
1476         spin_unlock(&obj->oo_guard);
1477 }
1478
1479 static void osd_oxc_del(struct osd_object *obj, const char *name)
1480 {
1481         struct osd_xattr_entry *oxe;
1482         size_t namelen = strlen(name);
1483
1484         spin_lock(&obj->oo_guard);
1485         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1486                 if (namelen == oxe->oxe_namelen &&
1487                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1488                         list_del_rcu(&oxe->oxe_list);
1489                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1490                         break;
1491                 }
1492         }
1493         spin_unlock(&obj->oo_guard);
1494 }
1495
1496 static void osd_oxc_fini(struct osd_object *obj)
1497 {
1498         struct osd_xattr_entry *oxe, *next;
1499
1500         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1501                 list_del(&oxe->oxe_list);
1502                 OBD_FREE(oxe, oxe->oxe_len);
1503         }
1504 }
1505
1506 /*
1507  * Concurrency: no concurrent access is possible that late in object
1508  * life-cycle.
1509  */
1510 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1511 {
1512         struct osd_object *obj = osd_obj(l);
1513         struct lu_object_header *h = obj->oo_header;
1514
1515         LINVRNT(osd_invariant(obj));
1516
1517         osd_oxc_fini(obj);
1518         dt_object_fini(&obj->oo_dt);
1519         if (obj->oo_hl_head != NULL)
1520                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1521         OBD_FREE_PTR(obj);
1522         if (unlikely(h)) {
1523                 lu_object_header_fini(h);
1524                 OBD_FREE_PTR(h);
1525         }
1526 }
1527
1528 /*
1529  * Concurrency: no concurrent access is possible that late in object
1530  * life-cycle.
1531  */
1532 static void osd_index_fini(struct osd_object *o)
1533 {
1534         struct iam_container *bag;
1535
1536         if (o->oo_dir != NULL) {
1537                 bag = &o->oo_dir->od_container;
1538                 if (o->oo_inode != NULL) {
1539                         if (bag->ic_object == o->oo_inode)
1540                                 iam_container_fini(bag);
1541                 }
1542                 OBD_FREE_PTR(o->oo_dir);
1543                 o->oo_dir = NULL;
1544         }
1545 }
1546
1547 /*
1548  * Concurrency: no concurrent access is possible that late in object
1549  * life-cycle (for all existing callers, that is. New callers have to provide
1550  * their own locking.)
1551  */
1552 static int osd_inode_unlinked(const struct inode *inode)
1553 {
1554         return inode->i_nlink == 0;
1555 }
1556
1557 enum {
1558         OSD_TXN_OI_DELETE_CREDITS    = 20,
1559         OSD_TXN_INODE_DELETE_CREDITS = 20
1560 };
1561
1562 /*
1563  * Journal
1564  */
1565
1566 #if OSD_THANDLE_STATS
1567 /**
1568  * Set time when the handle is allocated
1569  */
1570 static void osd_th_alloced(struct osd_thandle *oth)
1571 {
1572         oth->oth_alloced = ktime_get();
1573 }
1574
1575 /**
1576  * Set time when the handle started
1577  */
1578 static void osd_th_started(struct osd_thandle *oth)
1579 {
1580         oth->oth_started = ktime_get();
1581 }
1582
1583 /**
1584  * Check whether the we deal with this handle for too long.
1585  */
1586 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1587                                 ktime_t alloced, ktime_t started,
1588                                 ktime_t closed)
1589 {
1590         ktime_t now = ktime_get();
1591
1592         LASSERT(dev != NULL);
1593
1594         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1595                             ktime_us_delta(started, alloced));
1596         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1597                             ktime_us_delta(closed, started));
1598         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1599                             ktime_us_delta(now, closed));
1600
1601         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1602                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1603                       oth, now, alloced, started, closed);
1604                 libcfs_debug_dumpstack(NULL);
1605         }
1606 }
1607
1608 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1609 {                                                                       \
1610         ktime_t __closed = ktime_get();                                 \
1611         ktime_t __alloced = oth->oth_alloced;                           \
1612         ktime_t __started = oth->oth_started;                           \
1613                                                                         \
1614         expr;                                                           \
1615         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1616 }
1617
1618 #else /* OSD_THANDLE_STATS */
1619
1620 #define osd_th_alloced(h)                  do {} while(0)
1621 #define osd_th_started(h)                  do {} while(0)
1622 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1623
1624 #endif /* OSD_THANDLE_STATS */
1625
1626 /*
1627  * Concurrency: doesn't access mutable data.
1628  */
1629 static int osd_param_is_not_sane(const struct osd_device *dev,
1630                                  const struct thandle *th)
1631 {
1632         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1633
1634         return oh->ot_credits > osd_transaction_size(dev);
1635 }
1636
1637 /*
1638  * Concurrency: shouldn't matter.
1639  */
1640 static void osd_trans_commit_cb(struct super_block *sb,
1641                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1642 {
1643         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1644         struct thandle     *th  = &oh->ot_super;
1645         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1646         struct dt_txn_commit_cb *dcb, *tmp;
1647
1648         LASSERT(oh->ot_handle == NULL);
1649
1650         if (error)
1651                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1652
1653         dt_txn_hook_commit(th);
1654
1655         /* call per-transaction callbacks if any */
1656         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1657                                  dcb_linkage) {
1658                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1659                          "commit callback entry: magic=%x name='%s'\n",
1660                          dcb->dcb_magic, dcb->dcb_name);
1661                 list_del_init(&dcb->dcb_linkage);
1662                 dcb->dcb_func(NULL, th, dcb, error);
1663         }
1664
1665         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1666         lu_device_put(lud);
1667         th->th_dev = NULL;
1668
1669         OBD_FREE_PTR(oh);
1670 }
1671
1672 #ifndef HAVE_SB_START_WRITE
1673 # define sb_start_write(sb) do {} while (0)
1674 # define sb_end_write(sb) do {} while (0)
1675 #endif
1676
1677 static struct thandle *osd_trans_create(const struct lu_env *env,
1678                                         struct dt_device *d)
1679 {
1680         struct osd_thread_info  *oti = osd_oti_get(env);
1681         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1682         struct osd_thandle      *oh;
1683         struct thandle          *th;
1684         ENTRY;
1685
1686         if (d->dd_rdonly) {
1687                 CERROR("%s: someone try to start transaction under "
1688                        "readonly mode, should be disabled.\n",
1689                        osd_name(osd_dt_dev(d)));
1690                 dump_stack();
1691                 RETURN(ERR_PTR(-EROFS));
1692         }
1693
1694         /* on pending IO in this thread should left from prev. request */
1695         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1696
1697         sb_start_write(osd_sb(osd_dt_dev(d)));
1698
1699         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1700         if (!oh) {
1701                 sb_end_write(osd_sb(osd_dt_dev(d)));
1702                 RETURN(ERR_PTR(-ENOMEM));
1703         }
1704
1705         oh->ot_quota_trans = &oti->oti_quota_trans;
1706         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1707         th = &oh->ot_super;
1708         th->th_dev = d;
1709         th->th_result = 0;
1710         oh->ot_credits = 0;
1711         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1712         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1713         osd_th_alloced(oh);
1714
1715         memset(oti->oti_declare_ops, 0,
1716                sizeof(oti->oti_declare_ops));
1717         memset(oti->oti_declare_ops_cred, 0,
1718                sizeof(oti->oti_declare_ops_cred));
1719         memset(oti->oti_declare_ops_used, 0,
1720                sizeof(oti->oti_declare_ops_used));
1721
1722         oti->oti_ins_cache_depth++;
1723
1724         RETURN(th);
1725 }
1726
1727 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1728 {
1729         struct osd_thread_info  *oti = osd_oti_get(env);
1730         struct osd_thandle      *oh;
1731
1732         oh = container_of0(th, struct osd_thandle, ot_super);
1733         LASSERT(oh != NULL);
1734
1735         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1736               oti->oti_declare_ops[OSD_OT_CREATE],
1737               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1738               oti->oti_declare_ops_used[OSD_OT_CREATE],
1739               oti->oti_declare_ops[OSD_OT_DESTROY],
1740               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1741               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1742         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1743               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1744               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1745               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1746               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1747               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1748               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1749         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1750               oti->oti_declare_ops[OSD_OT_WRITE],
1751               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1752               oti->oti_declare_ops_used[OSD_OT_WRITE],
1753               oti->oti_declare_ops[OSD_OT_PUNCH],
1754               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1755               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1756               oti->oti_declare_ops[OSD_OT_QUOTA],
1757               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1758               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1759         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1760               oti->oti_declare_ops[OSD_OT_INSERT],
1761               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1762               oti->oti_declare_ops_used[OSD_OT_INSERT],
1763               oti->oti_declare_ops[OSD_OT_DELETE],
1764               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1765               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1766         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1767               oti->oti_declare_ops[OSD_OT_REF_ADD],
1768               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1769               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1770               oti->oti_declare_ops[OSD_OT_REF_DEL],
1771               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1772               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1773 }
1774
1775 /*
1776  * Concurrency: shouldn't matter.
1777  */
1778 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1779                            struct thandle *th)
1780 {
1781         struct osd_thread_info *oti = osd_oti_get(env);
1782         struct osd_device  *dev = osd_dt_dev(d);
1783         handle_t           *jh;
1784         struct osd_thandle *oh;
1785         int rc;
1786
1787         ENTRY;
1788
1789         LASSERT(current->journal_info == NULL);
1790
1791         oh = container_of0(th, struct osd_thandle, ot_super);
1792         LASSERT(oh != NULL);
1793         LASSERT(oh->ot_handle == NULL);
1794
1795         rc = dt_txn_hook_start(env, d, th);
1796         if (rc != 0)
1797                 GOTO(out, rc);
1798
1799         if (unlikely(osd_param_is_not_sane(dev, th))) {
1800                 static unsigned long last_printed;
1801                 static int last_credits;
1802
1803                 /* don't make noise on a tiny testing systems
1804                  * actual credits misuse will be caught anyway */
1805                 if (last_credits != oh->ot_credits &&
1806                     time_after(jiffies, last_printed +
1807                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1808                     osd_transaction_size(dev) > 512) {
1809                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1810                               oh->ot_credits, osd_transaction_size(dev));
1811                         osd_trans_dump_creds(env, th);
1812                         libcfs_debug_dumpstack(NULL);
1813                         last_credits = oh->ot_credits;
1814                         last_printed = jiffies;
1815                 }
1816                 /* XXX Limit the credits to 'max_transaction_buffers', and
1817                  *     let the underlying filesystem to catch the error if
1818                  *     we really need so many credits.
1819                  *
1820                  *     This should be removed when we can calculate the
1821                  *     credits precisely. */
1822                 oh->ot_credits = osd_transaction_size(dev);
1823         } else if (ldiskfs_track_declares_assert != 0) {
1824                 /* reserve few credits to prevent an assertion in JBD
1825                  * our debugging mechanism will be able to detected
1826                  * overuse. this can help to debug single-update
1827                  * transactions */
1828                 oh->ot_credits += 10;
1829                 if (unlikely(osd_param_is_not_sane(dev, th)))
1830                         oh->ot_credits = osd_transaction_size(dev);
1831         }
1832
1833         /*
1834          * XXX temporary stuff. Some abstraction layer should
1835          * be used.
1836          */
1837         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1838         osd_th_started(oh);
1839         if (!IS_ERR(jh)) {
1840                 oh->ot_handle = jh;
1841                 LASSERT(oti->oti_txns == 0);
1842
1843                 lu_device_get(&d->dd_lu_dev);
1844                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1845                               "osd-tx", th);
1846                 oti->oti_txns++;
1847                 rc = 0;
1848         } else {
1849                 rc = PTR_ERR(jh);
1850         }
1851 out:
1852         RETURN(rc);
1853 }
1854
1855 static int osd_seq_exists(const struct lu_env *env,
1856                           struct osd_device *osd, u64 seq)
1857 {
1858         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1859         struct seq_server_site  *ss = osd_seq_site(osd);
1860         int                     rc;
1861         ENTRY;
1862
1863         LASSERT(ss != NULL);
1864         LASSERT(ss->ss_server_fld != NULL);
1865
1866         rc = osd_fld_lookup(env, osd, seq, range);
1867         if (rc != 0) {
1868                 if (rc != -ENOENT)
1869                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1870                                osd_name(osd), seq, rc);
1871                 RETURN(0);
1872         }
1873
1874         RETURN(ss->ss_node_id == range->lsr_index);
1875 }
1876
1877 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1878 {
1879         struct dt_txn_commit_cb *dcb;
1880         struct dt_txn_commit_cb *tmp;
1881
1882         /* call per-transaction stop callbacks if any */
1883         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1884                                  dcb_linkage) {
1885                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1886                          "commit callback entry: magic=%x name='%s'\n",
1887                          dcb->dcb_magic, dcb->dcb_name);
1888                 list_del_init(&dcb->dcb_linkage);
1889                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1890         }
1891 }
1892
1893 /*
1894  * Concurrency: shouldn't matter.
1895  */
1896 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1897                           struct thandle *th)
1898 {
1899         int                     rc = 0, remove_agents = 0;
1900         struct osd_thandle     *oh;
1901         struct osd_thread_info *oti = osd_oti_get(env);
1902         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1903         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1904         struct qsd_instance    *qsd = osd->od_quota_slave;
1905         struct lquota_trans    *qtrans;
1906         ENTRY;
1907
1908         oh = container_of0(th, struct osd_thandle, ot_super);
1909
1910         remove_agents = oh->ot_remove_agents;
1911
1912         qtrans = oh->ot_quota_trans;
1913         oh->ot_quota_trans = NULL;
1914
1915         if (oh->ot_handle != NULL) {
1916                 int rc2;
1917                 handle_t *hdl = oh->ot_handle;
1918
1919                 /*
1920                  * add commit callback
1921                  * notice we don't do this in osd_trans_start()
1922                  * as underlying transaction can change during truncate
1923                  */
1924                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1925                                          &oh->ot_jcb);
1926
1927                 LASSERT(oti->oti_txns == 1);
1928                 oti->oti_txns--;
1929
1930                 rc = dt_txn_hook_stop(env, th);
1931                 if (rc != 0)
1932                         CERROR("%s: failed in transaction hook: rc = %d\n",
1933                                osd_name(osd), rc);
1934
1935                 osd_trans_stop_cb(oh, rc);
1936                 /* hook functions might modify th_sync */
1937                 hdl->h_sync = th->th_sync;
1938
1939                 oh->ot_handle = NULL;
1940                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1941                 if (rc2 != 0)
1942                         CERROR("%s: failed to stop transaction: rc = %d\n",
1943                                osd_name(osd), rc2);
1944                 if (!rc)
1945                         rc = rc2;
1946         } else {
1947                 osd_trans_stop_cb(oh, th->th_result);
1948                 OBD_FREE_PTR(oh);
1949         }
1950
1951         /* inform the quota slave device that the transaction is stopping */
1952         qsd_op_end(env, qsd, qtrans);
1953
1954         /* as we want IO to journal and data IO be concurrent, we don't block
1955          * awaiting data IO completion in osd_do_bio(), instead we wait here
1956          * once transaction is submitted to the journal. all reqular requests
1957          * don't do direct IO (except read/write), thus this wait_event becomes
1958          * no-op for them.
1959          *
1960          * IMPORTANT: we have to wait till any IO submited by the thread is
1961          * completed otherwise iobuf may be corrupted by different request
1962          */
1963         wait_event(iobuf->dr_wait,
1964                        atomic_read(&iobuf->dr_numreqs) == 0);
1965         osd_fini_iobuf(osd, iobuf);
1966         if (!rc)
1967                 rc = iobuf->dr_error;
1968
1969         if (unlikely(remove_agents != 0))
1970                 osd_process_scheduled_agent_removals(env, osd);
1971
1972         oti->oti_ins_cache_depth--;
1973         /* reset OI cache for safety */
1974         if (oti->oti_ins_cache_depth == 0)
1975                 oti->oti_ins_cache_used = 0;
1976
1977         sb_end_write(osd_sb(osd));
1978
1979         RETURN(rc);
1980 }
1981
1982 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1983 {
1984         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1985                                                ot_super);
1986
1987         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1988         LASSERT(&dcb->dcb_func != NULL);
1989         if (dcb->dcb_flags & DCB_TRANS_STOP)
1990                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1991         else
1992                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1993
1994         return 0;
1995 }
1996
1997 /*
1998  * Called just before object is freed. Releases all resources except for
1999  * object itself (that is released by osd_object_free()).
2000  *
2001  * Concurrency: no concurrent access is possible that late in object
2002  * life-cycle.
2003  */
2004 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2005 {
2006         struct osd_object *obj   = osd_obj(l);
2007         struct inode      *inode = obj->oo_inode;
2008
2009         LINVRNT(osd_invariant(obj));
2010
2011         /*
2012          * If object is unlinked remove fid->ino mapping from object index.
2013          */
2014
2015         osd_index_fini(obj);
2016         if (inode != NULL) {
2017                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
2018                 qid_t                    uid = i_uid_read(inode);
2019                 qid_t                    gid = i_gid_read(inode);
2020
2021                 obj->oo_inode = NULL;
2022                 iput(inode);
2023                 if (!obj->oo_header && qsd) {
2024                         struct osd_thread_info  *info = osd_oti_get(env);
2025                         struct lquota_id_info   *qi = &info->oti_qi;
2026
2027                         /* Release granted quota to master if necessary */
2028                         qi->lqi_id.qid_uid = uid;
2029                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2030
2031                         qi->lqi_id.qid_uid = gid;
2032                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2033
2034                         qi->lqi_id.qid_uid = i_projid_read(inode);
2035                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2036                 }
2037         }
2038 }
2039
2040 /*
2041  * Concurrency: ->loo_object_release() is called under site spin-lock.
2042  */
2043 static void osd_object_release(const struct lu_env *env,
2044                                struct lu_object *l)
2045 {
2046         struct osd_object *o = osd_obj(l);
2047         /* nobody should be releasing a non-destroyed object with nlink=0
2048          * the API allows this, but ldiskfs doesn't like and then report
2049          * this inode as deleted */
2050         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
2051                 LBUG();
2052 }
2053
2054 /*
2055  * Concurrency: shouldn't matter.
2056  */
2057 static int osd_object_print(const struct lu_env *env, void *cookie,
2058                             lu_printer_t p, const struct lu_object *l)
2059 {
2060         struct osd_object *o = osd_obj(l);
2061         struct iam_descr  *d;
2062
2063         if (o->oo_dir != NULL)
2064                 d = o->oo_dir->od_container.ic_descr;
2065         else
2066                 d = NULL;
2067         return (*p)(env, cookie,
2068                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2069                     o, o->oo_inode,
2070                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2071                     o->oo_inode ? o->oo_inode->i_generation : 0,
2072                     d ? d->id_ops->id_name : "plain");
2073 }
2074
2075 /*
2076  * Concurrency: shouldn't matter.
2077  */
2078 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2079                struct obd_statfs *sfs)
2080 {
2081         struct osd_device       *osd = osd_dt_dev(d);
2082         struct super_block      *sb = osd_sb(osd);
2083         struct kstatfs          *ksfs;
2084         __u64                    reserved;
2085         int                      result = 0;
2086
2087         if (unlikely(osd->od_mnt == NULL))
2088                 return -EINPROGRESS;
2089
2090         /* osd_lproc.c call this without env, allocate ksfs for that case */
2091         if (unlikely(env == NULL)) {
2092                 OBD_ALLOC_PTR(ksfs);
2093                 if (ksfs == NULL)
2094                         return -ENOMEM;
2095         } else {
2096                 ksfs = &osd_oti_get(env)->oti_ksfs;
2097         }
2098
2099         result = sb->s_op->statfs(sb->s_root, ksfs);
2100         if (result)
2101                 goto out;
2102
2103         statfs_pack(sfs, ksfs);
2104         if (unlikely(sb->s_flags & MS_RDONLY))
2105                 sfs->os_state |= OS_STATE_READONLY;
2106         if (ldiskfs_has_feature_extents(sb))
2107                 sfs->os_maxbytes = sb->s_maxbytes;
2108         else
2109                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2110
2111         /*
2112          * Reserve some space so to avoid fragmenting the filesystem too much.
2113          * Fragmentation not only impacts performance, but can also increase
2114          * metadata overhead significantly, causing grant calculation to be
2115          * wrong.
2116          *
2117          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2118          */
2119         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2120         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2121         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2122                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2123
2124         sfs->os_blocks -= reserved;
2125         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2126         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2127
2128 out:
2129         if (unlikely(env == NULL))
2130                 OBD_FREE_PTR(ksfs);
2131         return result;
2132 }
2133
2134 /**
2135  * Estimate space needed for file creations. We assume the largest filename
2136  * which is 2^64 - 1, hence a filename of 20 chars.
2137  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2138  */
2139 #ifdef __LDISKFS_DIR_REC_LEN
2140 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2141 #else
2142 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2143 #endif
2144
2145 /*
2146  * Concurrency: doesn't access mutable data.
2147  */
2148 static void osd_conf_get(const struct lu_env *env,
2149                          const struct dt_device *dev,
2150                          struct dt_device_param *param)
2151 {
2152         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2153         int                ea_overhead;
2154
2155         /*
2156          * XXX should be taken from not-yet-existing fs abstraction layer.
2157          */
2158         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2159         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2160         param->ddp_symlink_max  = sb->s_blocksize;
2161         param->ddp_mount_type     = LDD_MT_LDISKFS;
2162         if (ldiskfs_has_feature_extents(sb))
2163                 param->ddp_maxbytes = sb->s_maxbytes;
2164         else
2165                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2166         /* inode are statically allocated, so per-inode space consumption
2167          * is the space consumed by the directory entry */
2168         param->ddp_inodespace     = PER_OBJ_USAGE;
2169         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2170          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2171          * maximum length to not under count the actual number of extents
2172          * needed for writing a file. */
2173         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2174         /* worst-case extent insertion metadata overhead */
2175         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2176         param->ddp_mntopts      = 0;
2177         if (test_opt(sb, XATTR_USER))
2178                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2179         if (test_opt(sb, POSIX_ACL))
2180                 param->ddp_mntopts |= MNTOPT_ACL;
2181
2182         /* LOD might calculate the max stripe count based on max_ea_size,
2183          * so we need take account in the overhead as well,
2184          * xattr_header + magic + xattr_entry_head */
2185         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2186                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2187
2188 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2189         if (ldiskfs_has_feature_ea_inode(sb))
2190                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2191                                                                 ea_overhead;
2192         else
2193 #endif
2194                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2195
2196         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2197                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2198
2199         /* Preferred RPC size for efficient disk IO.  4MB shows good
2200          * all-around performance for ldiskfs, but use bigalloc chunk size
2201          * by default if larger. */
2202 #if defined(LDISKFS_CLUSTER_SIZE)
2203         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2204                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2205         else
2206 #endif
2207                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2208 }
2209
2210 /*
2211  * Concurrency: shouldn't matter.
2212  */
2213 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2214 {
2215         int rc;
2216
2217         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2218
2219         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2220
2221         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2222                rc);
2223
2224         return rc;
2225 }
2226
2227 /**
2228  * Start commit for OSD device.
2229  *
2230  * An implementation of dt_commit_async method for OSD device.
2231  * Asychronously starts underlayng fs sync and thereby a transaction
2232  * commit.
2233  *
2234  * \param env environment
2235  * \param d dt device
2236  *
2237  * \see dt_device_operations
2238  */
2239 static int osd_commit_async(const struct lu_env *env,
2240                             struct dt_device *d)
2241 {
2242         struct super_block *s = osd_sb(osd_dt_dev(d));
2243         ENTRY;
2244
2245         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2246         RETURN(s->s_op->sync_fs(s, 0));
2247 }
2248
2249 /* Our own copy of the set readonly functions if present, or NU if not. */
2250 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2251 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2252 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2253
2254 /*
2255  * Concurrency: shouldn't matter.
2256  */
2257 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2258 {
2259         struct super_block *sb = osd_sb(osd_dt_dev(d));
2260         struct block_device *dev = sb->s_bdev;
2261         int rc = -EOPNOTSUPP;
2262         ENTRY;
2263
2264         if (priv_dev_set_rdonly) {
2265                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2266
2267                 rc = 0;
2268                 CERROR("*** setting %s read-only ***\n",
2269                        osd_dt_dev(d)->od_svname);
2270
2271                 if (sb->s_op->freeze_fs) {
2272                         rc = sb->s_op->freeze_fs(sb);
2273                         if (rc)
2274                                 goto out;
2275                 }
2276
2277                 if (jdev && (jdev != dev)) {
2278                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2279                                (long)jdev);
2280                         priv_dev_set_rdonly(jdev);
2281                 }
2282                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2283                 priv_dev_set_rdonly(dev);
2284
2285                 if (sb->s_op->unfreeze_fs)
2286                         sb->s_op->unfreeze_fs(sb);
2287         }
2288
2289 out:
2290         if (rc)
2291                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2292                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2293
2294         RETURN(rc);
2295 }
2296
2297 /**
2298  * Note: we do not count into QUOTA here.
2299  * If we mount with --data_journal we may need more.
2300  */
2301 const int osd_dto_credits_noquota[DTO_NR] = {
2302         /**
2303          * Insert.
2304          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2305          * SINGLEDATA_TRANS_BLOCKS(8)
2306          * XXX Note: maybe iam need more, since iam have more level than
2307          *           EXT3 htree.
2308          */
2309         [DTO_INDEX_INSERT]  = 16,
2310         /**
2311          * Delete
2312          * just modify a single entry, probably merge few within a block
2313          */
2314         [DTO_INDEX_DELETE]  = 1,
2315         /**
2316          * Used for OI scrub
2317          */
2318         [DTO_INDEX_UPDATE]  = 16,
2319         /**
2320          * 4(inode, inode bits, groups, GDT)
2321          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2322          */
2323         [DTO_OBJECT_CREATE] = 4,
2324         /**
2325          * 4(inode, inode bits, groups, GDT)
2326          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2327          */
2328         [DTO_OBJECT_DELETE] = 4,
2329         /**
2330          * Attr set credits (inode)
2331          */
2332         [DTO_ATTR_SET_BASE] = 1,
2333         /**
2334          * Xattr set. The same as xattr of EXT3.
2335          * DATA_TRANS_BLOCKS(14)
2336          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2337          * are also counted in. Do not know why?
2338          */
2339         [DTO_XATTR_SET]     = 14,
2340         /**
2341          * credits for inode change during write.
2342          */
2343         [DTO_WRITE_BASE]    = 3,
2344         /**
2345          * credits for single block write.
2346          */
2347         [DTO_WRITE_BLOCK]   = 14,
2348         /**
2349          * Attr set credits for chown.
2350          * This is extra credits for setattr, and it is null without quota
2351          */
2352         [DTO_ATTR_SET_CHOWN] = 0
2353 };
2354
2355 static const struct dt_device_operations osd_dt_ops = {
2356         .dt_root_get       = osd_root_get,
2357         .dt_statfs         = osd_statfs,
2358         .dt_trans_create   = osd_trans_create,
2359         .dt_trans_start    = osd_trans_start,
2360         .dt_trans_stop     = osd_trans_stop,
2361         .dt_trans_cb_add   = osd_trans_cb_add,
2362         .dt_conf_get       = osd_conf_get,
2363         .dt_sync           = osd_sync,
2364         .dt_ro             = osd_ro,
2365         .dt_commit_async   = osd_commit_async,
2366 };
2367
2368 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2369                           unsigned role)
2370 {
2371         struct osd_object *obj = osd_dt_obj(dt);
2372         struct osd_thread_info *oti = osd_oti_get(env);
2373
2374         LINVRNT(osd_invariant(obj));
2375
2376         LASSERT(obj->oo_owner != env);
2377         down_read_nested(&obj->oo_sem, role);
2378
2379         LASSERT(obj->oo_owner == NULL);
2380         oti->oti_r_locks++;
2381 }
2382
2383 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2384                            unsigned role)
2385 {
2386         struct osd_object *obj = osd_dt_obj(dt);
2387         struct osd_thread_info *oti = osd_oti_get(env);
2388
2389         LINVRNT(osd_invariant(obj));
2390
2391         LASSERT(obj->oo_owner != env);
2392         down_write_nested(&obj->oo_sem, role);
2393
2394         LASSERT(obj->oo_owner == NULL);
2395         obj->oo_owner = env;
2396         oti->oti_w_locks++;
2397 }
2398
2399 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2400 {
2401         struct osd_object *obj = osd_dt_obj(dt);
2402         struct osd_thread_info *oti = osd_oti_get(env);
2403
2404         LINVRNT(osd_invariant(obj));
2405
2406         LASSERT(oti->oti_r_locks > 0);
2407         oti->oti_r_locks--;
2408         up_read(&obj->oo_sem);
2409 }
2410
2411 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2412 {
2413         struct osd_object *obj = osd_dt_obj(dt);
2414         struct osd_thread_info *oti = osd_oti_get(env);
2415
2416         LINVRNT(osd_invariant(obj));
2417
2418         LASSERT(obj->oo_owner == env);
2419         LASSERT(oti->oti_w_locks > 0);
2420         oti->oti_w_locks--;
2421         obj->oo_owner = NULL;
2422         up_write(&obj->oo_sem);
2423 }
2424
2425 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2426 {
2427         struct osd_object *obj = osd_dt_obj(dt);
2428
2429         LINVRNT(osd_invariant(obj));
2430
2431         return obj->oo_owner == env;
2432 }
2433
2434 static struct timespec *osd_inode_time(const struct lu_env *env,
2435                                        struct inode *inode, __u64 seconds)
2436 {
2437         struct osd_thread_info  *oti = osd_oti_get(env);
2438         struct timespec         *t   = &oti->oti_time;
2439
2440         t->tv_sec = seconds;
2441         t->tv_nsec = 0;
2442         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2443         return t;
2444 }
2445
2446 static void osd_inode_getattr(const struct lu_env *env,
2447                               struct inode *inode, struct lu_attr *attr)
2448 {
2449         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2450                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2451                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2452                            LA_BLKSIZE | LA_TYPE;
2453
2454         attr->la_atime   = LTIME_S(inode->i_atime);
2455         attr->la_mtime   = LTIME_S(inode->i_mtime);
2456         attr->la_ctime   = LTIME_S(inode->i_ctime);
2457         attr->la_mode    = inode->i_mode;
2458         attr->la_size    = i_size_read(inode);
2459         attr->la_blocks  = inode->i_blocks;
2460         attr->la_uid     = i_uid_read(inode);
2461         attr->la_gid     = i_gid_read(inode);
2462         attr->la_projid  = i_projid_read(inode);
2463         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2464         attr->la_nlink   = inode->i_nlink;
2465         attr->la_rdev    = inode->i_rdev;
2466         attr->la_blksize = 1 << inode->i_blkbits;
2467         attr->la_blkbits = inode->i_blkbits;
2468         /*
2469          * Ext4 did not transfer inherit flags from raw inode
2470          * to inode flags, and ext4 internally test raw inode
2471          * @i_flags directly. Instead of patching ext4, we do it here.
2472          */
2473         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2474                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2475 }
2476
2477 static int osd_attr_get(const struct lu_env *env,
2478                         struct dt_object *dt,
2479                         struct lu_attr *attr)
2480 {
2481         struct osd_object *obj = osd_dt_obj(dt);
2482
2483         if (unlikely(!dt_object_exists(dt)))
2484                 return -ENOENT;
2485         if (unlikely(obj->oo_destroyed))
2486                 return -ENOENT;
2487
2488         LASSERT(!dt_object_remote(dt));
2489         LINVRNT(osd_invariant(obj));
2490
2491         spin_lock(&obj->oo_guard);
2492         osd_inode_getattr(env, obj->oo_inode, attr);
2493         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2494                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2495         spin_unlock(&obj->oo_guard);
2496
2497         return 0;
2498 }
2499
2500 static int osd_declare_attr_qid(const struct lu_env *env,
2501                                 struct osd_object *obj,
2502                                 struct osd_thandle *oh, long long bspace,
2503                                 qid_t old_id, qid_t new_id, bool enforce,
2504                                 unsigned type, bool ignore_edquot)
2505 {
2506         int rc;
2507         struct osd_thread_info *info = osd_oti_get(env);
2508         struct lquota_id_info  *qi = &info->oti_qi;
2509
2510         qi->lqi_type = type;
2511         /* inode accounting */
2512         qi->lqi_is_blk = false;
2513
2514         /* one more inode for the new id ... */
2515         qi->lqi_id.qid_uid = new_id;
2516         qi->lqi_space      = 1;
2517         /* Reserve credits for the new id */
2518         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2519         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2520                 rc = 0;
2521         if (rc)
2522                 RETURN(rc);
2523
2524         /* and one less inode for the current id */
2525         qi->lqi_id.qid_uid = old_id;
2526         qi->lqi_space      = -1;
2527         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2528         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2529                 rc = 0;
2530         if (rc)
2531                 RETURN(rc);
2532
2533         /* block accounting */
2534         qi->lqi_is_blk = true;
2535
2536         /* more blocks for the new id ... */
2537         qi->lqi_id.qid_uid = new_id;
2538         qi->lqi_space      = bspace;
2539         /*
2540          * Credits for the new uid has been reserved, re-use "obj"
2541          * to save credit reservation.
2542          */
2543         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2544         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2545                 rc = 0;
2546         if (rc)
2547                 RETURN(rc);
2548
2549         /* and finally less blocks for the current uid */
2550         qi->lqi_id.qid_uid = old_id;
2551         qi->lqi_space      = -bspace;
2552         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2553         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2554                 rc = 0;
2555
2556         RETURN(rc);
2557 }
2558
2559 static int osd_declare_attr_set(const struct lu_env *env,
2560                                 struct dt_object *dt,
2561                                 const struct lu_attr *attr,
2562                                 struct thandle *handle)
2563 {
2564         struct osd_thandle     *oh;
2565         struct osd_object      *obj;
2566         qid_t                   uid;
2567         qid_t                   gid;
2568         long long               bspace;
2569         int                     rc = 0;
2570         bool                    enforce;
2571         ENTRY;
2572
2573         LASSERT(dt != NULL);
2574         LASSERT(handle != NULL);
2575
2576         obj = osd_dt_obj(dt);
2577         LASSERT(osd_invariant(obj));
2578
2579         oh = container_of0(handle, struct osd_thandle, ot_super);
2580         LASSERT(oh->ot_handle == NULL);
2581
2582         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2583                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2584
2585         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2586                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2587
2588         if (attr == NULL || obj->oo_inode == NULL)
2589                 RETURN(rc);
2590
2591         bspace   = obj->oo_inode->i_blocks << 9;
2592         bspace   = toqb(bspace);
2593
2594         /* Changing ownership is always preformed by super user, it should not
2595          * fail with EDQUOT unless required explicitly.
2596          *
2597          * We still need to call the osd_declare_qid() to calculate the journal
2598          * credits for updating quota accounting files and to trigger quota
2599          * space adjustment once the operation is completed.*/
2600         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2601                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2602
2603                 if (!ignore_edquot)
2604                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2605                                "(the quota space is %lld)\n",
2606                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2607                                attr->la_gid, bspace);
2608
2609                 /* USERQUOTA */
2610                 uid = i_uid_read(obj->oo_inode);
2611                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2612                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2613                                           attr->la_uid, enforce, USRQUOTA,
2614                                           true);
2615                 if (rc)
2616                         RETURN(rc);
2617
2618                 gid = i_gid_read(obj->oo_inode);
2619                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2620                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2621                                           i_gid_read(obj->oo_inode),
2622                                           attr->la_gid, enforce, GRPQUOTA,
2623                                           ignore_edquot);
2624                 if (rc)
2625                         RETURN(rc);
2626
2627         }
2628 #ifdef HAVE_PROJECT_QUOTA
2629         if (attr->la_valid & LA_PROJID) {
2630                 __u32 projid = i_projid_read(obj->oo_inode);
2631                 enforce = (attr->la_valid & LA_PROJID) &&
2632                                         (attr->la_projid != projid);
2633                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2634                                           (qid_t)projid, (qid_t)attr->la_projid,
2635                                           enforce, PRJQUOTA, true);
2636                 if (rc)
2637                         RETURN(rc);
2638         }
2639 #endif
2640         RETURN(rc);
2641 }
2642
2643 static int osd_inode_setattr(const struct lu_env *env,
2644                              struct inode *inode, const struct lu_attr *attr)
2645 {
2646         __u64 bits = attr->la_valid;
2647
2648         /* Only allow set size for regular file */
2649         if (!S_ISREG(inode->i_mode))
2650                 bits &= ~(LA_SIZE | LA_BLOCKS);
2651
2652         if (bits == 0)
2653                 return 0;
2654
2655         if (bits & LA_ATIME)
2656                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2657         if (bits & LA_CTIME)
2658                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2659         if (bits & LA_MTIME)
2660                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2661         if (bits & LA_SIZE) {
2662                 spin_lock(&inode->i_lock);
2663                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2664                 i_size_write(inode, attr->la_size);
2665                 spin_unlock(&inode->i_lock);
2666         }
2667
2668         /* OSD should not change "i_blocks" which is used by quota.
2669          * "i_blocks" should be changed by ldiskfs only. */
2670         if (bits & LA_MODE)
2671                 inode->i_mode = (inode->i_mode & S_IFMT) |
2672                                 (attr->la_mode & ~S_IFMT);
2673         if (bits & LA_UID)
2674                 i_uid_write(inode, attr->la_uid);
2675         if (bits & LA_GID)
2676                 i_gid_write(inode, attr->la_gid);
2677         if (bits & LA_PROJID)
2678                 i_projid_write(inode, attr->la_projid);
2679         if (bits & LA_NLINK)
2680                 set_nlink(inode, attr->la_nlink);
2681         if (bits & LA_RDEV)
2682                 inode->i_rdev = attr->la_rdev;
2683
2684         if (bits & LA_FLAGS) {
2685                 /* always keep S_NOCMTIME */
2686                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2687                                  S_NOCMTIME;
2688                 /*
2689                  * Ext4 did not transfer inherit flags from
2690                  * @inode->i_flags to raw inode i_flags when writing
2691                  * flags, we do it explictly here.
2692                  */
2693                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2694                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2695                 else
2696                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2697         }
2698         return 0;
2699 }
2700
2701 #ifdef HAVE_PROJECT_QUOTA
2702 static int osd_transfer_project(struct inode *inode, __u32 projid)
2703 {
2704         struct super_block *sb = inode->i_sb;
2705         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2706         int err;
2707         kprojid_t kprojid;
2708         struct ldiskfs_iloc iloc;
2709         struct ldiskfs_inode *raw_inode;
2710         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2711
2712         if (!ldiskfs_has_feature_project(sb)) {
2713                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2714                         == LDISKFS_DEF_PROJID);
2715                 if (projid != LDISKFS_DEF_PROJID)
2716                         return -EOPNOTSUPP;
2717                 else
2718                         return 0;
2719         }
2720
2721         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2722                 return -EOPNOTSUPP;
2723
2724         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2725         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2726                 return 0;
2727
2728         err = ldiskfs_get_inode_loc(inode, &iloc);
2729         if (err)
2730                 return err;
2731
2732         raw_inode = ldiskfs_raw_inode(&iloc);
2733         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2734                 err = -EOVERFLOW;
2735                 brelse(iloc.bh);
2736                 return err;
2737         }
2738         brelse(iloc.bh);
2739
2740         dquot_initialize(inode);
2741         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2742         if (transfer_to[PRJQUOTA]) {
2743                 err = __dquot_transfer(inode, transfer_to);
2744                 dqput(transfer_to[PRJQUOTA]);
2745                 if (err)
2746                         return err;
2747         }
2748
2749         return err;
2750 }
2751 #endif
2752
2753 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2754 {
2755         int rc;
2756
2757         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2758             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2759                 struct iattr    iattr;
2760
2761                 ll_vfs_dq_init(inode);
2762                 iattr.ia_valid = 0;
2763                 if (attr->la_valid & LA_UID)
2764                         iattr.ia_valid |= ATTR_UID;
2765                 if (attr->la_valid & LA_GID)
2766                         iattr.ia_valid |= ATTR_GID;
2767                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2768                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2769
2770                 rc = ll_vfs_dq_transfer(inode, &iattr);
2771                 if (rc) {
2772                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2773                                "enforcement enabled on the ldiskfs "
2774                                "filesystem?\n", inode->i_sb->s_id, rc);
2775                         return rc;
2776                 }
2777         }
2778
2779         /* Handle project id transfer here properly */
2780         if (attr->la_valid & LA_PROJID &&
2781             attr->la_projid != i_projid_read(inode)) {
2782 #ifdef HAVE_PROJECT_QUOTA
2783                 rc = osd_transfer_project(inode, attr->la_projid);
2784 #else
2785                 rc = -ENOTSUPP;
2786 #endif
2787                 if (rc) {
2788                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2789                                "enforcement enabled on the ldiskfs "
2790                                "filesystem?\n", inode->i_sb->s_id, rc);
2791                         return rc;
2792                 }
2793         }
2794         return 0;
2795 }
2796
2797 static int osd_attr_set(const struct lu_env *env,
2798                         struct dt_object *dt,
2799                         const struct lu_attr *attr,
2800                         struct thandle *handle)
2801 {
2802         struct osd_object *obj = osd_dt_obj(dt);
2803         struct inode      *inode;
2804         int rc;
2805
2806         if (!dt_object_exists(dt))
2807                 return -ENOENT;
2808
2809         LASSERT(handle != NULL);
2810         LASSERT(!dt_object_remote(dt));
2811         LASSERT(osd_invariant(obj));
2812
2813         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2814
2815         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2816             !osd_obj2dev(obj)->od_is_ost) {
2817                 struct osd_thread_info  *oti  = osd_oti_get(env);
2818                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2819                 struct lu_fid           *fid1 = &oti->oti_fid;
2820                 struct osd_inode_id     *id   = &oti->oti_id;
2821                 struct iam_path_descr   *ipd;
2822                 struct iam_container    *bag;
2823                 struct osd_thandle      *oh;
2824                 int                      rc;
2825
2826                 fid_cpu_to_be(fid1, fid0);
2827                 memset(id, 1, sizeof(*id));
2828                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2829                                   fid0)->oi_dir.od_container;
2830                 ipd = osd_idx_ipd_get(env, bag);
2831                 if (unlikely(ipd == NULL))
2832                         RETURN(-ENOMEM);
2833
2834                 oh = container_of0(handle, struct osd_thandle, ot_super);
2835                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2836                                 (const struct iam_rec *)id, ipd);
2837                 osd_ipd_put(env, bag, ipd);
2838                 return(rc > 0 ? 0 : rc);
2839         }
2840
2841         inode = obj->oo_inode;
2842
2843         rc = osd_quota_transfer(inode, attr);
2844         if (rc)
2845                 return rc;
2846
2847         spin_lock(&obj->oo_guard);
2848         rc = osd_inode_setattr(env, inode, attr);
2849         spin_unlock(&obj->oo_guard);
2850         if (rc != 0)
2851                 GOTO(out, rc);
2852
2853         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2854
2855         if (!(attr->la_valid & LA_FLAGS))
2856                 GOTO(out, rc);
2857
2858         /* Let's check if there are extra flags need to be set into LMA */
2859         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2860                 struct osd_thread_info *info = osd_oti_get(env);
2861                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2862
2863                 LASSERT(!obj->oo_pfid_in_lma);
2864
2865                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2866                                  &info->oti_ost_attrs);
2867                 if (rc)
2868                         GOTO(out, rc);
2869
2870                 lma->lma_incompat |=
2871                         lustre_to_lma_flags(attr->la_flags);
2872                 lustre_lma_swab(lma);
2873                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2874                                      lma, sizeof(*lma), XATTR_REPLACE);
2875                 if (rc != 0) {
2876                         struct osd_device *osd = osd_obj2dev(obj);
2877
2878                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2879                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2880                               lma->lma_incompat, rc);
2881                 } else {
2882                         obj->oo_lma_flags =
2883                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2884                 }
2885                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2886         }
2887 out:
2888         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2889
2890         return rc;
2891 }
2892
2893 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2894                                            struct osd_object *obj,
2895                                            const char *name, const int namelen)
2896 {
2897         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2898 }
2899
2900 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2901                       umode_t mode, struct dt_allocation_hint *hint,
2902                       struct thandle *th)
2903 {
2904         int result;
2905         struct osd_device  *osd = osd_obj2dev(obj);
2906         struct osd_thandle *oth;
2907         struct dt_object   *parent = NULL;
2908         struct inode       *inode;
2909
2910         LINVRNT(osd_invariant(obj));
2911         LASSERT(obj->oo_inode == NULL);
2912         LASSERT(obj->oo_hl_head == NULL);
2913
2914         if (S_ISDIR(mode) && ldiskfs_pdo) {
2915                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2916                 if (obj->oo_hl_head == NULL)
2917                         return -ENOMEM;
2918         }
2919
2920         oth = container_of(th, struct osd_thandle, ot_super);
2921         LASSERT(oth->ot_handle->h_transaction != NULL);
2922
2923         if (hint != NULL && hint->dah_parent != NULL &&
2924             !dt_object_remote(hint->dah_parent))
2925                 parent = hint->dah_parent;
2926
2927         inode = ldiskfs_create_inode(oth->ot_handle,
2928                                      parent ? osd_dt_obj(parent)->oo_inode :
2929                                               osd_sb(osd)->s_root->d_inode,
2930                                      mode);
2931         if (!IS_ERR(inode)) {
2932                 /* Do not update file c/mtime in ldiskfs. */
2933                 inode->i_flags |= S_NOCMTIME;
2934
2935                 /* For new created object, it must be consistent,
2936                  * and it is unnecessary to scrub against it. */
2937                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2938
2939                 obj->oo_inode = inode;
2940                 result = 0;
2941         } else {
2942                 if (obj->oo_hl_head != NULL) {
2943                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2944                         obj->oo_hl_head = NULL;
2945                 }
2946                 result = PTR_ERR(inode);
2947         }
2948         LINVRNT(osd_invariant(obj));
2949         return result;
2950 }
2951
2952 enum {
2953         OSD_NAME_LEN = 255
2954 };
2955
2956 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2957                      struct lu_attr *attr,
2958                      struct dt_allocation_hint *hint,
2959                      struct dt_object_format *dof,
2960                      struct thandle *th)
2961 {
2962         int result;
2963         struct osd_thandle *oth;
2964         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2965
2966         LASSERT(S_ISDIR(attr->la_mode));
2967
2968         oth = container_of(th, struct osd_thandle, ot_super);
2969         LASSERT(oth->ot_handle->h_transaction != NULL);
2970         result = osd_mkfile(info, obj, mode, hint, th);
2971
2972         return result;
2973 }
2974
2975 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2976                         struct lu_attr *attr,
2977                         struct dt_allocation_hint *hint,
2978                         struct dt_object_format *dof,
2979                         struct thandle *th)
2980 {
2981         int result;
2982         struct osd_thandle *oth;
2983         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2984
2985         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2986
2987         LASSERT(S_ISREG(attr->la_mode));
2988
2989         oth = container_of(th, struct osd_thandle, ot_super);
2990         LASSERT(oth->ot_handle->h_transaction != NULL);
2991
2992         result = osd_mkfile(info, obj, mode, hint, th);
2993         if (result == 0) {
2994                 LASSERT(obj->oo_inode != NULL);
2995                 if (feat->dif_flags & DT_IND_VARKEY)
2996                         result = iam_lvar_create(obj->oo_inode,
2997                                                  feat->dif_keysize_max,
2998                                                  feat->dif_ptrsize,
2999                                                  feat->dif_recsize_max,
3000                                                  oth->ot_handle);
3001                 else
3002                         result = iam_lfix_create(obj->oo_inode,
3003                                                  feat->dif_keysize_max,
3004                                                  feat->dif_ptrsize,
3005                                                  feat->dif_recsize_max,
3006                                                  oth->ot_handle);
3007
3008         }
3009         return result;
3010 }
3011
3012 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3013                      struct lu_attr *attr,
3014                      struct dt_allocation_hint *hint,
3015                      struct dt_object_format *dof,
3016                      struct thandle *th)
3017 {
3018         LASSERT(S_ISREG(attr->la_mode));
3019         return osd_mkfile(info, obj, (attr->la_mode &
3020                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3021 }
3022
3023 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3024                      struct lu_attr *attr,
3025                      struct dt_allocation_hint *hint,
3026                      struct dt_object_format *dof,
3027                      struct thandle *th)
3028 {
3029         LASSERT(S_ISLNK(attr->la_mode));
3030         return osd_mkfile(info, obj, (attr->la_mode &
3031                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3032 }
3033
3034 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3035                      struct lu_attr *attr,
3036                      struct dt_allocation_hint *hint,
3037                      struct dt_object_format *dof,
3038                      struct thandle *th)
3039 {
3040         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3041         int result;
3042
3043         LINVRNT(osd_invariant(obj));
3044         LASSERT(obj->oo_inode == NULL);
3045         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3046                 S_ISFIFO(mode) || S_ISSOCK(mode));
3047
3048         result = osd_mkfile(info, obj, mode, hint, th);
3049         if (result == 0) {
3050                 LASSERT(obj->oo_inode != NULL);
3051                 /*
3052                  * This inode should be marked dirty for i_rdev.  Currently
3053                  * that is done in the osd_attr_init().
3054                  */
3055                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3056                                    attr->la_rdev);
3057         }
3058         LINVRNT(osd_invariant(obj));
3059         return result;
3060 }
3061
3062 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3063                               struct lu_attr *,
3064                               struct dt_allocation_hint *hint,
3065                               struct dt_object_format *dof,
3066                               struct thandle *);
3067
3068 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3069 {
3070         osd_obj_type_f result;
3071
3072         switch (type) {
3073         case DFT_DIR:
3074                 result = osd_mkdir;
3075                 break;
3076         case DFT_REGULAR:
3077                 result = osd_mkreg;
3078                 break;
3079         case DFT_SYM:
3080                 result = osd_mksym;
3081                 break;
3082         case DFT_NODE:
3083                 result = osd_mknod;
3084                 break;
3085         case DFT_INDEX:
3086                 result = osd_mk_index;
3087                 break;
3088
3089         default:
3090                 LBUG();
3091                 break;
3092         }
3093         return result;
3094 }
3095
3096
3097 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3098                         struct dt_object *parent, struct dt_object *child,
3099                         umode_t child_mode)
3100 {
3101         LASSERT(ah);
3102
3103         ah->dah_parent = parent;
3104         ah->dah_mode = child_mode;
3105
3106         if (parent != NULL && !dt_object_remote(parent)) {
3107                 /* will help to find FID->ino at dt_insert("..") */
3108                 struct osd_object *pobj = osd_dt_obj(parent);
3109                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3110         }
3111 }
3112
3113 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3114                           struct lu_attr *attr, struct dt_object_format *dof)
3115 {
3116         struct inode   *inode = obj->oo_inode;
3117         __u64           valid = attr->la_valid;
3118         int             result;
3119
3120         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3121
3122         if (dof->dof_type != DFT_NODE)
3123                 attr->la_valid &= ~LA_RDEV;
3124         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3125                 attr->la_valid &= ~LA_ATIME;
3126         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3127                 attr->la_valid &= ~LA_CTIME;
3128         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3129                 attr->la_valid &= ~LA_MTIME;
3130
3131         result = osd_quota_transfer(inode, attr);
3132         if (result)
3133                 return;
3134
3135         if (attr->la_valid != 0) {
3136                 result = osd_inode_setattr(info->oti_env, inode, attr);
3137                 /*
3138                  * The osd_inode_setattr() should always succeed here.  The
3139                  * only error that could be returned is EDQUOT when we are
3140                  * trying to change the UID or GID of the inode. However, this
3141                  * should not happen since quota enforcement is no longer
3142                  * enabled on ldiskfs (lquota takes care of it).
3143                  */
3144                 LASSERTF(result == 0, "%d\n", result);
3145                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3146         }
3147
3148         attr->la_valid = valid;
3149 }
3150
3151 /**
3152  * Helper function for osd_create()
3153  *
3154  * \retval 0, on success
3155  */
3156 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3157                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3158                         struct dt_object_format *dof, struct thandle *th)
3159 {
3160         int     result;
3161         __u32   umask;
3162
3163         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3164
3165         /* we drop umask so that permissions we pass are not affected */
3166         umask = current->fs->umask;
3167         current->fs->umask = 0;
3168
3169         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3170                                                   th);
3171         if (likely(obj->oo_inode != NULL)) {
3172                 LASSERT(obj->oo_inode->i_state & I_NEW);
3173
3174                 /* Unlock the inode before attr initialization to avoid
3175                  * unnecessary dqget operations. LU-6378 */
3176                 unlock_new_inode(obj->oo_inode);
3177         }
3178
3179         if (likely(result == 0)) {
3180                 osd_attr_init(info, obj, attr, dof);
3181                 osd_object_init0(obj);
3182         }
3183
3184         /* restore previous umask value */
3185         current->fs->umask = umask;
3186
3187         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3188
3189         return result;
3190 }
3191
3192 /**
3193  * Helper function for osd_create()
3194  *
3195  * \retval 0, on success
3196  */
3197 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3198                            const struct lu_fid *fid, struct thandle *th)
3199 {
3200         struct osd_thread_info *info = osd_oti_get(env);
3201         struct osd_inode_id    *id   = &info->oti_id;
3202         struct osd_device      *osd  = osd_obj2dev(obj);
3203         struct osd_thandle     *oh;
3204         int                     rc;
3205
3206         LASSERT(obj->oo_inode != NULL);
3207
3208         oh = container_of0(th, struct osd_thandle, ot_super);
3209         LASSERT(oh->ot_handle);
3210         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3211
3212         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3213         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3214                            OI_CHECK_FLD, NULL);
3215         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3216
3217         return rc;
3218 }
3219
3220 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3221                    u64 seq, struct lu_seq_range *range)
3222 {
3223         struct seq_server_site  *ss = osd_seq_site(osd);
3224
3225         if (fid_seq_is_idif(seq)) {
3226                 fld_range_set_ost(range);
3227                 range->lsr_index = idif_ost_idx(seq);
3228                 return 0;
3229         }
3230
3231         if (!fid_seq_in_fldb(seq)) {
3232                 fld_range_set_mdt(range);
3233                 if (ss != NULL)
3234                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3235                          * at all */
3236                         range->lsr_index = ss->ss_node_id;
3237                 return 0;
3238         }
3239
3240         LASSERT(ss != NULL);
3241         fld_range_set_any(range);
3242         /* OSD will only do local fld lookup */
3243         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3244 }
3245
3246 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3247                               struct lu_attr *attr,
3248                               struct dt_allocation_hint *hint,
3249                               struct dt_object_format *dof,
3250                               struct thandle *handle)
3251 {
3252         struct osd_thandle *oh;
3253         int rc;
3254         ENTRY;
3255
3256         LASSERT(handle != NULL);
3257
3258         oh = container_of0(handle, struct osd_thandle, ot_super);
3259         LASSERT(oh->ot_handle == NULL);
3260
3261         /* EA object consumes more credits than regular object: osd_mk_index
3262          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3263          * leaf_node, could involves the block, block bitmap, groups, GDT
3264          * change for each block, so add 4 * 2 credits in that case. */
3265         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3266                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3267                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3268         /* Reuse idle OI block may cause additional one OI block
3269          * to be changed. */
3270         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3271                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3272
3273         if (!attr)
3274                 RETURN(0);
3275
3276         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3277                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3278                                    NULL, OSD_QID_INODE);
3279         if (rc != 0)
3280                 RETURN(rc);
3281
3282         /* will help to find FID->ino mapping at dt_insert() */
3283         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3284                                    osd_dt_obj(dt));
3285
3286         RETURN(rc);
3287 }
3288
3289 /**
3290  * Called to destroy on-disk representation of the object
3291  *
3292  * Concurrency: must be locked
3293  */
3294 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3295                                struct thandle *th)
3296 {
3297         struct osd_object  *obj = osd_dt_obj(dt);
3298         struct inode       *inode = obj->oo_inode;
3299         struct osd_thandle *oh;
3300         int                 rc;
3301         ENTRY;
3302
3303         if (inode == NULL)
3304                 RETURN(-ENOENT);
3305
3306         oh = container_of0(th, struct osd_thandle, ot_super);
3307         LASSERT(oh->ot_handle == NULL);
3308
3309         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3310                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3311
3312         /* For removing agent entry */
3313         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3314                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3315
3316         /* Recycle idle OI leaf may cause additional three OI blocks
3317          * to be changed. */
3318         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3319                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3320                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3321         /* one less inode */
3322         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3323                                    i_projid_read(inode), -1, oh, obj, NULL,
3324                                    OSD_QID_INODE);
3325         if (rc)
3326                 RETURN(rc);
3327         /* data to be truncated */
3328         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3329                                    i_projid_read(inode), 0, oh, obj, NULL,
3330                                    OSD_QID_BLK);
3331         if (rc)
3332                 RETURN(rc);
3333
3334         /* will help to find FID->ino when this object is being
3335          * added to PENDING/ */
3336         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3337
3338         RETURN(rc);
3339 }
3340
3341 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3342                        struct thandle *th)
3343 {
3344         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3345         struct osd_object *obj = osd_dt_obj(dt);
3346         struct inode *inode = obj->oo_inode;
3347         struct osd_device *osd = osd_obj2dev(obj);
3348         struct osd_thandle *oh;
3349         int result;
3350         ENTRY;
3351
3352         oh = container_of0(th, struct osd_thandle, ot_super);
3353         LASSERT(oh->ot_handle);
3354         LASSERT(inode);
3355         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3356
3357         if (unlikely(fid_is_acct(fid)))
3358                 RETURN(-EPERM);
3359
3360         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3361                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3362                 if (result != 0)
3363                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3364                                osd_name(osd), PFID(fid), result);
3365         }
3366
3367         if (S_ISDIR(inode->i_mode)) {
3368                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3369                         inode->i_nlink == 2);
3370
3371                 spin_lock(&obj->oo_guard);
3372                 clear_nlink(inode);
3373                 spin_unlock(&obj->oo_guard);
3374                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3375         }
3376
3377         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3378
3379         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3380
3381         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3382                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3383                                        oh->ot_handle, OI_CHECK_FLD);
3384
3385         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3386         /* XXX: add to ext3 orphan list */
3387         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3388
3389         /* not needed in the cache anymore */
3390         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3391         obj->oo_destroyed = 1;
3392
3393         RETURN(0);
3394 }
3395
3396 /**
3397  * Put the fid into lustre_mdt_attrs, and then place the structure
3398  * inode's ea. This fid should not be altered during the life time
3399  * of the inode.
3400  *
3401  * \retval +ve, on success
3402  * \retval -ve, on error
3403  *
3404  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3405  */
3406 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3407                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3408 {
3409         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3410         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3411         int rc;
3412         ENTRY;
3413
3414         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3415                 RETURN(0);
3416
3417         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3418                 rc = -ENOMEM;
3419
3420         lustre_loa_init(loa, fid, compat, incompat);
3421         lustre_loa_swab(loa, false);
3422
3423         /* For the OST device with 256 bytes inode size by default,
3424          * the PFID EA will be stored together with LMA EA to avoid
3425          * performance trouble. Otherwise the PFID EA can be stored
3426          * independently. LU-8998 */
3427         if ((compat & LMAC_FID_ON_OST) &&
3428             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3429                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3430                                      sizeof(*loa), XATTR_CREATE);
3431         else
3432                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3433                                      sizeof(*lma), XATTR_CREATE);
3434         /* LMA may already exist, but we need to check that all the
3435          * desired compat/incompat flags have been added. */
3436         if (unlikely(rc == -EEXIST)) {
3437                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3438                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3439                 if (rc < 0)
3440                         RETURN(rc);
3441
3442                 if (rc < sizeof(*lma))
3443                         RETURN(-EINVAL);
3444
3445                 lustre_loa_swab(loa, true);
3446                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3447                     ((compat == 0 && incompat == 0) ||
3448                      (!(~lma->lma_compat & compat) &&
3449                       !(~lma->lma_incompat & incompat))))
3450                         RETURN(0);
3451
3452                 lma->lma_self_fid = *fid;
3453                 lma->lma_compat |= compat;
3454                 lma->lma_incompat |= incompat;
3455                 if (rc == sizeof(*lma)) {
3456                         lustre_lma_swab(lma);
3457                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3458                                              sizeof(*lma), XATTR_REPLACE);
3459                 } else {
3460                         lustre_loa_swab(loa, false);
3461                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3462                                              sizeof(*loa), XATTR_REPLACE);
3463                 }
3464         }
3465
3466         RETURN(rc);
3467 }
3468
3469 /**
3470  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3471  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3472  * To have compatilibility with 1.8 ldiskfs driver we need to have
3473  * magic number at start of fid data.
3474  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3475  * its inmemory API.
3476  */
3477 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3478                                          const struct lu_fid *fid)
3479 {
3480         if (!fid_is_namespace_visible(fid) ||
3481             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3482                 param->edp_magic = 0;
3483                 return;
3484         }
3485
3486         param->edp_magic = LDISKFS_LUFID_MAGIC;
3487         param->edp_len =  sizeof(struct lu_fid) + 1;
3488         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3489 }
3490
3491 /**
3492  * Try to read the fid from inode ea into dt_rec.
3493  *
3494  * \param fid object fid.
3495  *
3496  * \retval 0 on success
3497  */
3498 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3499                           __u32 ino, struct lu_fid *fid,
3500                           struct osd_inode_id *id)
3501 {
3502         struct osd_thread_info *info  = osd_oti_get(env);
3503         struct inode           *inode;
3504         ENTRY;
3505
3506         osd_id_gen(id, ino, OSD_OII_NOGEN);
3507         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3508         if (IS_ERR(inode))
3509                 RETURN(PTR_ERR(inode));
3510
3511         iput(inode);
3512         RETURN(0);
3513 }
3514
3515 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3516                                         struct inode *dir,
3517                                         struct inode *parent_dir,
3518                                         const struct lu_fid *dot_fid,
3519                                         const struct lu_fid *dot_dot_fid,
3520                                         struct osd_thandle *oth)
3521 {
3522         struct ldiskfs_dentry_param *dot_ldp;
3523         struct ldiskfs_dentry_param *dot_dot_ldp;
3524         __u32 saved_nlink = dir->i_nlink;
3525         int rc;
3526
3527         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3528         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3529
3530         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3531         dot_ldp->edp_magic = 0;
3532
3533         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3534                                     dir, dot_ldp, dot_dot_ldp);
3535         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3536          * the subseqent ref_add() will increase the dir->i_nlink
3537          * as 3. That is incorrect for new created directory.
3538          *
3539          * It looks like hack, because we want to make the OSD API
3540          * to be order-independent for new created directory object
3541          * between dt_insert(..) and ref_add() operations.
3542          *
3543          * Here, we only restore the in-RAM dir-inode's nlink attr,
3544          * becuase if the nlink attr is not 2, then there will be
3545          * ref_add() called following the dt_insert(..), such call
3546          * will make both the in-RAM and on-disk dir-inode's nlink
3547          * attr to be set as 2. LU-7447 */
3548         set_nlink(dir, saved_nlink);
3549         return rc;
3550 }
3551
3552 /**
3553  * Create an local agent inode for remote entry
3554  */
3555 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3556                                                   struct osd_device *osd,
3557                                                   struct osd_object *pobj,
3558                                                   const struct lu_fid *fid,
3559                                                   __u32 type,
3560                                                   struct thandle *th)
3561 {
3562         struct osd_thread_info  *info = osd_oti_get(env);
3563         struct inode            *local;
3564         struct osd_thandle      *oh;
3565         int                     rc;
3566         ENTRY;
3567
3568         LASSERT(th);
3569         oh = container_of(th, struct osd_thandle, ot_super);
3570         LASSERT(oh->ot_handle->h_transaction != NULL);
3571
3572         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3573         if (IS_ERR(local)) {
3574                 CERROR("%s: create local error %d\n", osd_name(osd),
3575                        (int)PTR_ERR(local));
3576                 RETURN(local);
3577         }
3578
3579         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3580          * correct gid on remote file, not agent here */
3581         local->i_gid = current_fsgid();
3582         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3583         unlock_new_inode(local);
3584
3585         /* Agent inode should not have project ID */
3586 #ifdef  HAVE_PROJECT_QUOTA
3587         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3588             i_projid_read(pobj->oo_inode) != 0) {
3589                 rc = osd_transfer_project(local, 0);
3590                 if (rc) {
3591                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3592                                "quota enforcement enabled on the ldiskfs "
3593                                "filesystem?\n", local->i_sb->s_id, rc);
3594                         RETURN(ERR_PTR(rc));
3595                 }
3596         }
3597 #endif
3598         /* Set special LMA flag for local agent inode */
3599         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3600         if (rc != 0) {
3601                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3602                        osd_name(osd), PFID(fid), rc);
3603                 RETURN(ERR_PTR(rc));
3604         }
3605
3606         if (!S_ISDIR(type))
3607                 RETURN(local);
3608
3609         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3610                                          lu_object_fid(&pobj->oo_dt.do_lu),
3611                                          fid, oh);
3612         if (rc != 0) {
3613                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3614                         osd_name(osd), PFID(fid), rc);
3615                 RETURN(ERR_PTR(rc));
3616         }
3617
3618         RETURN(local);
3619 }
3620
3621 /**
3622  * when direntry is deleted, we have to take care of possible agent inode
3623  * referenced by that. unfortunately we can't do this at that point:
3624  * iget() within a running transaction leads to deadlock and we better do
3625  * not call that every delete declaration to save performance. so we put
3626  * a potention agent inode on a list and process that once the transaction
3627  * is over. Notice it's not any worse in terms of real orphans as regular
3628  * object destroy doesn't put inodes on the on-disk orphan list. this should
3629  * be addressed separately
3630  */
3631 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3632                                             struct osd_thandle *oh,
3633                                             __u32 ino)
3634 {
3635         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3636         struct osd_obj_orphan *oor;
3637
3638         OBD_ALLOC_PTR(oor);
3639         if (oor == NULL)
3640                 return -ENOMEM;
3641
3642         oor->oor_ino = ino;
3643         oor->oor_env = (struct lu_env *)env;
3644         spin_lock(&osd->od_osfs_lock);
3645         list_add(&oor->oor_list, &osd->od_orphan_list);
3646         spin_unlock(&osd->od_osfs_lock);
3647
3648         oh->ot_remove_agents = 1;
3649
3650         return 0;
3651
3652 }
3653
3654 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3655                                                 struct osd_device *osd)
3656 {
3657         struct osd_thread_info *info = osd_oti_get(env);
3658         struct osd_obj_orphan *oor, *tmp;
3659         struct osd_inode_id id;
3660         struct list_head list;
3661         struct inode *inode;
3662         struct lu_fid fid;
3663         handle_t *jh;
3664         __u32 ino;
3665
3666         INIT_LIST_HEAD(&list);
3667
3668         spin_lock(&osd->od_osfs_lock);
3669         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3670                 if (oor->oor_env == env) {
3671                         list_del(&oor->oor_list);
3672                         list_add(&oor->oor_list, &list);
3673                 }
3674         }
3675         spin_unlock(&osd->od_osfs_lock);
3676
3677         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3678
3679                 ino = oor->oor_ino;
3680
3681                 list_del(&oor->oor_list);
3682                 OBD_FREE_PTR(oor);
3683
3684                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3685                 inode = osd_iget_fid(info, osd, &id, &fid);
3686                 if (IS_ERR(inode))
3687                         continue;
3688
3689                 if (!osd_remote_fid(env, osd, &fid)) {
3690                         iput(inode);
3691                         continue;
3692                 }
3693
3694                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3695                 clear_nlink(inode);
3696                 mark_inode_dirty(inode);
3697                 ldiskfs_journal_stop(jh);
3698                 iput(inode);
3699         }
3700
3701         return 0;
3702 }
3703
3704 /**
3705  * OSD layer object create function for OST objects (b=11826).
3706  *
3707  * The FID is inserted into inode xattr here.
3708  *
3709  * \retval   0, on success
3710  * \retval -ve, on error
3711  */
3712 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3713                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3714                       struct dt_object_format *dof, struct thandle *th)
3715 {
3716         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3717         struct osd_object       *obj    = osd_dt_obj(dt);
3718         struct osd_thread_info  *info   = osd_oti_get(env);
3719         int                      result, on_ost = 0;
3720
3721         ENTRY;
3722
3723         if (dt_object_exists(dt))
3724                 RETURN(-EEXIST);
3725
3726         LINVRNT(osd_invariant(obj));
3727         LASSERT(!dt_object_remote(dt));
3728         LASSERT(osd_is_write_locked(env, obj));
3729         LASSERT(th != NULL);
3730
3731         if (unlikely(fid_is_acct(fid)))
3732                 /* Quota files can't be created from the kernel any more,
3733                  * 'tune2fs -O quota' will take care of creating them */
3734                 RETURN(-EPERM);
3735
3736         result = __osd_create(info, obj, attr, hint, dof, th);
3737         if (result == 0) {
3738                 if (fid_is_idif(fid) &&
3739                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3740                         struct lu_fid *tfid = &info->oti_fid;
3741                         struct ost_id *oi   = &info->oti_ostid;
3742
3743                         fid_to_ostid(fid, oi);
3744                         ostid_to_fid(tfid, oi, 0);
3745                         on_ost = 1;
3746                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3747                                                 LMAC_FID_ON_OST, 0);
3748                 } else {
3749                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3750                                                fid, OI_CHECK_FLD);
3751                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3752                                                 on_ost ? LMAC_FID_ON_OST : 0,
3753                                                 0);
3754                 }
3755                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3756                         obj->oo_dt.do_body_ops = &osd_body_ops;
3757         }
3758
3759         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3760                 result = __osd_oi_insert(env, obj, fid, th);
3761
3762         /* a small optimization - dt_insert() isn't usually applied
3763          * to OST objects, so we don't need to cache OI mapping for
3764          * OST objects */
3765         if (result == 0 && on_ost == 0) {
3766                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3767                 result = osd_idc_find_and_init(env, osd, obj);
3768                 LASSERT(result == 0);
3769         }
3770
3771         LASSERT(ergo(result == 0,
3772                      dt_object_exists(dt) && !dt_object_remote(dt)));
3773         LINVRNT(osd_invariant(obj));
3774         RETURN(result);
3775 }
3776
3777 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3778                                struct thandle *handle)
3779 {
3780         struct osd_thandle *oh;
3781
3782         /* it's possible that object doesn't exist yet */
3783         LASSERT(handle != NULL);
3784
3785         oh = container_of0(handle, struct osd_thandle, ot_super);
3786         LASSERT(oh->ot_handle == NULL);
3787
3788         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3789                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3790
3791         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3792
3793         return 0;
3794 }
3795
3796 /*
3797  * Concurrency: @dt is write locked.
3798  */
3799 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3800                        struct thandle *th)
3801 {
3802         struct osd_object  *obj = osd_dt_obj(dt);
3803         struct inode       *inode = obj->oo_inode;
3804         struct osd_thandle *oh;
3805         int                 rc = 0;
3806
3807         if (!dt_object_exists(dt) || obj->oo_destroyed)
3808                 return -ENOENT;
3809
3810         LINVRNT(osd_invariant(obj));
3811         LASSERT(!dt_object_remote(dt));
3812         LASSERT(osd_is_write_locked(env, obj));
3813         LASSERT(th != NULL);
3814
3815         oh = container_of0(th, struct osd_thandle, ot_super);
3816         LASSERT(oh->ot_handle != NULL);
3817
3818         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3819
3820         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3821                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3822         /*
3823          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3824          * (65000) subdirectories by storing "1" in i_nlink if the link count
3825          * would otherwise overflow. Directory tranversal tools understand
3826          * that (st_nlink == 1) indicates that the filesystem dose not track
3827          * hard links count on the directory, and will not abort subdirectory
3828          * scanning early once (st_nlink - 2) subdirs have been found.
3829          *
3830          * This also has to properly handle the case of inodes with nlink == 0
3831          * in case they are being linked into the PENDING directory
3832          */
3833         spin_lock(&obj->oo_guard);
3834         if (unlikely(inode->i_nlink == 0))
3835                 /* inc_nlink from 0 may cause WARN_ON */
3836                 set_nlink(inode, 1);
3837         else {
3838                 ldiskfs_inc_count(oh->ot_handle, inode);
3839                 if (!S_ISDIR(inode->i_mode))
3840                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3841         }
3842         spin_unlock(&obj->oo_guard);
3843
3844         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3845         LINVRNT(osd_invariant(obj));
3846
3847         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3848
3849         return rc;
3850 }
3851
3852 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3853                                struct thandle *handle)
3854 {
3855         struct osd_thandle *oh;
3856
3857         if (!dt_object_exists(dt))
3858                 return -ENOENT;
3859
3860         LASSERT(!dt_object_remote(dt));
3861         LASSERT(handle != NULL);
3862
3863         oh = container_of0(handle, struct osd_thandle, ot_super);
3864         LASSERT(oh->ot_handle == NULL);
3865
3866         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3867                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3868
3869         return 0;
3870 }
3871
3872 /*
3873  * Concurrency: @dt is write locked.
3874  */
3875 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3876                        struct thandle *th)
3877 {
3878         struct osd_object       *obj = osd_dt_obj(dt);
3879         struct inode            *inode = obj->oo_inode;
3880         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3881         struct osd_thandle      *oh;
3882
3883         if (!dt_object_exists(dt))
3884                 return -ENOENT;
3885
3886         LINVRNT(osd_invariant(obj));
3887         LASSERT(!dt_object_remote(dt));
3888         LASSERT(osd_is_write_locked(env, obj));
3889         LASSERT(th != NULL);
3890
3891         oh = container_of0(th, struct osd_thandle, ot_super);
3892         LASSERT(oh->ot_handle != NULL);
3893
3894         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3895
3896         spin_lock(&obj->oo_guard);
3897         /* That can be result of upgrade from old Lustre version and
3898          * applied only to local files.  Just skip this ref_del call.
3899          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3900         if (inode->i_nlink == 0) {
3901                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3902                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3903                              ", maybe an upgraded file? (LU-3915)\n",
3904                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3905                 spin_unlock(&obj->oo_guard);
3906                 return 0;
3907         }
3908
3909         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3910                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3911
3912         ldiskfs_dec_count(oh->ot_handle, inode);
3913         spin_unlock(&obj->oo_guard);
3914
3915         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3916         LINVRNT(osd_invariant(obj));
3917
3918         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3919
3920         return 0;
3921 }
3922
3923 /*
3924  * Concurrency: @dt is read locked.
3925  */
3926 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3927                          struct lu_buf *buf, const char *name)
3928 {
3929         struct osd_object      *obj    = osd_dt_obj(dt);
3930         struct inode           *inode  = obj->oo_inode;
3931         struct osd_thread_info *info   = osd_oti_get(env);
3932         struct dentry          *dentry = &info->oti_obj_dentry;
3933         bool                    cache_xattr = false;
3934         int                     rc;
3935
3936         LASSERT(buf);
3937
3938         /* version get is not real XATTR but uses xattr API */
3939         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3940                 dt_obj_version_t *ver = buf->lb_buf;
3941
3942                 /* for version we are just using xattr API but change inode
3943                  * field instead */
3944                 if (buf->lb_len == 0)
3945                         return sizeof(dt_obj_version_t);
3946
3947                 if (buf->lb_len < sizeof(dt_obj_version_t))
3948                         return -ERANGE;
3949
3950                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3951                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3952
3953                 *ver = LDISKFS_I(inode)->i_fs_version;
3954
3955                 return sizeof(dt_obj_version_t);
3956         }
3957
3958         if (!dt_object_exists(dt))
3959                 return -ENOENT;
3960
3961         LASSERT(!dt_object_remote(dt));
3962         LASSERT(inode->i_op != NULL);
3963 #ifdef HAVE_IOP_XATTR
3964         LASSERT(inode->i_op->getxattr != NULL);
3965 #endif
3966
3967         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3968             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3969                 cache_xattr = true;
3970
3971         if (cache_xattr) {
3972                 rc = osd_oxc_get(obj, name, buf);
3973                 if (rc != -ENOENT)
3974                         return rc;
3975         }
3976
3977         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
3978                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3979                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3980                 struct filter_fid *ff;
3981                 struct ost_layout *ol;
3982
3983                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3984                 if (rc)
3985                         return rc;
3986
3987                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
3988
3989                 rc = sizeof(*ff);
3990                 if (buf->lb_len == 0 || !buf->lb_buf)
3991                         return rc;
3992
3993                 if (buf->lb_len < rc)
3994                         return -ERANGE;
3995
3996                 ff = buf->lb_buf;
3997                 ol = &ff->ff_layout;
3998                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3999                                                   PFID_STRIPE_IDX_BITS);
4000                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4001                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4002                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4003                 if (lma->lma_compat & LMAC_COMP_INFO) {
4004                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4005                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4006                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4007                 } else {
4008                         ol->ol_comp_start = 0;
4009                         ol->ol_comp_end = 0;
4010                         ol->ol_comp_id = 0;
4011                 }
4012         } else {
4013                 rc = __osd_xattr_get(inode, dentry, name,
4014                                      buf->lb_buf, buf->lb_len);
4015         }
4016
4017         if (cache_xattr) {
4018                 if (rc == -ENOENT || rc == -ENODATA)
4019                         osd_oxc_add(obj, name, NULL, 0);
4020                 else if (rc > 0 && buf->lb_buf != NULL)
4021                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4022         }
4023
4024         return rc;
4025 }
4026
4027 static int osd_declare_xattr_set(const struct lu_env *env,
4028                                  struct dt_object *dt,
4029                                  const struct lu_buf *buf, const char *name,
4030                                  int fl, struct thandle *handle)
4031 {
4032         struct osd_thandle *oh;
4033         int credits = 0;
4034         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4035
4036         LASSERT(handle != NULL);
4037
4038         oh = container_of0(handle, struct osd_thandle, ot_super);
4039         LASSERT(oh->ot_handle == NULL);
4040
4041         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4042                 /* For non-upgrading case, the LMA is set first and
4043                  * usually fit inode. But for upgrade case, the LMA
4044                  * may be in another separated EA block. */
4045                 if (dt_object_exists(dt)) {
4046                         if (fl == LU_XATTR_REPLACE)
4047                                 credits = 1;
4048                         else
4049                                 goto upgrade;
4050                 }
4051         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4052                 credits = 1;
4053         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4054                 /* We may need to delete the old PFID EA. */
4055                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4056                 if (fl == LU_XATTR_REPLACE)
4057                         credits += 1;
4058                 else
4059                         goto upgrade;
4060         } else {
4061                 /* If some name entry resides on remote MDT, then will create
4062                  * agent entry under remote parent. On the other hand, if the
4063                  * remote entry will be removed, then related agent entry may
4064                  * need to be removed from the remote parent. So there may be
4065                  * kinds of cases, let's declare enough credits. The credits
4066                  * for create agent entry is enough for remove case. */
4067                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4068                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4069                         if (dt_object_exists(dt))
4070                                 credits += 1; /* For updating LMA */
4071                 }
4072
4073 upgrade:
4074                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4075
4076                 if (buf != NULL) {
4077                         ssize_t buflen;
4078
4079                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4080                                 /* learn xattr size from osd_xattr_get if
4081                                    attribute has not been read yet */
4082                                 buflen = __osd_xattr_get(
4083                                     osd_dt_obj(dt)->oo_inode,
4084                                     &osd_oti_get(env)->oti_obj_dentry,
4085                                     name, NULL, 0);
4086                                 if (buflen < 0)
4087                                         buflen = 0;
4088                         } else {
4089                                 buflen = buf->lb_len;
4090                         }
4091
4092                         if (buflen > sb->s_blocksize) {
4093                                 credits += osd_calc_bkmap_credits(
4094                                     sb, NULL, 0, -1,
4095                                     (buflen + sb->s_blocksize - 1) >>
4096                                     sb->s_blocksize_bits);
4097                         }
4098                 }
4099                 /*
4100                  * xattr set may involve inode quota change, reserve credits for
4101                  * dquot_initialize()
4102                  */
4103                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4104         }
4105
4106         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4107
4108         return 0;
4109 }
4110
4111 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4112                               const struct lu_buf *buf, int fl,
4113                               struct thandle *handle)
4114 {
4115         struct osd_thread_info *info = osd_oti_get(env);
4116         struct dentry *dentry = &info->oti_obj_dentry;
4117         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4118         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4119         struct inode *inode = obj->oo_inode;
4120         struct filter_fid *ff = buf->lb_buf;
4121         struct ost_layout *ol = &ff->ff_layout;
4122         int flags = XATTR_REPLACE;
4123         int rc;
4124         ENTRY;
4125
4126         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4127                 RETURN(-EINVAL);
4128
4129         rc = osd_get_lma(info, inode, dentry, loa);
4130         if (rc == -ENODATA) {
4131                 /* Usually for upgarding from old device */
4132                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4133                                 LMAC_FID_ON_OST, 0);
4134                 flags = XATTR_CREATE;
4135         } else if (rc) {
4136                 RETURN(rc);
4137         }
4138
4139         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4140                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4141                         RETURN(-EEXIST);
4142
4143                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4144                         /* Separate PFID EA from LMA */
4145                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4146                         lustre_lma_swab(lma);
4147                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4148                                              sizeof(*lma), XATTR_REPLACE);
4149                         if (!rc) {
4150                                 obj->oo_pfid_in_lma = 0;
4151                                 rc = LU_XATTR_CREATE;
4152                         }
4153
4154                         RETURN(rc);
4155                 }
4156         } else {
4157                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4158                         RETURN(fl);
4159
4160                 /* Old client does not send stripe information,
4161                  * then store the PFID EA on disk separatedly. */
4162                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4163                              ol->ol_stripe_size == 0))
4164                         RETURN(fl);
4165
4166                 /* Remove old PFID EA entry firstly. */
4167                 ll_vfs_dq_init(inode);
4168                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4169                 if (rc == -ENODATA) {
4170                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4171                                 RETURN(rc);
4172                 } else if (rc) {
4173                         RETURN(rc);
4174                 }
4175         }
4176
4177         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4178         if (likely(ol->ol_stripe_size != 0)) {
4179                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4180                                              PFID_STRIPE_IDX_BITS;
4181                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4182                 lma->lma_compat |= LMAC_STRIPE_INFO;
4183                 if (ol->ol_comp_id != 0) {
4184                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4185                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4186                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4187                         lma->lma_compat |= LMAC_COMP_INFO;
4188                 }
4189         }
4190
4191         lustre_loa_swab(loa, false);
4192
4193         /* Store the PFID EA inside LMA. */
4194         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4195                              flags);
4196         if (!rc)
4197                 obj->oo_pfid_in_lma = 1;
4198
4199         RETURN(rc);
4200 }
4201
4202 /*
4203  * In DNE environment, the object (in spite of regular file or directory)
4204  * and its name entry may reside on different MDTs. Under such case, we will
4205  * create an agent entry on the MDT where the object resides. The agent entry
4206  * references the object locally, that makes the object to be visible to the
4207  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4208  * such as 'tar' can handle the object properly.
4209  *
4210  * We handle the agent entry during set linkEA that is the common interface
4211  * for both regular file and directroy, can handle kinds of cases, such as
4212  * create/link/unlink/rename, and so on.
4213  *
4214  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4215  *
4216  * XXX: There are two known issues:
4217  * 1. For one object, we will create at most one agent entry even if there
4218  *    may be more than one cross-MDTs hard links on the object. So the local
4219  *    e2fsck may claim that the object's nlink is larger than the name entries
4220  *    that reference such inode. And in further, the e2fsck will fix the nlink
4221  *    attribute to match the local references. Then it will cause the object's
4222  *    nlink attribute to be inconsistent with the global references. it is bad
4223  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4224  *    other hand, the global namespace LFSCK can repair the object's attribute
4225  *    according to the linkEA.
4226  * 2. There may be too many hard links on the object as to its linkEA overflow,
4227  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4228  *    case happened, then at this point, we do not know whether there are some
4229  *    cross-MDTs reference. But there are local references, it guarantees that
4230  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4231  */
4232 static int osd_xattr_handle_linkea(const struct lu_env *env,
4233                                    struct osd_device *osd,
4234                                    struct osd_object *obj,
4235                                    const struct lu_buf *buf,
4236                                    struct thandle *handle)
4237 {
4238         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4239         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4240         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4241         struct lu_name tmpname;
4242         struct osd_thandle *oh;
4243         int rc;
4244         bool remote = false;
4245         ENTRY;
4246
4247         oh = container_of0(handle, struct osd_thandle, ot_super);
4248         LASSERT(oh->ot_handle != NULL);
4249
4250         rc = linkea_init_with_rec(&ldata);
4251         if (!rc) {
4252                 linkea_first_entry(&ldata);
4253                 while (ldata.ld_lee != NULL && !remote) {
4254                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4255                                             &tmpname, tfid);
4256                         if (osd_remote_fid(env, osd, tfid) > 0)
4257                                 remote = true;
4258                         else
4259                                 linkea_next_entry(&ldata);
4260                 }
4261         } else if (rc == -ENODATA) {
4262                 rc = 0;
4263         } else {
4264                 RETURN(rc);
4265         }
4266
4267         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4268                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4269                 if (rc)
4270                         CERROR("%s: failed to remove agent entry for "DFID
4271                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4272         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4273                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4274                 if (rc)
4275                         CERROR("%s: failed to create agent entry for "DFID
4276                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4277         }
4278
4279         RETURN(rc);
4280 }
4281
4282 /*
4283  * Concurrency: @dt is write locked.
4284  */
4285 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4286                          const struct lu_buf *buf, const char *name, int fl,
4287                          struct thandle *handle)
4288 {
4289         struct osd_object *obj = osd_dt_obj(dt);
4290         struct osd_device *osd = osd_obj2dev(obj);
4291         struct inode *inode = obj->oo_inode;
4292         struct osd_thread_info *info = osd_oti_get(env);
4293         int fs_flags = 0;
4294         int len;
4295         int rc;
4296         ENTRY;
4297
4298         LASSERT(handle);
4299         LASSERT(buf);
4300
4301         /* version set is not real XATTR */
4302         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4303                 dt_obj_version_t *version = buf->lb_buf;
4304
4305                 /* for version we are just using xattr API but change inode
4306                  * field instead */
4307                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4308
4309                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4310                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4311
4312                 LDISKFS_I(inode)->i_fs_version = *version;
4313                 /* Version is set after all inode operations are finished,
4314                  * so we should mark it dirty here */
4315                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4316
4317                 RETURN(0);
4318         }
4319
4320         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4321                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4322
4323         len = buf->lb_len;
4324         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4325
4326         /* For the OST device with 256 bytes inode size by default,
4327          * the PFID EA will be stored together with LMA EA to avoid
4328          * performance trouble. Otherwise the PFID EA can be stored
4329          * independently. LU-8998 */
4330         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4331             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4332                 LASSERT(buf->lb_buf);
4333
4334                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4335                 if (fl <= 0)
4336                         RETURN(fl);
4337         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4338                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4339                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4340
4341                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4342                 if (rc)
4343                         RETURN(rc);
4344
4345                 lma->lma_incompat |= LMAI_STRIPED;
4346                 lustre_lma_swab(lma);
4347                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4348                                      sizeof(*lma), XATTR_REPLACE);
4349                 if (rc != 0)
4350                         RETURN(rc);
4351         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4352                 LASSERT(!osd->od_is_ost);
4353
4354                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4355                 if (rc)
4356                         RETURN(rc);
4357         }
4358
4359         if (fl & LU_XATTR_REPLACE)
4360                 fs_flags |= XATTR_REPLACE;
4361
4362         if (fl & LU_XATTR_CREATE)
4363                 fs_flags |= XATTR_CREATE;
4364
4365         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4366         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4367
4368         if (rc == 0 &&
4369             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4370              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4371                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4372
4373         return rc;
4374 }
4375
4376 /*
4377  * Concurrency: @dt is read locked.
4378  */
4379 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4380                           const struct lu_buf *buf)
4381 {
4382         struct osd_object      *obj    = osd_dt_obj(dt);
4383         struct inode           *inode  = obj->oo_inode;
4384         struct osd_thread_info *info   = osd_oti_get(env);
4385         struct dentry          *dentry = &info->oti_obj_dentry;
4386
4387         if (!dt_object_exists(dt))
4388                 return -ENOENT;
4389
4390         LASSERT(!dt_object_remote(dt));
4391         LASSERT(inode->i_op != NULL);
4392         LASSERT(inode->i_op->listxattr != NULL);
4393
4394         dentry->d_inode = inode;
4395         dentry->d_sb = inode->i_sb;
4396         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4397 }
4398
4399 static int osd_declare_xattr_del(const struct lu_env *env,
4400                                  struct dt_object *dt, const char *name,
4401                                  struct thandle *handle)
4402 {
4403         struct osd_thandle *oh;
4404         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4405
4406         LASSERT(!dt_object_remote(dt));
4407         LASSERT(handle != NULL);
4408
4409         oh = container_of0(handle, struct osd_thandle, ot_super);
4410         LASSERT(oh->ot_handle == NULL);
4411
4412         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4413                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4414         /*
4415          * xattr del may involve inode quota change, reserve credits for
4416          * dquot_initialize()
4417          */
4418         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4419
4420         return 0;
4421 }
4422
4423 /*
4424  * Concurrency: @dt is write locked.
4425  */
4426 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4427                          const char *name, struct thandle *handle)
4428 {
4429         struct osd_object      *obj    = osd_dt_obj(dt);
4430         struct inode           *inode  = obj->oo_inode;
4431         struct osd_thread_info *info   = osd_oti_get(env);
4432         struct dentry          *dentry = &info->oti_obj_dentry;
4433         int                     rc;
4434
4435         if (!dt_object_exists(dt))
4436                 return -ENOENT;
4437
4438         LASSERT(!dt_object_remote(dt));
4439         LASSERT(inode->i_op != NULL);
4440         LASSERT(handle != NULL);
4441 #ifdef HAVE_IOP_XATTR
4442         LASSERT(inode->i_op->removexattr != NULL);
4443 #endif
4444
4445         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4446
4447         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4448                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4449
4450                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4451                                  &info->oti_ost_attrs);
4452                 if (!rc) {
4453                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4454
4455                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4456                         lustre_lma_swab(lma);
4457                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4458                                              sizeof(*lma), XATTR_REPLACE);
4459                         if (!rc)
4460                                 obj->oo_pfid_in_lma = 0;
4461                 }
4462         } else {
4463                 ll_vfs_dq_init(inode);
4464                 dentry->d_inode = inode;
4465                 dentry->d_sb = inode->i_sb;
4466                 rc = osd_removexattr(dentry, inode, name);
4467         }
4468
4469         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4470
4471         if (rc == 0 &&
4472             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4473              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4474                 osd_oxc_del(obj, name);
4475
4476         return rc;
4477 }
4478
4479 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4480                            __u64 start, __u64 end)
4481 {
4482         struct osd_object       *obj    = osd_dt_obj(dt);
4483         struct inode            *inode  = obj->oo_inode;
4484         struct osd_thread_info  *info   = osd_oti_get(env);
4485         struct dentry           *dentry = &info->oti_obj_dentry;
4486         struct file             *file   = &info->oti_file;
4487         int                     rc;
4488
4489         ENTRY;
4490
4491         dentry->d_inode = inode;
4492         dentry->d_sb = inode->i_sb;
4493         file->f_path.dentry = dentry;
4494         file->f_mapping = inode->i_mapping;
4495         file->f_op = inode->i_fop;
4496         set_file_inode(file, inode);
4497
4498         rc = ll_vfs_fsync_range(file, start, end, 0);
4499
4500         RETURN(rc);
4501 }
4502
4503 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4504 {
4505         return 0;
4506 }
4507
4508 /*
4509  * Index operations.
4510  */
4511
4512 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4513                            const struct dt_index_features *feat)
4514 {
4515         struct iam_descr *descr;
4516
4517         if (osd_object_is_root(o))
4518                 return feat == &dt_directory_features;
4519
4520         LASSERT(o->oo_dir != NULL);
4521
4522         descr = o->oo_dir->od_container.ic_descr;
4523         if (feat == &dt_directory_features) {
4524                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4525                         return 1;
4526                 else
4527                         return 0;
4528         } else {
4529                 return
4530                         feat->dif_keysize_min <= descr->id_key_size &&
4531                         descr->id_key_size <= feat->dif_keysize_max &&
4532                         feat->dif_recsize_min <= descr->id_rec_size &&
4533                         descr->id_rec_size <= feat->dif_recsize_max &&
4534                         !(feat->dif_flags & (DT_IND_VARKEY |
4535                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4536                         ergo(feat->dif_flags & DT_IND_UPDATE,
4537                              1 /* XXX check that object (and file system) is
4538                                 * writable */);
4539         }
4540 }
4541
4542 static int osd_iam_container_init(const struct lu_env *env,
4543                                   struct osd_object *obj,
4544                                   struct osd_directory *dir)
4545 {
4546         struct iam_container *bag = &dir->od_container;
4547         int result;
4548
4549         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4550         if (result != 0)
4551                 return result;
4552
4553         result = iam_container_setup(bag);
4554         if (result == 0)
4555                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4556         else
4557                 iam_container_fini(bag);
4558
4559         return result;
4560 }
4561
4562
4563 /*
4564  * Concurrency: no external locking is necessary.
4565  */
4566 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4567                          const struct dt_index_features *feat)
4568 {
4569         int                      result;
4570         int                      skip_iam = 0;
4571         struct osd_object       *obj = osd_dt_obj(dt);
4572
4573         LINVRNT(osd_invariant(obj));
4574
4575         if (osd_object_is_root(obj)) {
4576                 dt->do_index_ops = &osd_index_ea_ops;
4577                 result = 0;
4578         } else if (feat == &dt_directory_features) {
4579                 dt->do_index_ops = &osd_index_ea_ops;
4580                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4581                         result = 0;
4582                 else
4583                         result = -ENOTDIR;
4584                 skip_iam = 1;
4585         } else if (unlikely(feat == &dt_otable_features)) {
4586                 dt->do_index_ops = &osd_otable_ops;
4587                 return 0;
4588         } else if (unlikely(feat == &dt_acct_features)) {
4589                 dt->do_index_ops = &osd_acct_index_ops;
4590                 result = 0;
4591                 skip_iam = 1;
4592         } else if (!osd_has_index(obj)) {
4593                 struct osd_directory *dir;
4594                 struct osd_device *osd = osd_obj2dev(obj);
4595                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4596
4597                 OBD_ALLOC_PTR(dir);
4598                 if (dir) {
4599
4600                         spin_lock(&obj->oo_guard);
4601                         if (obj->oo_dir == NULL)
4602                                 obj->oo_dir = dir;
4603                         else
4604                                 /*
4605                                  * Concurrent thread allocated container data.
4606                                  */
4607                                 OBD_FREE_PTR(dir);
4608                         spin_unlock(&obj->oo_guard);
4609                         /*
4610                          * Now, that we have container data, serialize its
4611                          * initialization.
4612                          */
4613                         down_write(&obj->oo_ext_idx_sem);
4614                         /*
4615                          * recheck under lock.
4616                          */
4617
4618                         if (osd_has_index(obj)) {
4619                                 result = 0;
4620                                 goto unlock;
4621                         }
4622
4623                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4624                         if (result || feat == &dt_lfsck_namespace_features ||
4625                             feat == &dt_lfsck_layout_orphan_features ||
4626                             feat == &dt_lfsck_layout_dangling_features)
4627                                 goto unlock;
4628
4629                         result = osd_index_register(osd, fid,
4630                                                     feat->dif_keysize_max,
4631                                                     feat->dif_recsize_max);
4632                         if (result < 0)
4633                                 CWARN("%s: failed to register index "
4634                                       DFID": rc = %d\n",
4635                                       osd_name(osd), PFID(fid), result);
4636                         else if (result > 0)
4637                                 result = 0;
4638                         else
4639                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4640                                        " (%d/%d) registered\n",
4641                                        osd_name(osd), PFID(fid),
4642                                        (int)feat->dif_keysize_max,
4643                                        (int)feat->dif_recsize_max);
4644
4645 unlock:
4646                         up_write(&obj->oo_ext_idx_sem);
4647                 } else {
4648                         result = -ENOMEM;
4649                 }
4650         } else {
4651                 result = 0;
4652         }
4653
4654         if (result == 0 && skip_iam == 0) {
4655                 if (!osd_iam_index_probe(env, obj, feat))
4656                         result = -ENOTDIR;
4657         }
4658         LINVRNT(osd_invariant(obj));
4659
4660         return result;
4661 }
4662
4663 static int osd_otable_it_attr_get(const struct lu_env *env,
4664                                  struct dt_object *dt,
4665                                  struct lu_attr *attr)
4666 {
4667         attr->la_valid = 0;
4668         return 0;
4669 }
4670
4671 static const struct dt_object_operations osd_obj_ops = {
4672         .do_read_lock           = osd_read_lock,
4673         .do_write_lock          = osd_write_lock,
4674         .do_read_unlock         = osd_read_unlock,
4675         .do_write_unlock        = osd_write_unlock,
4676         .do_write_locked        = osd_write_locked,
4677         .do_attr_get            = osd_attr_get,
4678         .do_declare_attr_set    = osd_declare_attr_set,
4679         .do_attr_set            = osd_attr_set,
4680         .do_ah_init             = osd_ah_init,
4681         .do_declare_create      = osd_declare_create,
4682         .do_create              = osd_create,
4683         .do_declare_destroy     = osd_declare_destroy,
4684         .do_destroy             = osd_destroy,
4685         .do_index_try           = osd_index_try,
4686         .do_declare_ref_add     = osd_declare_ref_add,
4687         .do_ref_add             = osd_ref_add,
4688         .do_declare_ref_del     = osd_declare_ref_del,
4689         .do_ref_del             = osd_ref_del,
4690         .do_xattr_get           = osd_xattr_get,
4691         .do_declare_xattr_set   = osd_declare_xattr_set,
4692         .do_xattr_set           = osd_xattr_set,
4693         .do_declare_xattr_del   = osd_declare_xattr_del,
4694         .do_xattr_del           = osd_xattr_del,
4695         .do_xattr_list          = osd_xattr_list,
4696         .do_object_sync         = osd_object_sync,
4697         .do_invalidate          = osd_invalidate,
4698 };
4699
4700 static const struct dt_object_operations osd_obj_otable_it_ops = {
4701         .do_attr_get    = osd_otable_it_attr_get,
4702         .do_index_try   = osd_index_try,
4703 };
4704
4705 static int osd_index_declare_iam_delete(const struct lu_env *env,
4706                                         struct dt_object *dt,
4707                                         const struct dt_key *key,
4708                                         struct thandle *handle)
4709 {
4710         struct osd_thandle    *oh;
4711
4712         oh = container_of0(handle, struct osd_thandle, ot_super);
4713         LASSERT(oh->ot_handle == NULL);
4714
4715         /* Recycle  may cause additional three blocks to be changed. */
4716         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4717                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4718
4719         return 0;
4720 }
4721
4722 /**
4723  *      delete a (key, value) pair from index \a dt specified by \a key
4724  *
4725  *      \param  dt      osd index object
4726  *      \param  key     key for index
4727  *      \param  rec     record reference
4728  *      \param  handle  transaction handler
4729  *
4730  *      \retval  0  success
4731  *      \retval -ve   failure
4732  */
4733 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4734                                 const struct dt_key *key,
4735                                 struct thandle *handle)
4736 {
4737         struct osd_thread_info *oti = osd_oti_get(env);
4738         struct osd_object      *obj = osd_dt_obj(dt);
4739         struct osd_thandle     *oh;
4740         struct iam_path_descr  *ipd;
4741         struct iam_container   *bag = &obj->oo_dir->od_container;
4742         int                     rc;
4743         ENTRY;
4744
4745         if (!dt_object_exists(dt))
4746                 RETURN(-ENOENT);
4747
4748         LINVRNT(osd_invariant(obj));
4749         LASSERT(!dt_object_remote(dt));
4750         LASSERT(bag->ic_object == obj->oo_inode);
4751         LASSERT(handle != NULL);
4752
4753         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4754
4755         ipd = osd_idx_ipd_get(env, bag);
4756         if (unlikely(ipd == NULL))
4757                 RETURN(-ENOMEM);
4758
4759         oh = container_of0(handle, struct osd_thandle, ot_super);
4760         LASSERT(oh->ot_handle != NULL);
4761         LASSERT(oh->ot_handle->h_transaction != NULL);
4762
4763         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4764                 /* swab quota uid/gid provided by caller */
4765                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4766                 key = (const struct dt_key *)&oti->oti_quota_id;
4767         }
4768
4769         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4770         osd_ipd_put(env, bag, ipd);
4771         LINVRNT(osd_invariant(obj));
4772         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4773         RETURN(rc);
4774 }
4775
4776 static int osd_index_declare_ea_delete(const struct lu_env *env,
4777                                        struct dt_object *dt,
4778                                        const struct dt_key *key,
4779                                        struct thandle *handle)
4780 {
4781         struct osd_thandle *oh;
4782         struct inode       *inode;
4783         int                 rc, credits;
4784         ENTRY;
4785
4786         LASSERT(!dt_object_remote(dt));
4787         LASSERT(handle != NULL);
4788
4789         oh = container_of0(handle, struct osd_thandle, ot_super);
4790         LASSERT(oh->ot_handle == NULL);
4791
4792         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4793         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4794
4795         inode = osd_dt_obj(dt)->oo_inode;
4796         if (inode == NULL)
4797                 RETURN(-ENOENT);
4798
4799         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4800                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4801                                    NULL, OSD_QID_BLK);
4802         RETURN(rc);
4803 }
4804
4805 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4806                                           struct dt_rec *fid)
4807 {
4808         struct osd_fid_pack *rec;
4809         int                  rc = -ENODATA;
4810
4811         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4812                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4813                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4814                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4815                         rc = -EINVAL;
4816         }
4817         return rc;
4818 }
4819
4820 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4821                           const struct lu_fid *fid)
4822 {
4823         struct seq_server_site  *ss = osd_seq_site(osd);
4824         ENTRY;
4825
4826         /* FID seqs not in FLDB, must be local seq */
4827         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4828                 RETURN(0);
4829
4830         /* If FLD is not being initialized yet, it only happens during the
4831          * initialization, likely during mgs initialization, and we assume
4832          * this is local FID. */
4833         if (ss == NULL || ss->ss_server_fld == NULL)
4834                 RETURN(0);
4835
4836         /* Only check the local FLDB here */
4837         if (osd_seq_exists(env, osd, fid_seq(fid)))
4838                 RETURN(0);
4839
4840         RETURN(1);
4841 }
4842
4843 static void osd_take_care_of_agent(const struct lu_env *env,
4844                                    struct osd_device *osd,
4845                                    struct osd_thandle *oh,
4846                                    struct ldiskfs_dir_entry_2 *de)
4847 {
4848         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4849         struct osd_idmap_cache *idc;
4850         int rc, schedule = 0;
4851
4852         LASSERT(de != NULL);
4853
4854         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4855         if (likely(rc == 0)) {
4856                 idc = osd_idc_find_or_init(env, osd, fid);
4857                 if (IS_ERR(idc) || idc->oic_remote)
4858                         schedule = 1;
4859         } else if (rc == -ENODATA) {
4860                 /* can't get FID, postpone to the end of the
4861                  * transaction when iget() is safe */
4862                 schedule = 1;
4863         } else {
4864                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
4865         }
4866         if (schedule)
4867                 osd_schedule_agent_inode_removal(env, oh,
4868                                                  le32_to_cpu(de->inode));
4869 }
4870
4871 /**
4872  * Index delete function for interoperability mode (b11826).
4873  * It will remove the directory entry added by osd_index_ea_insert().
4874  * This entry is needed to maintain name->fid mapping.
4875  *
4876  * \param key,  key i.e. file entry to be deleted
4877  *
4878  * \retval   0, on success
4879  * \retval -ve, on error
4880  */
4881 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4882                                const struct dt_key *key, struct thandle *handle)
4883 {
4884         struct osd_object          *obj = osd_dt_obj(dt);
4885         struct inode               *dir = obj->oo_inode;
4886         struct dentry              *dentry;
4887         struct osd_thandle         *oh;
4888         struct ldiskfs_dir_entry_2 *de = NULL;
4889         struct buffer_head         *bh;
4890         struct htree_lock          *hlock = NULL;
4891         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4892         int                        rc;
4893         ENTRY;
4894
4895         if (!dt_object_exists(dt))
4896                 RETURN(-ENOENT);
4897
4898         LINVRNT(osd_invariant(obj));
4899         LASSERT(!dt_object_remote(dt));
4900         LASSERT(handle != NULL);
4901
4902         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4903
4904         oh = container_of(handle, struct osd_thandle, ot_super);
4905         LASSERT(oh->ot_handle != NULL);
4906         LASSERT(oh->ot_handle->h_transaction != NULL);
4907
4908         ll_vfs_dq_init(dir);
4909         dentry = osd_child_dentry_get(env, obj,
4910                                       (char *)key, strlen((char *)key));
4911
4912         if (obj->oo_hl_head != NULL) {
4913                 hlock = osd_oti_get(env)->oti_hlock;
4914                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4915                                    dir, LDISKFS_HLOCK_DEL);
4916         } else {
4917                 down_write(&obj->oo_ext_idx_sem);
4918         }
4919
4920         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4921         if (!IS_ERR(bh)) {
4922                 /* If this is not the ".." entry, it might be a remote DNE
4923                  * entry and  we need to check if the FID is for a remote
4924                  * MDT.  If the FID is  not in the directory entry (e.g.
4925                  * upgraded 1.8 filesystem without dirdata enabled) then
4926                  * we need to get the FID from the LMA. For a remote directory
4927                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4928                  * case.
4929                  *
4930                  * Delete the entry before the agent inode in order to
4931                  * simplify error handling.  At worst an error after deleting
4932                  * the entry first might leak the agent inode afterward. The
4933                  * reverse would need filesystem abort in case of error deleting
4934                  * the entry after the agent had been removed, or leave a
4935                  * dangling entry pointing at a random inode. */
4936                 if (strcmp((char *)key, dotdot) != 0)
4937                         osd_take_care_of_agent(env, osd, oh, de);
4938                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4939                 brelse(bh);
4940         } else {
4941                 rc = PTR_ERR(bh);
4942         }
4943         if (hlock != NULL)
4944                 ldiskfs_htree_unlock(hlock);
4945         else
4946                 up_write(&obj->oo_ext_idx_sem);
4947
4948         GOTO(out, rc);
4949
4950 out:
4951         LASSERT(osd_invariant(obj));
4952         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4953         RETURN(rc);
4954 }
4955
4956 /**
4957  *      Lookup index for \a key and copy record to \a rec.
4958  *
4959  *      \param  dt      osd index object
4960  *      \param  key     key for index
4961  *      \param  rec     record reference
4962  *
4963  *      \retval  +ve  success : exact mach
4964  *      \retval  0    return record with key not greater than \a key
4965  *      \retval -ve   failure
4966  */
4967 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4968                                 struct dt_rec *rec, const struct dt_key *key)
4969 {
4970         struct osd_object      *obj = osd_dt_obj(dt);
4971         struct iam_path_descr  *ipd;
4972         struct iam_container   *bag = &obj->oo_dir->od_container;
4973         struct osd_thread_info *oti = osd_oti_get(env);
4974         struct iam_iterator    *it = &oti->oti_idx_it;
4975         struct iam_rec         *iam_rec;
4976         int                     rc;
4977         ENTRY;
4978
4979         if (!dt_object_exists(dt))
4980                 RETURN(-ENOENT);
4981
4982         LASSERT(osd_invariant(obj));
4983         LASSERT(!dt_object_remote(dt));
4984         LASSERT(bag->ic_object == obj->oo_inode);
4985
4986         ipd = osd_idx_ipd_get(env, bag);
4987         if (IS_ERR(ipd))
4988                 RETURN(-ENOMEM);
4989
4990         /* got ipd now we can start iterator. */
4991         iam_it_init(it, bag, 0, ipd);
4992
4993         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4994                 /* swab quota uid/gid provided by caller */
4995                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4996                 key = (const struct dt_key *)&oti->oti_quota_id;
4997         }
4998
4999         rc = iam_it_get(it, (struct iam_key *)key);
5000         if (rc >= 0) {
5001                 if (S_ISDIR(obj->oo_inode->i_mode))
5002                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5003                 else
5004                         iam_rec = (struct iam_rec *) rec;
5005
5006                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5007
5008                 if (S_ISDIR(obj->oo_inode->i_mode))
5009                         osd_fid_unpack((struct lu_fid *) rec,
5010                                        (struct osd_fid_pack *)iam_rec);
5011                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5012                         osd_quota_unpack(obj, rec);
5013         }
5014
5015         iam_it_put(it);
5016         iam_it_fini(it);
5017         osd_ipd_put(env, bag, ipd);
5018
5019         LINVRNT(osd_invariant(obj));
5020
5021         RETURN(rc);
5022 }
5023
5024 static int osd_index_declare_iam_insert(const struct lu_env *env,
5025                                         struct dt_object *dt,
5026                                         const struct dt_rec *rec,
5027                                         const struct dt_key *key,
5028                                         struct thandle *handle)
5029 {
5030         struct osd_thandle *oh;
5031
5032         LASSERT(handle != NULL);
5033
5034         oh = container_of0(handle, struct osd_thandle, ot_super);
5035         LASSERT(oh->ot_handle == NULL);
5036
5037         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5038                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5039
5040         return 0;
5041 }
5042
5043 /**
5044  *      Inserts (key, value) pair in \a dt index object.
5045  *
5046  *      \param  dt      osd index object
5047  *      \param  key     key for index
5048  *      \param  rec     record reference
5049  *      \param  th      transaction handler
5050  *
5051  *      \retval  0  success
5052  *      \retval -ve failure
5053  */
5054 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5055                                 const struct dt_rec *rec,
5056                                 const struct dt_key *key, struct thandle *th,
5057                                 int ignore_quota)
5058 {
5059         struct osd_object     *obj = osd_dt_obj(dt);
5060         struct iam_path_descr *ipd;
5061         struct osd_thandle    *oh;
5062         struct iam_container  *bag;
5063         struct osd_thread_info *oti = osd_oti_get(env);
5064         struct iam_rec         *iam_rec;
5065         int                     rc;
5066         ENTRY;
5067
5068         if (!dt_object_exists(dt))
5069                 RETURN(-ENOENT);
5070
5071         LINVRNT(osd_invariant(obj));
5072         LASSERT(!dt_object_remote(dt));
5073
5074         bag = &obj->oo_dir->od_container;
5075         LASSERT(bag->ic_object == obj->oo_inode);
5076         LASSERT(th != NULL);
5077
5078         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5079
5080         ipd = osd_idx_ipd_get(env, bag);
5081         if (unlikely(ipd == NULL))
5082                 RETURN(-ENOMEM);
5083
5084         oh = container_of0(th, struct osd_thandle, ot_super);
5085         LASSERT(oh->ot_handle != NULL);
5086         LASSERT(oh->ot_handle->h_transaction != NULL);
5087         if (S_ISDIR(obj->oo_inode->i_mode)) {
5088                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5089                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
5090         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5091                 /* pack quota uid/gid */
5092                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5093                 key = (const struct dt_key *)&oti->oti_quota_id;
5094                 /* pack quota record */
5095                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5096                 iam_rec = (struct iam_rec *)rec;
5097         } else {
5098                 iam_rec = (struct iam_rec *)rec;
5099         }
5100
5101         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5102                         iam_rec, ipd);
5103         osd_ipd_put(env, bag, ipd);
5104         LINVRNT(osd_invariant(obj));
5105         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5106         RETURN(rc);
5107 }
5108
5109 /**
5110  * Calls ldiskfs_add_entry() to add directory entry
5111  * into the directory. This is required for
5112  * interoperability mode (b11826)
5113  *
5114  * \retval   0, on success
5115  * \retval -ve, on error
5116  */
5117 static int __osd_ea_add_rec(struct osd_thread_info *info,
5118                             struct osd_object *pobj, struct inode  *cinode,
5119                             const char *name, const struct lu_fid *fid,
5120                             struct htree_lock *hlock, struct thandle *th)
5121 {
5122         struct ldiskfs_dentry_param *ldp;
5123         struct dentry               *child;
5124         struct osd_thandle          *oth;
5125         int                          rc;
5126
5127         oth = container_of(th, struct osd_thandle, ot_super);
5128         LASSERT(oth->ot_handle != NULL);
5129         LASSERT(oth->ot_handle->h_transaction != NULL);
5130         LASSERT(pobj->oo_inode);
5131
5132         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5133         if (unlikely(pobj->oo_inode ==
5134                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5135                 ldp->edp_magic = 0;
5136         else
5137                 osd_get_ldiskfs_dirent_param(ldp, fid);
5138         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5139         child->d_fsdata = (void *)ldp;
5140         ll_vfs_dq_init(pobj->oo_inode);
5141         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5142                                    child, cinode, hlock);
5143         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5144                 struct ldiskfs_dir_entry_2      *de;
5145                 struct buffer_head              *bh;
5146                 int                              rc1;
5147
5148                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5149                                             NULL, hlock);
5150                 if (!IS_ERR(bh)) {
5151                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5152                                                                bh);
5153                         if (rc1 == 0) {
5154                                 if (S_ISDIR(cinode->i_mode))
5155                                         de->file_type = LDISKFS_DIRENT_LUFID |
5156                                                         LDISKFS_FT_REG_FILE;
5157                                 else
5158                                         de->file_type = LDISKFS_DIRENT_LUFID |
5159                                                         LDISKFS_FT_DIR;
5160                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5161                                                               NULL, bh);
5162                         }
5163                         brelse(bh);
5164                 }
5165         }
5166
5167         RETURN(rc);
5168 }
5169
5170 /**
5171  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5172  * into the directory.Also sets flags into osd object to
5173  * indicate dot and dotdot are created. This is required for
5174  * interoperability mode (b11826)
5175  *
5176  * \param dir   directory for dot and dotdot fixup.
5177  * \param obj   child object for linking
5178  *
5179  * \retval   0, on success
5180  * \retval -ve, on error
5181  */
5182 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5183                               struct osd_object *dir,
5184                               struct inode *parent_dir, const char *name,
5185                               const struct lu_fid *dot_fid,
5186                               const struct lu_fid *dot_dot_fid,
5187                               struct thandle *th)
5188 {
5189         struct inode                *inode = dir->oo_inode;
5190         struct osd_thandle          *oth;
5191         int result = 0;
5192
5193         oth = container_of(th, struct osd_thandle, ot_super);
5194         LASSERT(oth->ot_handle->h_transaction != NULL);
5195         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5196
5197         if (strcmp(name, dot) == 0) {
5198                 if (dir->oo_compat_dot_created) {
5199                         result = -EEXIST;
5200                 } else {
5201                         LASSERT(inode->i_ino == parent_dir->i_ino);
5202                         dir->oo_compat_dot_created = 1;
5203                         result = 0;
5204                 }
5205         } else if (strcmp(name, dotdot) == 0) {
5206                 if (!dir->oo_compat_dot_created)
5207                         return -EINVAL;
5208                 /* in case of rename, dotdot is already created */
5209                 if (dir->oo_compat_dotdot_created) {
5210                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5211                                                 dot_dot_fid, NULL, th);
5212                 }
5213
5214                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5215                         struct lu_fid tfid = *dot_dot_fid;
5216
5217                         tfid.f_oid--;
5218                         result = osd_add_dot_dotdot_internal(info,
5219                                         dir->oo_inode, parent_dir, dot_fid,
5220                                         &tfid, oth);
5221                 } else {
5222                         result = osd_add_dot_dotdot_internal(info,
5223                                         dir->oo_inode, parent_dir, dot_fid,
5224                                         dot_dot_fid, oth);
5225                 }
5226
5227                 if (result == 0)
5228                         dir->oo_compat_dotdot_created = 1;
5229         }
5230
5231         return result;
5232 }
5233
5234
5235 /**
5236  * It will call the appropriate osd_add* function and return the
5237  * value, return by respective functions.
5238  */
5239 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5240                           struct inode *cinode, const char *name,
5241                           const struct lu_fid *fid, struct thandle *th)
5242 {
5243         struct osd_thread_info *info   = osd_oti_get(env);
5244         struct htree_lock      *hlock;
5245         int                     rc;
5246
5247         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5248
5249         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
5250                                                    name[2] =='\0'))) {
5251                 if (hlock != NULL) {
5252                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5253                                            pobj->oo_inode, 0);
5254                 } else {
5255                         down_write(&pobj->oo_ext_idx_sem);
5256                 }
5257
5258                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5259                                         lu_object_fid(&pobj->oo_dt.do_lu),
5260                                         fid, th);
5261         } else {
5262                 if (hlock != NULL) {
5263                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5264                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5265                 } else {
5266                         down_write(&pobj->oo_ext_idx_sem);
5267                 }
5268
5269                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5270                         struct lu_fid *tfid = &info->oti_fid;
5271
5272                         *tfid = *fid;
5273                         tfid->f_ver = ~0;
5274                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5275                                               tfid, hlock, th);
5276                 } else {
5277                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5278                                               hlock, th);
5279                 }
5280         }
5281         if (hlock != NULL)
5282                 ldiskfs_htree_unlock(hlock);
5283         else
5284                 up_write(&pobj->oo_ext_idx_sem);
5285
5286         return rc;
5287 }
5288
5289 static int
5290 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5291                       struct osd_idmap_cache *oic)
5292 {
5293         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5294         struct lu_fid *fid = &oic->oic_fid;
5295         struct osd_inode_id *id = &oic->oic_lid;
5296         struct inode *inode = NULL;
5297         int once = 0;
5298         bool insert;
5299         int rc;
5300         ENTRY;
5301
5302         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5303                 RETURN(0);
5304
5305         if (thread_is_running(&scrub->os_thread) &&
5306             scrub->os_pos_current > id->oii_ino)
5307                 RETURN(0);
5308
5309         if (dev->od_auto_scrub_interval == AS_NEVER ||
5310             ktime_get_real_seconds() <
5311             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5312                 RETURN(0);
5313
5314 again:
5315         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5316         if (rc == -ENOENT) {
5317                 __u32 gen = id->oii_gen;
5318
5319                 insert = true;
5320                 if (inode != NULL)
5321                         goto trigger;
5322
5323                 inode = osd_iget(oti, dev, id);
5324                 /* The inode has been removed (by race maybe). */
5325                 if (IS_ERR(inode)) {
5326                         rc = PTR_ERR(inode);
5327
5328                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5329                 }
5330
5331                 /* The OI mapping is lost. */
5332                 if (gen != OSD_OII_NOGEN)
5333                         goto trigger;
5334
5335                 /* The inode may has been reused by others, we do not know,
5336                  * leave it to be handled by subsequent osd_fid_lookup(). */
5337                 GOTO(out, rc = 0);
5338         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5339                 GOTO(out, rc);
5340         }
5341
5342         insert = false;
5343
5344 trigger:
5345         if (thread_is_running(&scrub->os_thread)) {
5346                 if (inode == NULL) {
5347                         inode = osd_iget(oti, dev, id);
5348                         /* The inode has been removed (by race maybe). */
5349                         if (IS_ERR(inode)) {
5350                                 rc = PTR_ERR(inode);
5351
5352                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5353                         }
5354                 }
5355
5356                 rc = osd_oii_insert(dev, oic, insert);
5357                 /* There is race condition between osd_oi_lookup and OI scrub.
5358                  * The OI scrub finished just after osd_oi_lookup() failure.
5359                  * Under such case, it is unnecessary to trigger OI scrub again,
5360                  * but try to call osd_oi_lookup() again. */
5361                 if (unlikely(rc == -EAGAIN))
5362                         goto again;
5363
5364                 if (!S_ISDIR(inode->i_mode))
5365                         rc = 0;
5366                 else
5367                         rc = osd_check_lmv(oti, dev, inode, oic);
5368
5369                 GOTO(out, rc);
5370         }
5371
5372         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5373                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5374                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5375                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5376                        "%s: trigger partial OI scrub for RPC inconsistency "
5377                        "checking FID "DFID": rc = %d\n",
5378                        osd_dev2name(dev), PFID(fid), rc);
5379                 if (rc == 0 || rc == -EALREADY)
5380                         goto again;
5381         }
5382
5383         GOTO(out, rc);
5384
5385 out:
5386         if (inode)
5387                 iput(inode);
5388
5389         RETURN(rc);
5390 }
5391
5392 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5393                                struct osd_device *dev,
5394                                struct osd_idmap_cache *oic,
5395                                struct lu_fid *fid, __u32 ino)
5396 {
5397         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5398         struct inode *inode;
5399         int rc;
5400
5401         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5402         inode = osd_iget(oti, dev, &oic->oic_lid);
5403         if (IS_ERR(inode)) {
5404                 fid_zero(&oic->oic_fid);
5405                 return PTR_ERR(inode);
5406         }
5407
5408         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5409         iput(inode);
5410         if (rc != 0)
5411                 fid_zero(&oic->oic_fid);
5412         else
5413                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5414         return rc;
5415 }
5416
5417 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5418                       struct osd_inode_id *id, const struct lu_fid *fid)
5419 {
5420         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5421                id->oii_ino, id->oii_gen, info);
5422         info->oti_cache.oic_lid = *id;
5423         info->oti_cache.oic_fid = *fid;
5424         info->oti_cache.oic_dev = osd;
5425 }
5426
5427 /**
5428  * Get parent FID from the linkEA.
5429  *
5430  * For a directory which parent resides on remote MDT, to satisfy the
5431  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5432  * the other hand, to make the lookup(..) on the directory can return
5433  * the real parent FID, we append the real parent FID after its ".."
5434  * name entry in the /REMOTE_PARENT_DIR.
5435  *
5436  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5437  * backup. So after the restore, we cannot get the right parent FID from
5438  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5439  * we have stored the real parent FID in the directory object's linkEA,
5440  * we can parse the linkEA for the real parent FID.
5441  *
5442  * \param[in] env       pointer to the thread context
5443  * \param[in] obj       pointer to the object to be handled
5444  * \param[out]fid       pointer to the buffer to hold the parent FID
5445  *
5446  * \retval              0 for getting the real parent FID successfully
5447  * \retval              negative error number on failure
5448  */
5449 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5450                                     struct osd_object *obj,
5451                                     struct lu_fid *fid)
5452 {
5453         struct osd_thread_info  *oti    = osd_oti_get(env);
5454         struct lu_buf           *buf    = &oti->oti_big_buf;
5455         struct dentry           *dentry = &oti->oti_obj_dentry;
5456         struct inode            *inode  = obj->oo_inode;
5457         struct linkea_data       ldata  = { NULL };
5458         int                      rc;
5459         ENTRY;
5460
5461         fid_zero(fid);
5462         if (!S_ISDIR(inode->i_mode))
5463                 RETURN(-EIO);
5464
5465 again:
5466         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5467                              buf->lb_buf, buf->lb_len);
5468         if (rc == -ERANGE) {
5469                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5470                                      NULL, 0);
5471                 if (rc > 0) {
5472                         lu_buf_realloc(buf, rc);
5473                         if (buf->lb_buf == NULL)
5474                                 RETURN(-ENOMEM);
5475
5476                         goto again;
5477                 }
5478         }
5479
5480         if (unlikely(rc == 0))
5481                 RETURN(-ENODATA);
5482
5483         if (rc < 0)
5484                 RETURN(rc);
5485
5486         if (unlikely(buf->lb_buf == NULL)) {
5487                 lu_buf_realloc(buf, rc);
5488                 if (buf->lb_buf == NULL)
5489                         RETURN(-ENOMEM);
5490
5491                 goto again;
5492         }
5493
5494         ldata.ld_buf = buf;
5495         rc = linkea_init_with_rec(&ldata);
5496         if (!rc) {
5497                 linkea_first_entry(&ldata);
5498                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5499         }
5500
5501         RETURN(rc);
5502 }
5503
5504 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5505                                     struct inode *inode,
5506                                     const struct lu_fid *pfid,
5507                                     const char *name, const int namelen)
5508 {
5509         struct osd_thread_info *oti = osd_oti_get(env);
5510         struct lu_buf *buf = &oti->oti_big_buf;
5511         struct dentry *dentry = &oti->oti_obj_dentry;
5512         struct linkea_data ldata = { NULL };
5513         struct lu_name cname = { .ln_name = name,
5514                                  .ln_namelen = namelen };
5515         int rc;
5516         ENTRY;
5517
5518 again:
5519         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5520                              buf->lb_buf, buf->lb_len);
5521         if (rc == -ERANGE)
5522                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5523
5524         if (rc < 0)
5525                 RETURN(rc);
5526
5527         if (unlikely(rc == 0))
5528                 RETURN(-ENODATA);
5529
5530         if (buf->lb_len < rc) {
5531                 lu_buf_realloc(buf, rc);
5532                 if (buf->lb_buf == NULL)
5533                         RETURN(-ENOMEM);
5534
5535                 goto again;
5536         }
5537
5538         ldata.ld_buf = buf;
5539         rc = linkea_init_with_rec(&ldata);
5540         if (!rc)
5541                 rc = linkea_links_find(&ldata, &cname, pfid);
5542
5543         RETURN(rc);
5544 }
5545
5546 /**
5547  * Calls ->lookup() to find dentry. From dentry get inode and
5548  * read inode's ea to get fid. This is required for  interoperability
5549  * mode (b11826)
5550  *
5551  * \retval   0, on success
5552  * \retval -ve, on error
5553  */
5554 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5555                              struct dt_rec *rec, const struct dt_key *key)
5556 {
5557         struct inode                    *dir    = obj->oo_inode;
5558         struct dentry                   *dentry;
5559         struct ldiskfs_dir_entry_2      *de;
5560         struct buffer_head              *bh;
5561         struct lu_fid                   *fid = (struct lu_fid *) rec;
5562         struct htree_lock               *hlock = NULL;
5563         int                             ino;
5564         int                             rc;
5565         ENTRY;
5566
5567         LASSERT(dir->i_op != NULL);
5568         LASSERT(dir->i_op->lookup != NULL);
5569
5570         dentry = osd_child_dentry_get(env, obj,
5571                                       (char *)key, strlen((char *)key));
5572
5573         if (obj->oo_hl_head != NULL) {
5574                 hlock = osd_oti_get(env)->oti_hlock;
5575                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5576                                    dir, LDISKFS_HLOCK_LOOKUP);
5577         } else {
5578                 down_read(&obj->oo_ext_idx_sem);
5579         }
5580
5581         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5582         if (!IS_ERR(bh)) {
5583                 struct osd_thread_info *oti = osd_oti_get(env);
5584                 struct osd_inode_id *id = &oti->oti_id;
5585                 struct osd_idmap_cache *oic = &oti->oti_cache;
5586                 struct osd_device *dev = osd_obj2dev(obj);
5587
5588                 ino = le32_to_cpu(de->inode);
5589                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5590                         brelse(bh);
5591                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5592                         GOTO(out, rc);
5593                 }
5594
5595                 rc = osd_get_fid_from_dentry(de, rec);
5596
5597                 /* done with de, release bh */
5598                 brelse(bh);
5599                 if (rc != 0) {
5600                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5601                                 const char *name = (const char *)key;
5602
5603                                 /* If the parent is on remote MDT, and there
5604                                  * is no FID-in-dirent, then we have to get
5605                                  * the parent FID from the linkEA.  */
5606                                 if (likely(strlen(name) == 2 &&
5607                                            name[0] == '.' && name[1] == '.'))
5608                                         rc = osd_get_pfid_from_linkea(env, obj,
5609                                                                       fid);
5610                         } else {
5611                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5612                         }
5613                 } else {
5614                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5615                 }
5616
5617                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5618                         fid_zero(&oic->oic_fid);
5619
5620                         GOTO(out, rc);
5621                 }
5622
5623                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5624                 rc = osd_consistency_check(oti, dev, oic);
5625                 if (rc == -ENOENT)
5626                         fid_zero(&oic->oic_fid);
5627                 else
5628                         /* Other error should not affect lookup result. */
5629                         rc = 0;
5630         } else {
5631                 rc = PTR_ERR(bh);
5632         }
5633
5634         GOTO(out, rc);
5635
5636 out:
5637         if (hlock != NULL)
5638                 ldiskfs_htree_unlock(hlock);
5639         else
5640                 up_read(&obj->oo_ext_idx_sem);
5641         return rc;
5642 }
5643
5644 static int osd_index_declare_ea_insert(const struct lu_env *env,
5645                                        struct dt_object *dt,
5646                                        const struct dt_rec *rec,
5647                                        const struct dt_key *key,
5648                                        struct thandle *handle)
5649 {
5650         struct osd_thandle      *oh;
5651         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5652         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5653         const struct lu_fid     *fid    = rec1->rec_fid;
5654         int                      credits, rc = 0;
5655         struct osd_idmap_cache  *idc;
5656         ENTRY;
5657
5658         LASSERT(!dt_object_remote(dt));
5659         LASSERT(handle != NULL);
5660         LASSERT(fid != NULL);
5661         LASSERT(rec1->rec_type != 0);
5662
5663         oh = container_of0(handle, struct osd_thandle, ot_super);
5664         LASSERT(oh->ot_handle == NULL);
5665
5666         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5667
5668         /* we can't call iget() while a transactions is running
5669          * (this can lead to a deadlock), but we need to know
5670          * inum and object type. so we find this information at
5671          * declaration and cache in per-thread info */
5672         idc = osd_idc_find_or_init(env, osd, fid);
5673         if (IS_ERR(idc))
5674                 RETURN(PTR_ERR(idc));
5675         if (idc->oic_remote) {
5676                 /* a reference to remote inode is represented by an
5677                  * agent inode which we have to create */
5678                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5679                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5680         }
5681
5682         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5683
5684         if (osd_dt_obj(dt)->oo_inode != NULL) {
5685                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5686
5687                 /* We ignore block quota on meta pool (MDTs), so needn't
5688                  * calculate how many blocks will be consumed by this index
5689                  * insert */
5690                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5691                                            i_gid_read(inode),
5692                                            i_projid_read(inode), 0,
5693                                            oh, osd_dt_obj(dt), NULL,
5694                                            OSD_QID_BLK);
5695                 if (rc)
5696                         RETURN(rc);
5697
5698 #ifdef HAVE_PROJECT_QUOTA
5699                 /* Reserve credits for local agent inode to transfer
5700                  * to 0, quota enforcement is ignored in this case.
5701                  */
5702                 if (idc->oic_remote &&
5703                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5704                     i_projid_read(inode) != 0)
5705                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5706                                                   0, i_projid_read(inode),
5707                                                   0, false, PRJQUOTA, true);
5708 #endif
5709         }
5710
5711
5712         RETURN(rc);
5713 }
5714
5715 /**
5716  * Index add function for interoperability mode (b11826).
5717  * It will add the directory entry.This entry is needed to
5718  * maintain name->fid mapping.
5719  *
5720  * \param key it is key i.e. file entry to be inserted
5721  * \param rec it is value of given key i.e. fid
5722  *
5723  * \retval   0, on success
5724  * \retval -ve, on error
5725  */
5726 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5727                                const struct dt_rec *rec,
5728                                const struct dt_key *key, struct thandle *th,
5729                                int ignore_quota)
5730 {
5731         struct osd_object       *obj = osd_dt_obj(dt);
5732         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5733         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5734         const struct lu_fid     *fid    = rec1->rec_fid;
5735         const char              *name = (const char *)key;
5736         struct osd_thread_info  *oti   = osd_oti_get(env);
5737         struct inode            *child_inode = NULL;
5738         struct osd_idmap_cache  *idc;
5739         int                     rc;
5740         ENTRY;
5741
5742         if (!dt_object_exists(dt))
5743                 RETURN(-ENOENT);
5744
5745         LASSERT(osd_invariant(obj));
5746         LASSERT(!dt_object_remote(dt));
5747         LASSERT(th != NULL);
5748
5749         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5750
5751         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5752
5753         idc = osd_idc_find(env, osd, fid);
5754         if (unlikely(idc == NULL)) {
5755                 /* this dt_insert() wasn't declared properly, so
5756                  * FID is missing in OI cache. we better do not
5757                  * lookup FID in FLDB/OI and don't risk to deadlock,
5758                  * but in some special cases (lfsck testing, etc)
5759                  * it's much simpler than fixing a caller */
5760                 CERROR("%s: "DFID" wasn't declared for insert\n",
5761                        osd_name(osd), PFID(fid));
5762                 dump_stack();
5763                 idc = osd_idc_find_or_init(env, osd, fid);
5764                 if (IS_ERR(idc))
5765                         RETURN(PTR_ERR(idc));
5766         }
5767
5768         if (idc->oic_remote) {
5769                 /* Insert remote entry */
5770                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5771                         child_inode =
5772                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
5773                 } else {
5774                         child_inode = osd_create_local_agent_inode(env, osd,
5775                                         obj, fid, rec1->rec_type & S_IFMT, th);
5776                         if (IS_ERR(child_inode))
5777                                 RETURN(PTR_ERR(child_inode));
5778                 }
5779         } else {
5780                 /* Insert local entry */
5781                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5782                         /* for a reason OI cache wasn't filled properly */
5783                         CERROR("%s: OIC for "DFID" isn't filled\n",
5784                                osd_name(osd), PFID(fid));
5785                         RETURN(-EINVAL);
5786                 }
5787                 child_inode = oti->oti_inode;
5788                 if (unlikely(child_inode == NULL)) {
5789                         struct ldiskfs_inode_info *lii;
5790                         OBD_ALLOC_PTR(lii);
5791                         if (lii == NULL)
5792                                 RETURN(-ENOMEM);
5793                         child_inode = oti->oti_inode = &lii->vfs_inode;
5794                 }
5795                 child_inode->i_sb = osd_sb(osd);
5796                 child_inode->i_ino = idc->oic_lid.oii_ino;
5797                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5798         }
5799
5800         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5801
5802         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5803                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5804
5805         if (child_inode && child_inode != oti->oti_inode)
5806                 iput(child_inode);
5807         LASSERT(osd_invariant(obj));
5808         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5809         RETURN(rc);
5810 }
5811
5812 /**
5813  *  Initialize osd Iterator for given osd index object.
5814  *
5815  *  \param  dt      osd index object
5816  */
5817
5818 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5819                                      struct dt_object *dt,
5820                                      __u32 unused)
5821 {
5822         struct osd_it_iam      *it;
5823         struct osd_object      *obj = osd_dt_obj(dt);
5824         struct lu_object       *lo  = &dt->do_lu;
5825         struct iam_path_descr  *ipd;
5826         struct iam_container   *bag = &obj->oo_dir->od_container;
5827
5828         if (!dt_object_exists(dt))
5829                 return ERR_PTR(-ENOENT);
5830
5831         OBD_ALLOC_PTR(it);
5832         if (it == NULL)
5833                 return ERR_PTR(-ENOMEM);
5834
5835         ipd = osd_it_ipd_get(env, bag);
5836         if (likely(ipd != NULL)) {
5837                 it->oi_obj = obj;
5838                 it->oi_ipd = ipd;
5839                 lu_object_get(lo);
5840                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5841                 return (struct dt_it *)it;
5842         } else {
5843                 OBD_FREE_PTR(it);
5844                 return ERR_PTR(-ENOMEM);
5845         }
5846 }
5847
5848 /**
5849  * free given Iterator.
5850  */
5851 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5852 {
5853         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5854         struct osd_object       *obj = it->oi_obj;
5855
5856         iam_it_fini(&it->oi_it);
5857         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5858         osd_object_put(env, obj);
5859         OBD_FREE_PTR(it);
5860 }
5861
5862 /**
5863  *  Move Iterator to record specified by \a key
5864  *
5865  *  \param  di      osd iterator
5866  *  \param  key     key for index
5867  *
5868  *  \retval +ve  di points to record with least key not larger than key
5869  *  \retval  0   di points to exact matched key
5870  *  \retval -ve  failure
5871  */
5872
5873 static int osd_it_iam_get(const struct lu_env *env,
5874                           struct dt_it *di, const struct dt_key *key)
5875 {
5876         struct osd_thread_info  *oti = osd_oti_get(env);
5877         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5878
5879         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5880                 /* swab quota uid/gid */
5881                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5882                 key = (struct dt_key *)&oti->oti_quota_id;
5883         }
5884
5885         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5886 }
5887
5888 /**
5889  *  Release Iterator
5890  *
5891  *  \param  di      osd iterator
5892  */
5893 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5894 {
5895         struct osd_it_iam *it = (struct osd_it_iam *)di;
5896
5897         iam_it_put(&it->oi_it);
5898 }
5899
5900 /**
5901  *  Move iterator by one record
5902  *
5903  *  \param  di      osd iterator
5904  *
5905  *  \retval +1   end of container reached
5906  *  \retval  0   success
5907  *  \retval -ve  failure
5908  */
5909
5910 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5911 {
5912         struct osd_it_iam *it = (struct osd_it_iam *)di;
5913
5914         return iam_it_next(&it->oi_it);
5915 }
5916
5917 /**
5918  * Return pointer to the key under iterator.
5919  */
5920
5921 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5922                                  const struct dt_it *di)
5923 {
5924         struct osd_thread_info *oti = osd_oti_get(env);
5925         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5926         struct osd_object      *obj = it->oi_obj;
5927         struct dt_key          *key;
5928
5929         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5930
5931         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5932                 /* swab quota uid/gid */
5933                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5934                 key = (struct dt_key *)&oti->oti_quota_id;
5935         }
5936
5937         return key;
5938 }
5939
5940 /**
5941  * Return size of key under iterator (in bytes)
5942  */
5943
5944 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5945 {
5946         struct osd_it_iam *it = (struct osd_it_iam *)di;
5947
5948         return iam_it_key_size(&it->oi_it);
5949 }
5950
5951 static inline void
5952 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5953 {
5954         /* check if file type is required */
5955         if (ent->lde_attrs & LUDA_TYPE) {
5956                 struct luda_type *lt;
5957                 int align = sizeof(*lt) - 1;
5958
5959                 len = (len + align) & ~align;
5960                 lt = (struct luda_type *)(ent->lde_name + len);
5961                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5962         }
5963
5964         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5965 }
5966
5967 /**
5968  * build lu direct from backend fs dirent.
5969  */
5970
5971 static inline void
5972 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5973                    char *name, __u16 namelen, __u16 type, __u32 attr)
5974 {
5975         ent->lde_attrs = attr | LUDA_FID;
5976         fid_cpu_to_le(&ent->lde_fid, fid);
5977
5978         ent->lde_hash = cpu_to_le64(offset);
5979         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5980
5981         strncpy(ent->lde_name, name, namelen);
5982         ent->lde_name[namelen] = '\0';
5983         ent->lde_namelen = cpu_to_le16(namelen);
5984
5985         /* append lustre attributes */
5986         osd_it_append_attrs(ent, namelen, type);
5987 }
5988
5989 /**
5990  * Return pointer to the record under iterator.
5991  */
5992 static int osd_it_iam_rec(const struct lu_env *env,
5993                           const struct dt_it *di,
5994                           struct dt_rec *dtrec, __u32 attr)
5995 {
5996         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5997         struct osd_thread_info *info = osd_oti_get(env);
5998         ENTRY;
5999
6000         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6001                 const struct osd_fid_pack *rec;
6002                 struct lu_fid             *fid = &info->oti_fid;
6003                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
6004                 char                      *name;
6005                 int                        namelen;
6006                 __u64                      hash;
6007                 int                        rc;
6008
6009                 name = (char *)iam_it_key_get(&it->oi_it);
6010                 if (IS_ERR(name))
6011                         RETURN(PTR_ERR(name));
6012
6013                 namelen = iam_it_key_size(&it->oi_it);
6014
6015                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6016                 if (IS_ERR(rec))
6017                         RETURN(PTR_ERR(rec));
6018
6019                 rc = osd_fid_unpack(fid, rec);
6020                 if (rc)
6021                         RETURN(rc);
6022
6023                 hash = iam_it_store(&it->oi_it);
6024
6025                 /* IAM does not store object type in IAM index (dir) */
6026                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6027                                    0, LUDA_FID);
6028         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6029                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6030                            (struct iam_rec *)dtrec);
6031                 osd_quota_unpack(it->oi_obj, dtrec);
6032         } else {
6033                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6034                            (struct iam_rec *)dtrec);
6035         }
6036
6037         RETURN(0);
6038 }
6039
6040 /**
6041  * Returns cookie for current Iterator position.
6042  */
6043 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6044 {
6045         struct osd_it_iam *it = (struct osd_it_iam *)di;
6046
6047         return iam_it_store(&it->oi_it);
6048 }
6049
6050 /**
6051  * Restore iterator from cookie.
6052  *
6053  * \param  di      osd iterator
6054  * \param  hash    Iterator location cookie
6055  *
6056  * \retval +ve  di points to record with least key not larger than key.
6057  * \retval  0   di points to exact matched key
6058  * \retval -ve  failure
6059  */
6060
6061 static int osd_it_iam_load(const struct lu_env *env,
6062                            const struct dt_it *di, __u64 hash)
6063 {
6064         struct osd_it_iam *it = (struct osd_it_iam *)di;
6065
6066         return iam_it_load(&it->oi_it, hash);
6067 }
6068
6069 static const struct dt_index_operations osd_index_iam_ops = {
6070         .dio_lookup         = osd_index_iam_lookup,
6071         .dio_declare_insert = osd_index_declare_iam_insert,
6072         .dio_insert         = osd_index_iam_insert,
6073         .dio_declare_delete = osd_index_declare_iam_delete,
6074         .dio_delete         = osd_index_iam_delete,
6075         .dio_it     = {
6076                 .init     = osd_it_iam_init,
6077                 .fini     = osd_it_iam_fini,
6078                 .get      = osd_it_iam_get,
6079                 .put      = osd_it_iam_put,
6080                 .next     = osd_it_iam_next,
6081                 .key      = osd_it_iam_key,
6082                 .key_size = osd_it_iam_key_size,
6083                 .rec      = osd_it_iam_rec,
6084                 .store    = osd_it_iam_store,
6085                 .load     = osd_it_iam_load
6086         }
6087 };
6088
6089
6090 /**
6091  * Creates or initializes iterator context.
6092  *
6093  * \retval struct osd_it_ea, iterator structure on success
6094  *
6095  */
6096 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6097                                     struct dt_object *dt,
6098                                     __u32 attr)
6099 {
6100         struct osd_object       *obj  = osd_dt_obj(dt);
6101         struct osd_thread_info  *info = osd_oti_get(env);
6102         struct osd_it_ea        *oie;
6103         struct file             *file;
6104         struct lu_object        *lo   = &dt->do_lu;
6105         struct dentry           *obj_dentry;
6106         ENTRY;
6107
6108         if (!dt_object_exists(dt) || obj->oo_destroyed)
6109                 RETURN(ERR_PTR(-ENOENT));
6110
6111         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6112         if (oie == NULL)
6113                 RETURN(ERR_PTR(-ENOMEM));
6114         obj_dentry = &oie->oie_dentry;
6115
6116         obj_dentry->d_inode = obj->oo_inode;
6117         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6118         obj_dentry->d_name.hash = 0;
6119
6120         oie->oie_rd_dirent       = 0;
6121         oie->oie_it_dirent       = 0;
6122         oie->oie_dirent          = NULL;
6123         if (unlikely(!info->oti_it_ea_buf_used)) {
6124                 oie->oie_buf = info->oti_it_ea_buf;
6125                 info->oti_it_ea_buf_used = 1;
6126         } else {
6127                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6128                 if (oie->oie_buf == NULL)
6129                         RETURN(ERR_PTR(-ENOMEM));
6130         }
6131         oie->oie_obj             = obj;
6132
6133         file = &oie->oie_file;
6134
6135         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6136         if (attr & LUDA_64BITHASH)
6137                 file->f_mode    = FMODE_64BITHASH;
6138         else
6139                 file->f_mode    = FMODE_32BITHASH;
6140         file->f_path.dentry     = obj_dentry;
6141         file->f_mapping         = obj->oo_inode->i_mapping;
6142         file->f_op              = obj->oo_inode->i_fop;
6143         set_file_inode(file, obj->oo_inode);
6144
6145         lu_object_get(lo);
6146         RETURN((struct dt_it *) oie);
6147 }
6148
6149 /**
6150  * Destroy or finishes iterator context.
6151  *
6152  * \param di iterator structure to be destroyed
6153  */
6154 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6155 {
6156         struct osd_thread_info  *info = osd_oti_get(env);
6157         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
6158         struct osd_object       *obj    = oie->oie_obj;
6159         struct inode            *inode  = obj->oo_inode;
6160
6161         ENTRY;
6162         oie->oie_file.f_op->release(inode, &oie->oie_file);
6163         osd_object_put(env, obj);
6164         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6165                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6166         else
6167                 info->oti_it_ea_buf_used = 0;
6168         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6169         EXIT;
6170 }
6171
6172 /**
6173  * It position the iterator at given key, so that next lookup continues from
6174  * that key Or it is similar to dio_it->load() but based on a key,
6175  * rather than file position.
6176  *
6177  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6178  * to the beginning.
6179  *
6180  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6181  */
6182 static int osd_it_ea_get(const struct lu_env *env,
6183                          struct dt_it *di, const struct dt_key *key)
6184 {
6185         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
6186
6187         ENTRY;
6188         LASSERT(((const char *)key)[0] == '\0');
6189         it->oie_file.f_pos      = 0;
6190         it->oie_rd_dirent       = 0;
6191         it->oie_it_dirent       = 0;
6192         it->oie_dirent          = NULL;
6193
6194         RETURN(+1);
6195 }
6196
6197 /**
6198  * Does nothing
6199  */
6200 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6201 {
6202 }
6203
6204 struct osd_filldir_cbs {
6205 #ifdef HAVE_DIR_CONTEXT
6206         struct dir_context ctx;
6207 #endif
6208         struct osd_it_ea  *it;
6209 };
6210 /**
6211  * It is called internally by ->readdir(). It fills the
6212  * iterator's in-memory data structure with required
6213  * information i.e. name, namelen, rec_size etc.
6214  *
6215  * \param buf in which information to be filled in.
6216  * \param name name of the file in given dir
6217  *
6218  * \retval 0 on success
6219  * \retval 1 on buffer full
6220  */
6221 #ifdef HAVE_FILLDIR_USE_CTX
6222 static int osd_ldiskfs_filldir(struct dir_context *buf,
6223 #else
6224 static int osd_ldiskfs_filldir(void *buf,
6225 #endif
6226                                const char *name, int namelen,
6227                                loff_t offset, __u64 ino, unsigned d_type)
6228 {
6229         struct osd_it_ea        *it   =
6230                 ((struct osd_filldir_cbs *)buf)->it;
6231         struct osd_object       *obj  = it->oie_obj;
6232         struct osd_it_ea_dirent *ent  = it->oie_dirent;
6233         struct lu_fid           *fid  = &ent->oied_fid;
6234         struct osd_fid_pack     *rec;
6235         ENTRY;
6236
6237         /* this should never happen */
6238         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6239                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6240                 RETURN(-EIO);
6241         }
6242
6243         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
6244             OSD_IT_EA_BUFSIZE)
6245                 RETURN(1);
6246
6247         /* "." is just the object itself. */
6248         if (namelen == 1 && name[0] == '.') {
6249                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6250         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6251                 rec = (struct osd_fid_pack*) (name + namelen + 1);
6252                 if (osd_fid_unpack(fid, rec) != 0)
6253                         fid_zero(fid);
6254         } else {
6255                 fid_zero(fid);
6256         }
6257         d_type &= ~LDISKFS_DIRENT_LUFID;
6258
6259         /* NOT export local root. */
6260         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6261                 ino = obj->oo_inode->i_ino;
6262                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6263         }
6264
6265         ent->oied_ino     = ino;
6266         ent->oied_off     = offset;
6267         ent->oied_namelen = namelen;
6268         ent->oied_type    = d_type;
6269
6270         memcpy(ent->oied_name, name, namelen);
6271
6272         it->oie_rd_dirent++;
6273         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6274         RETURN(0);
6275 }
6276
6277 /**
6278  * Calls ->readdir() to load a directory entry at a time
6279  * and stored it in iterator's in-memory data structure.
6280  *
6281  * \param di iterator's in memory structure
6282  *
6283  * \retval   0 on success
6284  * \retval -ve on error
6285  * \retval +1 reach the end of entry
6286  */
6287 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6288                                const struct dt_it *di)
6289 {
6290         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
6291         struct osd_object  *obj   = it->oie_obj;
6292         struct inode       *inode = obj->oo_inode;
6293         struct htree_lock  *hlock = NULL;
6294         struct file        *filp  = &it->oie_file;
6295         int                 rc = 0;
6296         struct osd_filldir_cbs buf = {
6297 #ifdef HAVE_DIR_CONTEXT
6298                 .ctx.actor = osd_ldiskfs_filldir,
6299 #endif
6300                 .it = it
6301         };
6302
6303         ENTRY;
6304         it->oie_dirent = it->oie_buf;
6305         it->oie_rd_dirent = 0;
6306
6307         if (obj->oo_hl_head != NULL) {
6308                 hlock = osd_oti_get(env)->oti_hlock;
6309                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6310                                    inode, LDISKFS_HLOCK_READDIR);
6311         } else {
6312                 down_read(&obj->oo_ext_idx_sem);
6313         }
6314
6315 #ifdef HAVE_DIR_CONTEXT
6316         buf.ctx.pos = filp->f_pos;
6317         rc = inode->i_fop->iterate(filp, &buf.ctx);
6318         filp->f_pos = buf.ctx.pos;
6319 #else
6320         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6321 #endif
6322
6323         if (hlock != NULL)
6324                 ldiskfs_htree_unlock(hlock);
6325         else
6326                 up_read(&obj->oo_ext_idx_sem);
6327
6328         if (it->oie_rd_dirent == 0) {
6329                 /*If it does not get any dirent, it means it has been reached
6330                  *to the end of the dir */
6331                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6332                 if (rc == 0)
6333                         rc = 1;
6334         } else {
6335                 it->oie_dirent = it->oie_buf;
6336                 it->oie_it_dirent = 1;
6337         }
6338
6339         RETURN(rc);
6340 }
6341
6342 /**
6343  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6344  * to load a directory entry at a time and stored it in
6345  * iterator's in-memory data structure.
6346  *
6347  * \param di iterator's in memory structure
6348  *
6349  * \retval +ve iterator reached to end
6350  * \retval   0 iterator not reached to end
6351  * \retval -ve on error
6352  */
6353 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6354 {
6355         struct osd_it_ea *it = (struct osd_it_ea *)di;
6356         int rc;
6357
6358         ENTRY;
6359
6360         if (it->oie_it_dirent < it->oie_rd_dirent) {
6361                 it->oie_dirent =
6362                         (void *) it->oie_dirent +
6363                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6364                                        it->oie_dirent->oied_namelen);
6365                 it->oie_it_dirent++;
6366                 RETURN(0);
6367         } else {
6368                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6369                         rc = +1;
6370                 else
6371                         rc = osd_ldiskfs_it_fill(env, di);
6372         }
6373
6374         RETURN(rc);
6375 }
6376
6377 /**
6378  * Returns the key at current position from iterator's in memory structure.
6379  *
6380  * \param di iterator's in memory structure
6381  *
6382  * \retval key i.e. struct dt_key on success
6383  */
6384 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6385                                     const struct dt_it *di)
6386 {
6387         struct osd_it_ea *it = (struct osd_it_ea *)di;
6388
6389         return (struct dt_key *)it->oie_dirent->oied_name;
6390 }
6391
6392 /**
6393  * Returns the key's size at current position from iterator's in memory structure.
6394  *
6395  * \param di iterator's in memory structure
6396  *
6397  * \retval key_size i.e. struct dt_key on success
6398  */
6399 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6400 {
6401         struct osd_it_ea *it = (struct osd_it_ea *)di;
6402
6403         return it->oie_dirent->oied_namelen;
6404 }
6405
6406 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6407 {
6408         if (LDISKFS_DIR_REC_LEN(de) >=
6409             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6410                 return true;
6411
6412         return false;
6413 }
6414
6415 static inline bool
6416 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6417                      unsigned blocksize, bool dotdot)
6418 {
6419         if (dotdot)
6420                 return osd_dotdot_has_space(de);
6421
6422         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6423             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6424                 return true;
6425
6426         return false;
6427 }
6428
6429 static int
6430 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6431                     handle_t *jh, struct dentry *dentry,
6432                     const struct lu_fid *fid, struct buffer_head *bh,
6433                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6434                     bool dotdot)
6435 {
6436         struct inode                *dir        = dentry->d_parent->d_inode;
6437         struct inode                *inode      = dentry->d_inode;
6438         struct osd_fid_pack         *rec;
6439         struct ldiskfs_dentry_param *ldp;
6440         int                          namelen    = dentry->d_name.len;
6441         int                          rc;
6442         struct osd_thread_info     *info        = osd_oti_get(env);
6443         ENTRY;
6444
6445         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6446                 RETURN(0);
6447
6448         /* There is enough space to hold the FID-in-dirent. */
6449         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6450                 rc = ldiskfs_journal_get_write_access(jh, bh);
6451                 if (rc != 0)
6452                         RETURN(rc);
6453
6454                 de->name[namelen] = 0;
6455                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6456                 rec->fp_len = sizeof(struct lu_fid) + 1;
6457                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6458                 de->file_type |= LDISKFS_DIRENT_LUFID;
6459                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6460
6461                 RETURN(rc);
6462         }
6463
6464         LASSERT(!dotdot);
6465
6466         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6467         if (rc != 0)
6468                 RETURN(rc);
6469
6470         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6471         osd_get_ldiskfs_dirent_param(ldp, fid);
6472         dentry->d_fsdata = (void *)ldp;
6473         ll_vfs_dq_init(dir);
6474         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6475         /* It is too bad, we cannot reinsert the name entry back.
6476          * That means we lose it! */
6477         if (rc != 0)
6478                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6479                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6480                        osd_ino2name(inode),
6481                        dir->i_ino, dir->i_generation, namelen,
6482                        dentry->d_name.name, PFID(fid), rc);
6483
6484         RETURN(rc);
6485 }
6486
6487 static int
6488 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6489                         struct osd_it_ea *it, struct lu_fid *fid,
6490                         struct osd_inode_id *id, __u32 *attr)
6491 {
6492         struct osd_thread_info     *info        = osd_oti_get(env);
6493         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6494         struct osd_device          *dev         = osd_obj2dev(obj);
6495         struct super_block         *sb          = osd_sb(dev);
6496         const char                 *devname     = osd_name(dev);
6497         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6498         struct inode               *dir         = obj->oo_inode;
6499         struct htree_lock          *hlock       = NULL;
6500         struct buffer_head         *bh          = NULL;
6501         handle_t                   *jh          = NULL;
6502         struct ldiskfs_dir_entry_2 *de;
6503         struct dentry              *dentry;
6504         struct inode               *inode;
6505         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6506         int                         credits;
6507         int                         rc;
6508         bool                        dotdot      = false;
6509         bool                        dirty       = false;
6510         ENTRY;
6511
6512         if (ent->oied_name[0] == '.') {
6513                 if (ent->oied_namelen == 1)
6514                         RETURN(0);
6515
6516                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6517                         dotdot = true;
6518         }
6519
6520         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6521         inode = osd_iget(info, dev, id);
6522         if (IS_ERR(inode)) {
6523                 rc = PTR_ERR(inode);
6524                 if (rc == -ENOENT || rc == -ESTALE) {
6525                         /* Maybe dangling name entry, or
6526                          * corrupted directory entry. */
6527                         *attr |= LUDA_UNKNOWN;
6528                         rc = 0;
6529                 } else {
6530                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6531                                "check_repair, dir = %lu/%u, name = %.*s, "
6532                                "ino = %llu, rc = %d\n",
6533                                devname, dir->i_ino, dir->i_generation,
6534                                ent->oied_namelen, ent->oied_name,
6535                                ent->oied_ino, rc);
6536                 }
6537
6538                 RETURN(rc);
6539         }
6540
6541         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6542                                            ent->oied_namelen);
6543         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6544         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6545                 lma = NULL;
6546         else if (rc != 0)
6547                 GOTO(out, rc);
6548
6549         /* We need to ensure that the name entry is still valid.
6550          * Because it may be removed or renamed by other already.
6551          *
6552          * The unlink or rename operation will start journal before PDO lock,
6553          * so to avoid deadlock, here we need to start journal handle before
6554          * related PDO lock also. But because we do not know whether there
6555          * will be something to be repaired before PDO lock, we just start
6556          * journal without conditions.
6557          *
6558          * We may need to remove the name entry firstly, then insert back.
6559          * One credit is for user quota file update.
6560          * One credit is for group quota file update.
6561          * Two credits are for dirty inode. */
6562         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6563                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6564
6565         if (dev->od_dirent_journal != 0) {
6566
6567 again:
6568                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6569                 if (IS_ERR(jh)) {
6570                         rc = PTR_ERR(jh);
6571                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6572                                "check_repair, dir = %lu/%u, credits = %d, "
6573                                "name = %.*s, ino = %llu: rc = %d\n",
6574                                devname, dir->i_ino, dir->i_generation, credits,
6575                                ent->oied_namelen, ent->oied_name,
6576                                ent->oied_ino, rc);
6577
6578                         GOTO(out_inode, rc);
6579                 }
6580
6581                 if (obj->oo_hl_head != NULL) {
6582                         hlock = osd_oti_get(env)->oti_hlock;
6583                         /* "0" means exclusive lock for the whole directory.
6584                          * We need to prevent others access such name entry
6585                          * during the delete + insert. Neither HLOCK_ADD nor
6586                          * HLOCK_DEL cannot guarantee the atomicity. */
6587                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6588                 } else {
6589                         down_write(&obj->oo_ext_idx_sem);
6590                 }
6591         } else {
6592                 if (obj->oo_hl_head != NULL) {
6593                         hlock = osd_oti_get(env)->oti_hlock;
6594                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6595                                            LDISKFS_HLOCK_LOOKUP);
6596                 } else {
6597                         down_read(&obj->oo_ext_idx_sem);
6598                 }
6599         }
6600
6601         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6602         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6603                 *attr |= LUDA_IGNORE;
6604
6605                 GOTO(out, rc = 0);
6606         }
6607
6608         /* For dotdot entry, if there is not enough space to hold the
6609          * FID-in-dirent, just keep them there. It only happens when the
6610          * device upgraded from 1.8 or restored from MDT file-level backup.
6611          * For the whole directory, only dotdot entry have no FID-in-dirent
6612          * and needs to get FID from LMA when readdir, it will not affect the
6613          * performance much. */
6614         if (dotdot && !osd_dotdot_has_space(de)) {
6615                 *attr |= LUDA_UNKNOWN;
6616
6617                 GOTO(out, rc = 0);
6618         }
6619
6620         if (lma != NULL) {
6621                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6622                         GOTO(out, rc = 0);
6623
6624                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6625                         struct lu_fid *tfid = &lma->lma_self_fid;
6626
6627                         if (likely(dotdot &&
6628                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6629                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6630                                 /* It must be REMOTE_PARENT_DIR and as the
6631                                  * 'dotdot' entry of remote directory */
6632                                 *attr |= LUDA_IGNORE;
6633                         } else {
6634                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6635                                        "parent directory, but got %.*s under "
6636                                        "dir = %lu/%u with the FID "DFID"\n",
6637                                        devname, ent->oied_namelen,
6638                                        ent->oied_name, dir->i_ino,
6639                                        dir->i_generation, PFID(tfid));
6640
6641                                 *attr |= LUDA_UNKNOWN;
6642                         }
6643
6644                         GOTO(out, rc = 0);
6645                 }
6646         }
6647
6648         if (!fid_is_zero(fid)) {
6649                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6650                                               ent->oied_namelen);
6651                 if (rc == -ENOENT ||
6652                     (rc == -ENODATA &&
6653                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6654                         /* linkEA does not recognize the dirent entry,
6655                          * it may because the dirent entry corruption
6656                          * and points to other's inode. */
6657                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6658                                "recognize the dirent, dir = %lu/%u, "
6659                                " name = %.*s, ino = %llu, "
6660                                DFID": rc = %d\n", devname, dir->i_ino,
6661                                dir->i_generation, ent->oied_namelen,
6662                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6663                         *attr |= LUDA_UNKNOWN;
6664
6665                         GOTO(out, rc = 0);
6666                 }
6667
6668                 if (rc && rc != -ENODATA) {
6669                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6670                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6671                                DFID": rc = %d\n", devname, dir->i_ino,
6672                                dir->i_generation, ent->oied_namelen,
6673                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6674                         *attr |= LUDA_UNKNOWN;
6675
6676                         GOTO(out, rc = 0);
6677                 }
6678         }
6679
6680         if (lma != NULL) {
6681                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6682                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6683                 if (*attr & LUDA_VERIFY_DRYRUN) {
6684                         *fid = lma->lma_self_fid;
6685                         *attr |= LUDA_REPAIR;
6686
6687                         GOTO(out, rc = 0);
6688                 }
6689
6690                 if (jh == NULL) {
6691                         brelse(bh);
6692                         dev->od_dirent_journal = 1;
6693                         if (hlock != NULL) {
6694                                 ldiskfs_htree_unlock(hlock);
6695                                 hlock = NULL;
6696                         } else {
6697                                 up_read(&obj->oo_ext_idx_sem);
6698                         }
6699
6700                         goto again;
6701                 }
6702
6703                 *fid = lma->lma_self_fid;
6704                 dirty = true;
6705                 /* Update or append the FID-in-dirent. */
6706                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6707                                          bh, de, hlock, dotdot);
6708                 if (rc == 0)
6709                         *attr |= LUDA_REPAIR;
6710                 else
6711                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6712                                "the dirent, dir = %lu/%u, name = %.*s, "
6713                                "ino = %llu, "DFID": rc = %d\n",
6714                                devname, dir->i_ino, dir->i_generation,
6715                                ent->oied_namelen, ent->oied_name,
6716                                ent->oied_ino, PFID(fid), rc);
6717         } else {
6718                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6719                 if (*attr & LUDA_VERIFY_DRYRUN) {
6720                         if (fid_is_sane(fid)) {
6721                                 *attr |= LUDA_REPAIR;
6722                         } else if (dev->od_index == 0) {
6723                                 lu_igif_build(fid, inode->i_ino,
6724                                               inode->i_generation);
6725                                 *attr |= LUDA_UPGRADE;
6726                         }
6727
6728                         GOTO(out, rc = 0);
6729                 }
6730
6731                 if (jh == NULL) {
6732                         brelse(bh);
6733                         dev->od_dirent_journal = 1;
6734                         if (hlock != NULL) {
6735                                 ldiskfs_htree_unlock(hlock);
6736                                 hlock = NULL;
6737                         } else {
6738                                 up_read(&obj->oo_ext_idx_sem);
6739                         }
6740
6741                         goto again;
6742                 }
6743
6744                 dirty = true;
6745                 if (unlikely(fid_is_sane(fid))) {
6746                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6747                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6748                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6749                         if (rc == 0)
6750                                 *attr |= LUDA_REPAIR;
6751                         else
6752                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6753                                        "update dirent, dir = %lu/%u, "
6754                                        "name = %.*s, ino = %llu, "
6755                                        DFID": rc = %d\n",
6756                                        devname, dir->i_ino, dir->i_generation,
6757                                        ent->oied_namelen, ent->oied_name,
6758                                        ent->oied_ino, PFID(fid), rc);
6759                 } else if (dev->od_index == 0) {
6760                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6761                         /* It is probably IGIF object. Only aappend the
6762                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6763                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6764                                                  bh, de, hlock, dotdot);
6765                         if (rc == 0)
6766                                 *attr |= LUDA_UPGRADE;
6767                         else
6768                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6769                                        "after the dirent, dir = %lu/%u, "
6770                                        "name = %.*s, ino = %llu, "
6771                                        DFID": rc = %d\n",
6772                                        devname, dir->i_ino, dir->i_generation,
6773                                        ent->oied_namelen, ent->oied_name,
6774                                        ent->oied_ino, PFID(fid), rc);
6775                 }
6776         }
6777
6778         GOTO(out, rc);
6779
6780 out:
6781         if (!IS_ERR(bh))
6782                 brelse(bh);
6783         if (hlock != NULL) {
6784                 ldiskfs_htree_unlock(hlock);
6785         } else {
6786                 if (dev->od_dirent_journal != 0)
6787                         up_write(&obj->oo_ext_idx_sem);
6788                 else
6789                         up_read(&obj->oo_ext_idx_sem);
6790         }
6791
6792         if (jh != NULL)
6793                 ldiskfs_journal_stop(jh);
6794
6795 out_inode:
6796         iput(inode);
6797         if (rc >= 0 && !dirty)
6798                 dev->od_dirent_journal = 0;
6799
6800         return rc;
6801 }
6802
6803 /**
6804  * Returns the value at current position from iterator's in memory structure.
6805  *
6806  * \param di struct osd_it_ea, iterator's in memory structure
6807  * \param attr attr requested for dirent.
6808  * \param lde lustre dirent
6809  *
6810  * \retval   0 no error and \param lde has correct lustre dirent.
6811  * \retval -ve on error
6812  */
6813 static inline int osd_it_ea_rec(const struct lu_env *env,
6814                                 const struct dt_it *di,
6815                                 struct dt_rec *dtrec, __u32 attr)
6816 {
6817         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6818         struct osd_object      *obj   = it->oie_obj;
6819         struct osd_device      *dev   = osd_obj2dev(obj);
6820         struct osd_thread_info *oti   = osd_oti_get(env);
6821         struct osd_inode_id    *id    = &oti->oti_id;
6822         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6823         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6824         __u32                   ino   = it->oie_dirent->oied_ino;
6825         int                     rc    = 0;
6826         ENTRY;
6827
6828         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
6829
6830         if (attr & LUDA_VERIFY) {
6831                 if (unlikely(is_remote_parent_ino(dev, ino))) {
6832                         attr |= LUDA_IGNORE;
6833                         /* If the parent is on remote MDT, and there
6834                          * is no FID-in-dirent, then we have to get
6835                          * the parent FID from the linkEA.  */
6836                         if (!fid_is_sane(fid) &&
6837                             it->oie_dirent->oied_namelen == 2 &&
6838                             it->oie_dirent->oied_name[0] == '.' &&
6839                             it->oie_dirent->oied_name[1] == '.')
6840                                 osd_get_pfid_from_linkea(env, obj, fid);
6841                 } else {
6842                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6843                                                      &attr);
6844                 }
6845
6846                 if (!fid_is_sane(fid))
6847                         attr |= LUDA_UNKNOWN;
6848         } else {
6849                 attr &= ~LU_DIRENT_ATTRS_MASK;
6850                 if (!fid_is_sane(fid)) {
6851                         bool is_dotdot = false;
6852                         if (it->oie_dirent->oied_namelen == 2 &&
6853                             it->oie_dirent->oied_name[0] == '.' &&
6854                             it->oie_dirent->oied_name[1] == '.')
6855                                 is_dotdot = true;
6856                         /* If the parent is on remote MDT, and there
6857                          * is no FID-in-dirent, then we have to get
6858                          * the parent FID from the linkEA.  */
6859                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
6860                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6861                         } else {
6862                                 if (is_dotdot == false &&
6863                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6864                                         RETURN(-ENOENT);
6865
6866                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6867                         }
6868                 } else {
6869                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6870                 }
6871         }
6872
6873         /* Pack the entry anyway, at least the offset is right. */
6874         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6875                            it->oie_dirent->oied_name,
6876                            it->oie_dirent->oied_namelen,
6877                            it->oie_dirent->oied_type, attr);
6878
6879         if (rc < 0)
6880                 RETURN(rc);
6881
6882         if (osd_remote_fid(env, dev, fid))
6883                 RETURN(0);
6884
6885         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6886                 osd_add_oi_cache(oti, dev, id, fid);
6887
6888         RETURN(rc > 0 ? 0 : rc);
6889 }
6890
6891 /**
6892  * Returns the record size size at current position.
6893  *
6894  * This function will return record(lu_dirent) size in bytes.
6895  *
6896  * \param[in] env       execution environment
6897  * \param[in] di        iterator's in memory structure
6898  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6899  *                      calculate the lu_dirent size.
6900  *
6901  * \retval      record size(in bytes & in memory) of the current lu_dirent
6902  *              entry.
6903  */
6904 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6905                               __u32 attr)
6906 {
6907         struct osd_it_ea *it = (struct osd_it_ea *)di;
6908
6909         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6910 }
6911
6912 /**
6913  * Returns a cookie for current position of the iterator head, so that
6914  * user can use this cookie to load/start the iterator next time.
6915  *
6916  * \param di iterator's in memory structure
6917  *
6918  * \retval cookie for current position, on success
6919  */
6920 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6921 {
6922         struct osd_it_ea *it = (struct osd_it_ea *)di;
6923
6924         return it->oie_dirent->oied_off;
6925 }
6926
6927 /**
6928  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6929  * to load a directory entry at a time and stored it i inn,
6930  * in iterator's in-memory data structure.
6931  *
6932  * \param di struct osd_it_ea, iterator's in memory structure
6933  *
6934  * \retval +ve on success
6935  * \retval -ve on error
6936  */
6937 static int osd_it_ea_load(const struct lu_env *env,
6938                           const struct dt_it *di, __u64 hash)
6939 {
6940         struct osd_it_ea *it = (struct osd_it_ea *)di;
6941         int rc;
6942
6943         ENTRY;
6944         it->oie_file.f_pos = hash;
6945
6946         rc =  osd_ldiskfs_it_fill(env, di);
6947         if (rc > 0)
6948                 rc = -ENODATA;
6949
6950         if (rc == 0)
6951                 rc = +1;
6952
6953         RETURN(rc);
6954 }
6955
6956 /**
6957  * Index lookup function for interoperability mode (b11826).
6958  *
6959  * \param key,  key i.e. file name to be searched
6960  *
6961  * \retval +ve, on success
6962  * \retval -ve, on error
6963  */
6964 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6965                                struct dt_rec *rec, const struct dt_key *key)
6966 {
6967         struct osd_object *obj = osd_dt_obj(dt);
6968         int rc = 0;
6969
6970         ENTRY;
6971
6972         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6973         LINVRNT(osd_invariant(obj));
6974
6975         rc = osd_ea_lookup_rec(env, obj, rec, key);
6976         if (rc == 0)
6977                 rc = +1;
6978         RETURN(rc);
6979 }
6980
6981 /**
6982  * Index and Iterator operations for interoperability
6983  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6984  */
6985 static const struct dt_index_operations osd_index_ea_ops = {
6986         .dio_lookup         = osd_index_ea_lookup,
6987         .dio_declare_insert = osd_index_declare_ea_insert,
6988         .dio_insert         = osd_index_ea_insert,
6989         .dio_declare_delete = osd_index_declare_ea_delete,
6990         .dio_delete         = osd_index_ea_delete,
6991         .dio_it     = {
6992                 .init     = osd_it_ea_init,
6993                 .fini     = osd_it_ea_fini,
6994                 .get      = osd_it_ea_get,
6995                 .put      = osd_it_ea_put,
6996                 .next     = osd_it_ea_next,
6997                 .key      = osd_it_ea_key,
6998                 .key_size = osd_it_ea_key_size,
6999                 .rec      = osd_it_ea_rec,
7000                 .rec_size = osd_it_ea_rec_size,
7001                 .store    = osd_it_ea_store,
7002                 .load     = osd_it_ea_load
7003         }
7004 };
7005
7006 static void *osd_key_init(const struct lu_context *ctx,
7007                           struct lu_context_key *key)
7008 {
7009         struct osd_thread_info *info;
7010
7011         OBD_ALLOC_PTR(info);
7012         if (info == NULL)
7013                 return ERR_PTR(-ENOMEM);
7014
7015         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7016         if (info->oti_it_ea_buf == NULL)
7017                 goto out_free_info;
7018
7019         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7020
7021         info->oti_hlock = ldiskfs_htree_lock_alloc();
7022         if (info->oti_hlock == NULL)
7023                 goto out_free_ea;
7024
7025         return info;
7026
7027  out_free_ea:
7028         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7029  out_free_info:
7030         OBD_FREE_PTR(info);
7031         return ERR_PTR(-ENOMEM);
7032 }
7033
7034 static void osd_key_fini(const struct lu_context *ctx,
7035                          struct lu_context_key *key, void* data)
7036 {
7037         struct osd_thread_info *info = data;
7038         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7039         struct osd_idmap_cache  *idc = info->oti_ins_cache;
7040
7041         if (info->oti_inode != NULL)
7042                 OBD_FREE_PTR(lli);
7043         if (info->oti_hlock != NULL)
7044                 ldiskfs_htree_lock_free(info->oti_hlock);
7045         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7046         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7047         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7048         lu_buf_free(&info->oti_big_buf);
7049         if (idc != NULL) {
7050                 LASSERT(info->oti_ins_cache_size > 0);
7051                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7052                 info->oti_ins_cache = NULL;
7053                 info->oti_ins_cache_size = 0;
7054         }
7055         OBD_FREE_PTR(info);
7056 }
7057
7058 static void osd_key_exit(const struct lu_context *ctx,
7059                          struct lu_context_key *key, void *data)
7060 {
7061         struct osd_thread_info *info = data;
7062
7063         LASSERT(info->oti_r_locks == 0);
7064         LASSERT(info->oti_w_locks == 0);
7065         LASSERT(info->oti_txns    == 0);
7066 }
7067
7068 /* type constructor/destructor: osd_type_init, osd_type_fini */
7069 LU_TYPE_INIT_FINI(osd, &osd_key);
7070
7071 struct lu_context_key osd_key = {
7072         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7073         .lct_init = osd_key_init,
7074         .lct_fini = osd_key_fini,
7075         .lct_exit = osd_key_exit
7076 };
7077
7078
7079 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7080                            const char *name, struct lu_device *next)
7081 {
7082         struct osd_device *osd = osd_dev(d);
7083
7084         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
7085             >= sizeof(osd->od_svname))
7086                 return -E2BIG;
7087         return osd_procfs_init(osd, name);
7088 }
7089
7090 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7091 {
7092         struct seq_server_site  *ss = osd_seq_site(osd);
7093         int                     rc;
7094         ENTRY;
7095
7096         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7097                 RETURN(0);
7098
7099         if (unlikely(ss == NULL))
7100                 RETURN(-ENODEV);
7101
7102         OBD_ALLOC_PTR(osd->od_cl_seq);
7103         if (osd->od_cl_seq == NULL)
7104                 RETURN(-ENOMEM);
7105
7106         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7107                              osd->od_svname, ss->ss_server_seq);
7108         if (rc != 0) {
7109                 OBD_FREE_PTR(osd->od_cl_seq);
7110                 osd->od_cl_seq = NULL;
7111                 RETURN(rc);
7112         }
7113
7114         if (ss->ss_node_id == 0) {
7115                 /* If the OSD on the sequence controller(MDT0), then allocate
7116                  * sequence here, otherwise allocate sequence after connected
7117                  * to MDT0 (see mdt_register_lwp_callback()). */
7118                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7119                                    &osd->od_cl_seq->lcs_space, env);
7120         }
7121
7122         RETURN(rc);
7123 }
7124
7125 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7126 {
7127         if (osd->od_cl_seq == NULL)
7128                 return;
7129
7130         seq_client_fini(osd->od_cl_seq);
7131         OBD_FREE_PTR(osd->od_cl_seq);
7132         osd->od_cl_seq = NULL;
7133 }
7134
7135 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7136 {
7137         ENTRY;
7138
7139         /* shutdown quota slave instance associated with the device */
7140         if (o->od_quota_slave != NULL) {
7141                 struct qsd_instance *qsd = o->od_quota_slave;
7142
7143                 o->od_quota_slave = NULL;
7144                 qsd_fini(env, qsd);
7145         }
7146
7147         osd_fid_fini(env, o);
7148         osd_scrub_cleanup(env, o);
7149
7150         RETURN(0);
7151 }
7152
7153 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7154 {
7155         ENTRY;
7156
7157         if (o->od_mnt != NULL) {
7158                 shrink_dcache_sb(osd_sb(o));
7159                 osd_sync(env, &o->od_dt_dev);
7160
7161                 mntput(o->od_mnt);
7162                 o->od_mnt = NULL;
7163         }
7164
7165         EXIT;
7166 }
7167
7168 static int osd_mount(const struct lu_env *env,
7169                      struct osd_device *o, struct lustre_cfg *cfg)
7170 {
7171         const char              *name  = lustre_cfg_string(cfg, 0);
7172         const char              *dev  = lustre_cfg_string(cfg, 1);
7173         const char              *opts;
7174         unsigned long            page, s_flags, lmd_flags = 0;
7175         struct page             *__page;
7176         struct file_system_type *type;
7177         char                    *options = NULL;
7178         char                    *str;
7179         struct osd_thread_info  *info = osd_oti_get(env);
7180         struct lu_fid           *fid = &info->oti_fid;
7181         struct inode            *inode;
7182         int                      rc = 0, force_over_512tb = 0;
7183         ENTRY;
7184
7185         if (o->od_mnt != NULL)
7186                 RETURN(0);
7187
7188         if (strlen(dev) >= sizeof(o->od_mntdev))
7189                 RETURN(-E2BIG);
7190         strcpy(o->od_mntdev, dev);
7191
7192         str = lustre_cfg_string(cfg, 2);
7193         s_flags = simple_strtoul(str, NULL, 0);
7194         str = strstr(str, ":");
7195         if (str)
7196                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7197         opts = lustre_cfg_string(cfg, 3);
7198 #ifdef __BIG_ENDIAN
7199         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7200                 CERROR("%s: device %s extents feature is not guaranteed to "
7201                        "work on big-endian systems. Use \"bigendian_extents\" "
7202                        "mount option to override.\n", name, dev);
7203                 RETURN(-EINVAL);
7204         }
7205 #endif
7206 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7207         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7208                 CWARN("force_over_128tb option is deprecated. "
7209                       "Filesystems less than 512TB can be created without any "
7210                       "force options. Use force_over_512tb option for "
7211                       "filesystems greater than 512TB.\n");
7212         }
7213 #endif
7214 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7215         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7216                 CWARN("force_over_256tb option is deprecated. "
7217                       "Filesystems less than 512TB can be created without any "
7218                       "force options. Use force_over_512tb option for "
7219                       "filesystems greater than 512TB.\n");
7220         }
7221 #endif
7222
7223         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7224                 force_over_512tb = 1;
7225
7226         __page = alloc_page(GFP_KERNEL);
7227         if (__page == NULL)
7228                 GOTO(out, rc = -ENOMEM);
7229         page = (unsigned long)page_address(__page);
7230         options = (char *)page;
7231         *options = '\0';
7232         if (opts != NULL) {
7233                 /* strip out the options for back compatiblity */
7234                 static char *sout[] = {
7235                         "mballoc",
7236                         "iopen",
7237                         "noiopen",
7238                         "iopen_nopriv",
7239                         "extents",
7240                         "noextents",
7241                         /* strip out option we processed in osd */
7242                         "bigendian_extents",
7243                         "force_over_128tb",
7244                         "force_over_256tb",
7245                         "force_over_512tb",
7246                         NULL
7247                 };
7248                 strcat(options, opts);
7249                 for (rc = 0, str = options; sout[rc]; ) {
7250                         char *op = strstr(str, sout[rc]);
7251                         if (op == NULL) {
7252                                 rc++;
7253                                 str = options;
7254                                 continue;
7255                         }
7256                         if (op == options || *(op - 1) == ',') {
7257                                 str = op + strlen(sout[rc]);
7258                                 if (*str == ',' || *str == '\0') {
7259                                         *str == ',' ? str++ : str;
7260                                         memmove(op, str, strlen(str) + 1);
7261                                 }
7262                         }
7263                         for (str = op; *str != ',' && *str != '\0'; str++)
7264                                 ;
7265                 }
7266         } else {
7267                 strncat(options, "user_xattr,acl", 14);
7268         }
7269
7270         /* Glom up mount options */
7271         if (*options != '\0')
7272                 strcat(options, ",");
7273         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7274
7275         type = get_fs_type("ldiskfs");
7276         if (!type) {
7277                 CERROR("%s: cannot find ldiskfs module\n", name);
7278                 GOTO(out, rc = -ENODEV);
7279         }
7280
7281         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7282         module_put(type->owner);
7283
7284         if (IS_ERR(o->od_mnt)) {
7285                 rc = PTR_ERR(o->od_mnt);
7286                 o->od_mnt = NULL;
7287                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7288                 GOTO(out, rc);
7289         }
7290
7291         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7292                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7293                                  force_over_512tb == 0) {
7294                 CERROR("%s: device %s LDISKFS does not support filesystems "
7295                        "greater than 512TB and can cause data corruption. "
7296                        "Use \"force_over_512tb\" mount option to override.\n",
7297                        name, dev);
7298                 GOTO(out_mnt, rc = -EINVAL);
7299         }
7300
7301         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7302                 if (priv_dev_set_rdonly) {
7303                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7304                         o->od_dt_dev.dd_rdonly = 1;
7305                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7306                                       name);
7307                 } else {
7308                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7309                                       name);
7310
7311                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7312                 }
7313         } else if (priv_dev_check_rdonly &&
7314                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7315                 CERROR("%s: underlying device %s is marked as "
7316                        "read-only. Setup failed\n", name, dev);
7317
7318                 GOTO(out_mnt, rc = -EROFS);
7319         }
7320
7321         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7322                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7323                 GOTO(out_mnt, rc = -EINVAL);
7324         }
7325
7326 #ifdef LDISKFS_MOUNT_DIRDATA
7327         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7328                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7329         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7330                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7331                       "enabling the dirdata feature. If you do not want to "
7332                       "downgrade to Lustre-1.x again, you can enable it via "
7333                       "'tune2fs -O dirdata device'\n", name, dev);
7334 #endif
7335         inode = osd_sb(o)->s_root->d_inode;
7336         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7337         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7338         if (rc != 0) {
7339                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7340                 GOTO(out_mnt, rc);
7341         }
7342
7343         if (lmd_flags & LMD_FLG_NOSCRUB)
7344                 o->od_auto_scrub_interval = AS_NEVER;
7345
7346         GOTO(out, rc = 0);
7347
7348 out_mnt:
7349         mntput(o->od_mnt);
7350         o->od_mnt = NULL;
7351
7352 out:
7353         if (__page)
7354                 __free_page(__page);
7355
7356         return rc;
7357 }
7358
7359 static struct lu_device *osd_device_fini(const struct lu_env *env,
7360                                          struct lu_device *d)
7361 {
7362         struct osd_device *o = osd_dev(d);
7363         ENTRY;
7364
7365         osd_index_backup(env, o, false);
7366         osd_shutdown(env, o);
7367         osd_procfs_fini(o);
7368         osd_obj_map_fini(o);
7369         osd_umount(env, o);
7370
7371         RETURN(NULL);
7372 }
7373
7374 static int osd_device_init0(const struct lu_env *env,
7375                             struct osd_device *o,
7376                             struct lustre_cfg *cfg)
7377 {
7378         struct lu_device        *l = osd2lu_dev(o);
7379         struct osd_thread_info *info;
7380         int                     rc;
7381         int                     cplen = 0;
7382
7383         /* if the module was re-loaded, env can loose its keys */
7384         rc = lu_env_refill((struct lu_env *) env);
7385         if (rc)
7386                 GOTO(out, rc);
7387         info = osd_oti_get(env);
7388         LASSERT(info);
7389
7390         l->ld_ops = &osd_lu_ops;
7391         o->od_dt_dev.dd_ops = &osd_dt_ops;
7392
7393         spin_lock_init(&o->od_osfs_lock);
7394         mutex_init(&o->od_otable_mutex);
7395         INIT_LIST_HEAD(&o->od_orphan_list);
7396         INIT_LIST_HEAD(&o->od_index_backup_list);
7397         INIT_LIST_HEAD(&o->od_index_restore_list);
7398         spin_lock_init(&o->od_lock);
7399         o->od_index_backup_policy = LIBP_NONE;
7400
7401         o->od_read_cache = 1;
7402         o->od_writethrough_cache = 1;
7403         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7404         o->od_auto_scrub_interval = AS_DEFAULT;
7405
7406         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7407                         sizeof(o->od_svname));
7408         if (cplen >= sizeof(o->od_svname)) {
7409                 rc = -E2BIG;
7410                 GOTO(out, rc);
7411         }
7412
7413         o->od_index_backup_stop = 0;
7414         o->od_index = -1; /* -1 means index is invalid */
7415         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7416         if (rc == LDD_F_SV_TYPE_OST)
7417                 o->od_is_ost = 1;
7418
7419         o->od_full_scrub_ratio = OFSR_DEFAULT;
7420         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7421         rc = osd_mount(env, o, cfg);
7422         if (rc != 0)
7423                 GOTO(out, rc);
7424
7425         rc = osd_obj_map_init(env, o);
7426         if (rc != 0)
7427                 GOTO(out_mnt, rc);
7428
7429         rc = lu_site_init(&o->od_site, l);
7430         if (rc != 0)
7431                 GOTO(out_compat, rc);
7432         o->od_site.ls_bottom_dev = l;
7433
7434         rc = lu_site_init_finish(&o->od_site);
7435         if (rc != 0)
7436                 GOTO(out_site, rc);
7437
7438         INIT_LIST_HEAD(&o->od_ios_list);
7439         /* setup scrub, including OI files initialization */
7440         o->od_in_init = 1;
7441         rc = osd_scrub_setup(env, o);
7442         o->od_in_init = 0;
7443         if (rc < 0)
7444                 GOTO(out_site, rc);
7445
7446         rc = osd_procfs_init(o, o->od_svname);
7447         if (rc != 0) {
7448                 CERROR("%s: can't initialize procfs: rc = %d\n",
7449                        o->od_svname, rc);
7450                 GOTO(out_scrub, rc);
7451         }
7452
7453         LASSERT(l->ld_site->ls_linkage.next != NULL);
7454         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7455
7456         /* initialize quota slave instance */
7457         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7458                                      o->od_proc_entry);
7459         if (IS_ERR(o->od_quota_slave)) {
7460                 rc = PTR_ERR(o->od_quota_slave);
7461                 o->od_quota_slave = NULL;
7462                 GOTO(out_procfs, rc);
7463         }
7464
7465         RETURN(0);
7466
7467 out_procfs:
7468         osd_procfs_fini(o);
7469 out_scrub:
7470         osd_scrub_cleanup(env, o);
7471 out_site:
7472         lu_site_fini(&o->od_site);
7473 out_compat:
7474         osd_obj_map_fini(o);
7475 out_mnt:
7476         osd_umount(env, o);
7477 out:
7478         return rc;
7479 }
7480
7481 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7482                                           struct lu_device_type *t,
7483                                           struct lustre_cfg *cfg)
7484 {
7485         struct osd_device *o;
7486         int                rc;
7487
7488         OBD_ALLOC_PTR(o);
7489         if (o == NULL)
7490                 return ERR_PTR(-ENOMEM);
7491
7492         rc = dt_device_init(&o->od_dt_dev, t);
7493         if (rc == 0) {
7494                 /* Because the ctx might be revived in dt_device_init,
7495                  * refill the env here */
7496                 lu_env_refill((struct lu_env *)env);
7497                 rc = osd_device_init0(env, o, cfg);
7498                 if (rc)
7499                         dt_device_fini(&o->od_dt_dev);
7500         }
7501
7502         if (unlikely(rc != 0))
7503                 OBD_FREE_PTR(o);
7504
7505         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7506 }
7507
7508 static struct lu_device *osd_device_free(const struct lu_env *env,
7509                                          struct lu_device *d)
7510 {
7511         struct osd_device *o = osd_dev(d);
7512         ENTRY;
7513
7514         /* XXX: make osd top device in order to release reference */
7515         d->ld_site->ls_top_dev = d;
7516         lu_site_purge(env, d->ld_site, -1);
7517         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7518                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7519                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7520         }
7521         lu_site_fini(&o->od_site);
7522         dt_device_fini(&o->od_dt_dev);
7523         OBD_FREE_PTR(o);
7524         RETURN(NULL);
7525 }
7526
7527 static int osd_process_config(const struct lu_env *env,
7528                               struct lu_device *d, struct lustre_cfg *cfg)
7529 {
7530         struct osd_device               *o = osd_dev(d);
7531         int                             rc;
7532         ENTRY;
7533
7534         switch (cfg->lcfg_command) {
7535         case LCFG_SETUP:
7536                 rc = osd_mount(env, o, cfg);
7537                 break;
7538         case LCFG_CLEANUP:
7539                 /* For the case LCFG_PRE_CLEANUP is not called in advance,
7540                  * that may happend if hit failure during mount process. */
7541                 osd_index_backup(env, o, false);
7542                 lu_dev_del_linkage(d->ld_site, d);
7543                 rc = osd_shutdown(env, o);
7544                 break;
7545         case LCFG_PARAM:
7546                 LASSERT(&o->od_dt_dev);
7547                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7548                                               cfg, &o->od_dt_dev);
7549                 if (rc > 0 || rc == -ENOSYS) {
7550                         rc = class_process_proc_param(PARAM_OST,
7551                                                       lprocfs_osd_obd_vars,
7552                                                       cfg, &o->od_dt_dev);
7553                         if (rc > 0)
7554                                 rc = 0;
7555                 }
7556                 break;
7557         case LCFG_PRE_CLEANUP:
7558                 osd_scrub_stop(o);
7559                 osd_index_backup(env, o,
7560                                  o->od_index_backup_policy != LIBP_NONE);
7561                 rc = 0;
7562                 break;
7563         default:
7564                 rc = -ENOSYS;
7565         }
7566
7567         RETURN(rc);
7568 }
7569
7570 static int osd_recovery_complete(const struct lu_env *env,
7571                                  struct lu_device *d)
7572 {
7573         struct osd_device       *osd = osd_dev(d);
7574         int                      rc = 0;
7575         ENTRY;
7576
7577         if (osd->od_quota_slave == NULL)
7578                 RETURN(0);
7579
7580         /* start qsd instance on recovery completion, this notifies the quota
7581          * slave code that we are about to process new requests now */
7582         rc = qsd_start(env, osd->od_quota_slave);
7583         RETURN(rc);
7584 }
7585
7586 /*
7587  * we use exports to track all osd users
7588  */
7589 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7590                            struct obd_device *obd, struct obd_uuid *cluuid,
7591                            struct obd_connect_data *data, void *localdata)
7592 {
7593         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7594         struct lustre_handle  conn;
7595         int                   rc;
7596         ENTRY;
7597
7598         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7599
7600         rc = class_connect(&conn, obd, cluuid);
7601         if (rc)
7602                 RETURN(rc);
7603
7604         *exp = class_conn2export(&conn);
7605
7606         spin_lock(&osd->od_osfs_lock);
7607         osd->od_connects++;
7608         spin_unlock(&osd->od_osfs_lock);
7609
7610         RETURN(0);
7611 }
7612
7613 /*
7614  * once last export (we don't count self-export) disappeared
7615  * osd can be released
7616  */
7617 static int osd_obd_disconnect(struct obd_export *exp)
7618 {
7619         struct obd_device *obd = exp->exp_obd;
7620         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7621         int                rc, release = 0;
7622         ENTRY;
7623
7624         /* Only disconnect the underlying layers on the final disconnect. */
7625         spin_lock(&osd->od_osfs_lock);
7626         osd->od_connects--;
7627         if (osd->od_connects == 0)
7628                 release = 1;
7629         spin_unlock(&osd->od_osfs_lock);
7630
7631         rc = class_disconnect(exp); /* bz 9811 */
7632
7633         if (rc == 0 && release)
7634                 class_manual_cleanup(obd);
7635         RETURN(rc);
7636 }
7637
7638 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7639                        struct lu_device *dev)
7640 {
7641         struct osd_device       *osd    = osd_dev(dev);
7642         struct lr_server_data   *lsd    =
7643                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7644         int                      result = 0;
7645         ENTRY;
7646
7647         if (osd->od_quota_slave != NULL) {
7648                 /* set up quota slave objects */
7649                 result = qsd_prepare(env, osd->od_quota_slave);
7650                 if (result != 0)
7651                         RETURN(result);
7652         }
7653
7654         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7655 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7656                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7657                         osd->od_index_in_idif = 1;
7658                 } else {
7659                         osd->od_index_in_idif = 0;
7660                         result = osd_register_proc_index_in_idif(osd);
7661                         if (result != 0)
7662                                 RETURN(result);
7663                 }
7664 #else
7665                 osd->od_index_in_idif = 1;
7666 #endif
7667         }
7668
7669         result = osd_fid_init(env, osd);
7670
7671         RETURN(result);
7672 }
7673
7674 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7675                          struct lu_fid *fid, struct md_op_data *op_data)
7676 {
7677         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7678
7679         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7680 }
7681
7682 static const struct lu_object_operations osd_lu_obj_ops = {
7683         .loo_object_init      = osd_object_init,
7684         .loo_object_delete    = osd_object_delete,
7685         .loo_object_release   = osd_object_release,
7686         .loo_object_free      = osd_object_free,
7687         .loo_object_print     = osd_object_print,
7688         .loo_object_invariant = osd_object_invariant
7689 };
7690
7691 const struct lu_device_operations osd_lu_ops = {
7692         .ldo_object_alloc      = osd_object_alloc,
7693         .ldo_process_config    = osd_process_config,
7694         .ldo_recovery_complete = osd_recovery_complete,
7695         .ldo_prepare           = osd_prepare,
7696 };
7697
7698 static const struct lu_device_type_operations osd_device_type_ops = {
7699         .ldto_init = osd_type_init,
7700         .ldto_fini = osd_type_fini,
7701
7702         .ldto_start = osd_type_start,
7703         .ldto_stop  = osd_type_stop,
7704
7705         .ldto_device_alloc = osd_device_alloc,
7706         .ldto_device_free  = osd_device_free,
7707
7708         .ldto_device_init    = osd_device_init,
7709         .ldto_device_fini    = osd_device_fini
7710 };
7711
7712 static struct lu_device_type osd_device_type = {
7713         .ldt_tags     = LU_DEVICE_DT,
7714         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7715         .ldt_ops      = &osd_device_type_ops,
7716         .ldt_ctx_tags = LCT_LOCAL,
7717 };
7718
7719 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7720 {
7721         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7722         struct super_block *sb = osd_sb(osd);
7723
7724         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7725 }
7726
7727 /*
7728  * lprocfs legacy support.
7729  */
7730 static struct obd_ops osd_obd_device_ops = {
7731         .o_owner = THIS_MODULE,
7732         .o_connect      = osd_obd_connect,
7733         .o_disconnect   = osd_obd_disconnect,
7734         .o_fid_alloc    = osd_fid_alloc,
7735         .o_health_check = osd_health_check,
7736 };
7737
7738 static int __init osd_init(void)
7739 {
7740         int rc;
7741
7742         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7743 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7744         /* please, try to keep osd_thread_info smaller than a page */
7745         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7746 #endif
7747
7748         osd_oi_mod_init();
7749
7750         rc = lu_kmem_init(ldiskfs_caches);
7751         if (rc)
7752                 return rc;
7753
7754 #ifdef CONFIG_KALLSYMS
7755         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7756         priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7757         /* Clear readonly is unused at this time */
7758         /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7759 #endif
7760
7761         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7762                                  lprocfs_osd_module_vars,
7763                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7764         if (rc)
7765                 lu_kmem_fini(ldiskfs_caches);
7766         return rc;
7767 }
7768
7769 static void __exit osd_exit(void)
7770 {
7771         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7772         lu_kmem_fini(ldiskfs_caches);
7773 }
7774
7775 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7776 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7777 MODULE_VERSION(LUSTRE_VERSION_STRING);
7778 MODULE_LICENSE("GPL");
7779
7780 module_init(osd_init);
7781 module_exit(osd_exit);