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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76 #include <lustre_linkea.h>
  77
  78 int ldiskfs_pdo = 1;
  79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  80                 "ldiskfs with parallel directory operations");
  81
  82 int ldiskfs_track_declares_assert;
  83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  84                 "LBUG during tracking of declares");
  85
  86 /* Slab to allocate dynlocks */
  87 struct kmem_cache *dynlock_cachep;
  88
  89 static struct lu_kmem_descr ldiskfs_caches[] = {
  90         {
  91                 .ckd_cache = &dynlock_cachep,
  92                 .ckd_name  = "dynlock_cache",
  93                 .ckd_size  = sizeof(struct dynlock_handle)
  94         },
  95         {
  96                 .ckd_cache = NULL
  97         }
  98 };
  99
 100 static const char dot[] = ".";
 101 static const char dotdot[] = "..";
 102 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 103
 104 static const struct lu_object_operations      osd_lu_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ops;
 106 static const struct dt_object_operations      osd_obj_ea_ops;
 107 static const struct dt_object_operations      osd_obj_otable_it_ops;
 108 static const struct dt_index_operations       osd_index_iam_ops;
 109 static const struct dt_index_operations       osd_index_ea_ops;
 110
 111 int osd_trans_declare_op2rb[] = {
 112         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 113         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 114         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 115         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 116         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 117         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 118         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 119         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 120         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 121         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 122         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 123         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 124 };
 125
 126 static int osd_has_index(const struct osd_object *obj)
 127 {
 128         return obj->oo_dt.do_index_ops != NULL;
 129 }
 130
 131 static int osd_object_invariant(const struct lu_object *l)
 132 {
 133         return osd_invariant(osd_obj(l));
 134 }
 135
 136 /*
 137  * Concurrency: doesn't matter
 138  */
 139
 140 /*
 141  * Concurrency: doesn't matter
 142  */
 143 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 144 {
 145         struct osd_thread_info *oti = osd_oti_get(env);
 146         return oti->oti_w_locks > 0 && o->oo_owner == env;
 147 }
 148
 149 /*
 150  * Concurrency: doesn't access mutable data
 151  */
 152 static int osd_root_get(const struct lu_env *env,
 153                         struct dt_device *dev, struct lu_fid *f)
 154 {
 155         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 156         return 0;
 157 }
 158
 159 /*
 160  * OSD object methods.
 161  */
 162
 163 /*
 164  * Concurrency: no concurrent access is possible that early in object
 165  * life-cycle.
 166  */
 167 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 168                                           const struct lu_object_header *hdr,
 169                                           struct lu_device *d)
 170 {
 171         struct osd_object *mo;
 172
 173         OBD_ALLOC_PTR(mo);
 174         if (mo != NULL) {
 175                 struct lu_object *l;
 176
 177                 l = &mo->oo_dt.do_lu;
 178                 dt_object_init(&mo->oo_dt, NULL, d);
 179                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 180                 l->lo_ops = &osd_lu_obj_ops;
 181                 init_rwsem(&mo->oo_sem);
 182                 init_rwsem(&mo->oo_ext_idx_sem);
 183                 spin_lock_init(&mo->oo_guard);
 184                 return l;
 185         } else {
 186                 return NULL;
 187         }
 188 }
 189
 190 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 191                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 192 {
 193         int rc;
 194
 195         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 196         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 197                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 198         if (rc > 0) {
 199                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 200                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 201                 rc = 0;
 202                 lustre_lma_swab(lma);
 203                 /* Check LMA compatibility */
 204                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 205                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 206                               "for fid = "DFID", ino = %lu\n",
 207                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 208                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 209                               PFID(&lma->lma_self_fid), inode->i_ino);
 210                         rc = -EOPNOTSUPP;
 211                 }
 212         } else if (rc == 0) {
 213                 rc = -ENODATA;
 214         }
 215
 216         return rc;
 217 }
 218
 219 /*
 220  * retrieve object from backend ext fs.
 221  **/
 222 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 223                        struct osd_inode_id *id)
 224 {
 225         struct inode *inode = NULL;
 226
 227         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 228         if (IS_ERR(inode)) {
 229                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 230                        id->oii_ino, PTR_ERR(inode));
 231         } else if (id->oii_gen != OSD_OII_NOGEN &&
 232                    inode->i_generation != id->oii_gen) {
 233                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 234                        "i_generation = %u\n",
 235                        id->oii_ino, id->oii_gen, inode->i_generation);
 236                 iput(inode);
 237                 inode = ERR_PTR(-ESTALE);
 238         } else if (inode->i_nlink == 0) {
 239                 /* due to parallel readdir and unlink,
 240                 * we can have dead inode here. */
 241                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 242                 iput(inode);
 243                 inode = ERR_PTR(-ESTALE);
 244         } else if (is_bad_inode(inode)) {
 245                 CWARN("%.16s: bad inode: ino = %u\n",
 246                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 247                 iput(inode);
 248                 inode = ERR_PTR(-ENOENT);
 249         } else {
 250                 if (id->oii_gen == OSD_OII_NOGEN)
 251                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 252
 253                 /* Do not update file c/mtime in ldiskfs.
 254                  * NB: we don't have any lock to protect this because we don't
 255                  * have reference on osd_object now, but contention with
 256                  * another lookup + attr_set can't happen in the tiny window
 257                  * between if (...) and set S_NOCMTIME. */
 258                 if (!(inode->i_flags & S_NOCMTIME))
 259                         inode->i_flags |= S_NOCMTIME;
 260         }
 261         return inode;
 262 }
 263
 264 static struct inode *
 265 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 266              struct osd_inode_id *id, struct lu_fid *fid)
 267 {
 268         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 269         struct inode            *inode;
 270         int                      rc;
 271
 272         inode = osd_iget(info, dev, id);
 273         if (IS_ERR(inode))
 274                 return inode;
 275
 276         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 277         if (rc == 0) {
 278                 *fid = lma->lma_self_fid;
 279         } else if (rc == -ENODATA) {
 280                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 281                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 282                 else
 283                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 284         } else {
 285                 iput(inode);
 286                 inode = ERR_PTR(rc);
 287         }
 288         return inode;
 289 }
 290
 291 static struct inode *osd_iget_check(struct osd_thread_info *info,
 292                                     struct osd_device *dev,
 293                                     const struct lu_fid *fid,
 294                                     struct osd_inode_id *id,
 295                                     bool in_oi)
 296 {
 297         struct inode    *inode;
 298         int              rc     = 0;
 299         ENTRY;
 300
 301         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 302         if (IS_ERR(inode)) {
 303                 rc = PTR_ERR(inode);
 304                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 305                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 306                                id->oii_ino, rc);
 307
 308                         GOTO(put, rc);
 309                 }
 310
 311                 goto check_oi;
 312         }
 313
 314         if (is_bad_inode(inode)) {
 315                 rc = -ENOENT;
 316                 if (!in_oi) {
 317                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 318
 319                         GOTO(put, rc);
 320                 }
 321
 322                 goto check_oi;
 323         }
 324
 325         if (id->oii_gen != OSD_OII_NOGEN &&
 326             inode->i_generation != id->oii_gen) {
 327                 rc = -ESTALE;
 328                 if (!in_oi) {
 329                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 330                                "oii_gen = %u, i_generation = %u\n",
 331                                id->oii_ino, id->oii_gen, inode->i_generation);
 332
 333                         GOTO(put, rc);
 334                 }
 335
 336                 goto check_oi;
 337         }
 338
 339         if (inode->i_nlink == 0) {
 340                 rc = -ENOENT;
 341                 if (!in_oi) {
 342                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 343
 344                         GOTO(put, rc);
 345                 }
 346
 347                 goto check_oi;
 348         }
 349
 350 check_oi:
 351         if (rc != 0) {
 352                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 353
 354                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 355                 /* XXX: There are some possible cases:
 356                  *      1. rc = 0.
 357                  *         Backup/restore caused the OI invalid.
 358                  *      2. rc = 0.
 359                  *         Someone unlinked the object but NOT removed
 360                  *         the OI mapping, such as mount target device
 361                  *         as ldiskfs, and modify something directly.
 362                  *      3. rc = -ENOENT.
 363                  *         Someone just removed the object between the
 364                  *         former oi_lookup and the iget. It is normal.
 365                  *      4. Other failure cases.
 366                  *
 367                  *      Generally, when the device is mounted, it will
 368                  *      auto check whether the system is restored from
 369                  *      file-level backup or not. We trust such detect
 370                  *      to distinguish the 1st case from the 2nd case. */
 371                 if (rc == 0) {
 372                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 373                             inode->i_generation == id->oii_gen)
 374                                 rc = -ENOENT;
 375                         else
 376                                 rc = -EREMCHG;
 377                 }
 378         } else {
 379                 if (id->oii_gen == OSD_OII_NOGEN)
 380                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 381
 382                 /* Do not update file c/mtime in ldiskfs.
 383                  * NB: we don't have any lock to protect this because we don't
 384                  * have reference on osd_object now, but contention with
 385                  * another lookup + attr_set can't happen in the tiny window
 386                  * between if (...) and set S_NOCMTIME. */
 387                 if (!(inode->i_flags & S_NOCMTIME))
 388                         inode->i_flags |= S_NOCMTIME;
 389         }
 390
 391         GOTO(put, rc);
 392
 393 put:
 394         if (rc != 0) {
 395                 if (!IS_ERR(inode))
 396                         iput(inode);
 397
 398                 inode = ERR_PTR(rc);
 399         }
 400
 401         return inode;
 402 }
 403
 404 /**
 405  * \retval +v: new filter_fid, does not contain self-fid
 406  * \retval 0:  filter_fid_old, contains self-fid
 407  * \retval -v: other failure cases
 408  */
 409 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 410                  struct dentry *dentry, struct lu_fid *fid)
 411 {
 412         struct filter_fid_old   *ff     = &info->oti_ff;
 413         struct ost_id           *ostid  = &info->oti_ostid;
 414         int                      rc;
 415
 416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 417         if (rc == sizeof(*ff)) {
 418                 rc = 0;
 419                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 420                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 421                 /* XXX: should use real OST index in the future. LU-3569 */
 422                 ostid_to_fid(fid, ostid, 0);
 423         } else if (rc == sizeof(struct filter_fid)) {
 424                 rc = 1;
 425         } else if (rc >= 0) {
 426                 rc = -EINVAL;
 427         }
 428
 429         return rc;
 430 }
 431
 432 static int osd_lma_self_repair(struct osd_thread_info *info,
 433                                struct osd_device *osd, struct inode *inode,
 434                                const struct lu_fid *fid, __u32 compat)
 435 {
 436         handle_t *jh;
 437         int       rc;
 438
 439         LASSERT(current->journal_info == NULL);
 440
 441         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 442                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 443         if (IS_ERR(jh)) {
 444                 rc = PTR_ERR(jh);
 445                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 446                       osd_name(osd), rc);
 447                 return rc;
 448         }
 449
 450         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 451         if (rc != 0)
 452                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 453                       osd_name(osd), rc);
 454         ldiskfs_journal_stop(jh);
 455         return rc;
 456 }
 457
 458 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 459 {
 460         struct osd_thread_info  *info   = osd_oti_get(env);
 461         struct osd_device       *osd    = osd_obj2dev(obj);
 462         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 463         struct inode            *inode  = obj->oo_inode;
 464         struct dentry           *dentry = &info->oti_obj_dentry;
 465         struct lu_fid           *fid    = NULL;
 466         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 467         int                      rc;
 468         ENTRY;
 469
 470         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 471         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 472                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 473         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 474                 fid = &lma->lma_self_fid;
 475                 rc = osd_get_idif(info, inode, dentry, fid);
 476                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 477                         /* For the given OST-object, if it has neither LMA nor
 478                          * FID in XATTR_NAME_FID, then the given FID (which is
 479                          * contained in the @obj, from client RPC for locating
 480                          * the OST-object) is trusted. We use it to generate
 481                          * the LMA. */
 482                         osd_lma_self_repair(info, osd, inode, rfid,
 483                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 484                                 LMAC_FID_ON_OST : 0);
 485                         RETURN(0);
 486                 }
 487         }
 488
 489         if (unlikely(rc == -ENODATA))
 490                 RETURN(0);
 491
 492         if (rc < 0)
 493                 RETURN(rc);
 494
 495         if (rc > 0) {
 496                 rc = 0;
 497                 lustre_lma_swab(lma);
 498                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 499                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 500                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 501                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 502                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 503                               PFID(rfid), inode->i_ino);
 504                         rc = -EOPNOTSUPP;
 505                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 506                         fid = &lma->lma_self_fid;
 507                 }
 508         }
 509
 510         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 511                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 512                         struct ost_id   *oi   = &info->oti_ostid;
 513                         struct lu_fid   *fid1 = &info->oti_fid3;
 514                         __u32            idx  = fid_idif_ost_idx(rfid);
 515
 516                         /* For old IDIF, the OST index is not part of the IDIF,
 517                          * Means that different OSTs may have the same IDIFs.
 518                          * Under such case, we need to make some compatible
 519                          * check to make sure to trigger OI scrub properly. */
 520                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 521                                 /* Given @rfid is new, LMA is old. */
 522                                 fid_to_ostid(fid, oi);
 523                                 ostid_to_fid(fid1, oi, idx);
 524                                 if (lu_fid_eq(fid1, rfid)) {
 525                                         if (osd->od_lma_self_repair)
 526                                                 osd_lma_self_repair(info, osd,
 527                                                         inode, rfid,
 528                                                         LMAC_FID_ON_OST);
 529                                         RETURN(0);
 530                                 }
 531                         }
 532                 }
 533
 534                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 535                        osd_name(osd), PFID(rfid), PFID(fid));
 536                 rc = -EREMCHG;
 537         }
 538
 539         RETURN(rc);
 540 }
 541
 542 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 543                           const struct lu_fid *fid,
 544                           const struct lu_object_conf *conf)
 545 {
 546         struct osd_thread_info *info;
 547         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 548         struct osd_device      *dev;
 549         struct osd_idmap_cache *oic;
 550         struct osd_inode_id    *id;
 551         struct inode           *inode;
 552         struct osd_scrub       *scrub;
 553         struct scrub_file      *sf;
 554         int                     result;
 555         int                     saved  = 0;
 556         bool                    in_oi  = false;
 557         bool                    triggered = false;
 558         ENTRY;
 559
 560         LINVRNT(osd_invariant(obj));
 561         LASSERT(obj->oo_inode == NULL);
 562         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 563
 564         dev = osd_dev(ldev);
 565         scrub = &dev->od_scrub;
 566         sf = &scrub->os_file;
 567         info = osd_oti_get(env);
 568         LASSERT(info);
 569         oic = &info->oti_cache;
 570
 571         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 572                 RETURN(-ENOENT);
 573
 574         /* For the object is created as locking anchor, or for the object to
 575          * be created on disk. No need to osd_oi_lookup() at here because FID
 576          * shouldn't never be re-used, if it's really a duplicate FID from
 577          * unexpected reason, we should be able to detect it later by calling
 578          * do_create->osd_oi_insert(). */
 579         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 580                 GOTO(out, result = 0);
 581
 582         /* Search order: 1. per-thread cache. */
 583         if (lu_fid_eq(fid, &oic->oic_fid) &&
 584             likely(oic->oic_dev == dev)) {
 585                 id = &oic->oic_lid;
 586                 goto iget;
 587         }
 588
 589         id = &info->oti_id;
 590         if (!list_empty(&scrub->os_inconsistent_items)) {
 591                 /* Search order: 2. OI scrub pending list. */
 592                 result = osd_oii_lookup(dev, fid, id);
 593                 if (result == 0)
 594                         goto iget;
 595         }
 596
 597         /* Search order: 3. OI files. */
 598         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 599         if (result == -ENOENT) {
 600                 if (!fid_is_norm(fid) ||
 601                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 602                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 603                                       sf->sf_oi_bitmap))
 604                         GOTO(out, result = 0);
 605
 606                 goto trigger;
 607         }
 608
 609         if (result != 0)
 610                 GOTO(out, result);
 611
 612         in_oi = true;
 613
 614 iget:
 615         inode = osd_iget_check(info, dev, fid, id, in_oi);
 616         if (IS_ERR(inode)) {
 617                 result = PTR_ERR(inode);
 618                 if (result == -ENOENT || result == -ESTALE) {
 619                         if (!in_oi)
 620                                 fid_zero(&oic->oic_fid);
 621
 622                         GOTO(out, result = -ENOENT);
 623                 } else if (result == -EREMCHG) {
 624
 625 trigger:
 626                         if (!in_oi)
 627                                 fid_zero(&oic->oic_fid);
 628
 629                         if (unlikely(triggered))
 630                                 GOTO(out, result = saved);
 631
 632                         triggered = true;
 633                         if (thread_is_running(&scrub->os_thread)) {
 634                                 result = -EINPROGRESS;
 635                         } else if (!dev->od_noscrub) {
 636                                 /* Since we do not know the right OI mapping,
 637                                  * we have to trigger OI scrub to scan the
 638                                  * whole device. */
 639                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
 640                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 641                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 642                                               "for "DFID", rc = %d [1]\n",
 643                                               osd_name(dev), PFID(fid),result);
 644                                 if (result == 0 || result == -EALREADY)
 645                                         result = -EINPROGRESS;
 646                                 else
 647                                         result = -EREMCHG;
 648                         }
 649
 650                         /* We still have chance to get the valid inode: for the
 651                          * object which is referenced by remote name entry, the
 652                          * object on the local MDT will be linked under the dir
 653                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 654                          *
 655                          * We do not know whether the object for the given FID
 656                          * is referenced by some remote name entry or not, and
 657                          * especially for DNE II, a multiple-linked object may
 658                          * have many name entries reside on many MDTs.
 659                          *
 660                          * To simplify the operation, OSD will not distinguish
 661                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 662                          * only happened for the RPC from other MDT during the
 663                          * OI scrub, or for the client side RPC with FID only,
 664                          * such as FID to path, or from old connected client. */
 665                         saved = result;
 666                         result = osd_lookup_in_remote_parent(info, dev,
 667                                                              fid, id);
 668                         if (result == 0) {
 669                                 in_oi = false;
 670                                 goto iget;
 671                         }
 672
 673                         result = saved;
 674                 }
 675
 676                 GOTO(out, result);
 677         }
 678
 679         obj->oo_inode = inode;
 680         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 681
 682         result = osd_check_lma(env, obj);
 683         if (result != 0) {
 684                 iput(inode);
 685                 obj->oo_inode = NULL;
 686                 if (result == -EREMCHG) {
 687                         if (!in_oi) {
 688                                 result = osd_oi_lookup(info, dev, fid, id,
 689                                                        OI_CHECK_FLD);
 690                                 if (result != 0) {
 691                                         fid_zero(&oic->oic_fid);
 692                                         GOTO(out, result);
 693                                 }
 694                         }
 695
 696                         goto trigger;
 697                 }
 698
 699                 GOTO(out, result);
 700         }
 701
 702         obj->oo_compat_dot_created = 1;
 703         obj->oo_compat_dotdot_created = 1;
 704
 705         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 706                 GOTO(out, result = 0);
 707
 708         LASSERT(obj->oo_hl_head == NULL);
 709         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 710         if (obj->oo_hl_head == NULL) {
 711                 obj->oo_inode = NULL;
 712                 iput(inode);
 713                 GOTO(out, result = -ENOMEM);
 714         }
 715         GOTO(out, result = 0);
 716
 717 out:
 718         LINVRNT(osd_invariant(obj));
 719         return result;
 720 }
 721
 722 /*
 723  * Concurrency: shouldn't matter.
 724  */
 725 static void osd_object_init0(struct osd_object *obj)
 726 {
 727         LASSERT(obj->oo_inode != NULL);
 728         obj->oo_dt.do_body_ops = &osd_body_ops;
 729         obj->oo_dt.do_lu.lo_header->loh_attr |=
 730                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 731 }
 732
 733 /*
 734  * Concurrency: no concurrent access is possible that early in object
 735  * life-cycle.
 736  */
 737 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 738                            const struct lu_object_conf *conf)
 739 {
 740         struct osd_object *obj = osd_obj(l);
 741         int result;
 742
 743         LINVRNT(osd_invariant(obj));
 744
 745         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 746                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 747                 l->lo_header->loh_attr |= LOHA_EXISTS;
 748                 return 0;
 749         }
 750
 751         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 752         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 753         if (result == 0 && obj->oo_inode != NULL)
 754                 osd_object_init0(obj);
 755
 756         LINVRNT(osd_invariant(obj));
 757         return result;
 758 }
 759
 760 /*
 761  * Concurrency: no concurrent access is possible that late in object
 762  * life-cycle.
 763  */
 764 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 765 {
 766         struct osd_object *obj = osd_obj(l);
 767
 768         LINVRNT(osd_invariant(obj));
 769
 770         dt_object_fini(&obj->oo_dt);
 771         if (obj->oo_hl_head != NULL)
 772                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 773         OBD_FREE_PTR(obj);
 774 }
 775
 776 /*
 777  * Concurrency: no concurrent access is possible that late in object
 778  * life-cycle.
 779  */
 780 static void osd_index_fini(struct osd_object *o)
 781 {
 782         struct iam_container *bag;
 783
 784         if (o->oo_dir != NULL) {
 785                 bag = &o->oo_dir->od_container;
 786                 if (o->oo_inode != NULL) {
 787                         if (bag->ic_object == o->oo_inode)
 788                                 iam_container_fini(bag);
 789                 }
 790                 OBD_FREE_PTR(o->oo_dir);
 791                 o->oo_dir = NULL;
 792         }
 793 }
 794
 795 /*
 796  * Concurrency: no concurrent access is possible that late in object
 797  * life-cycle (for all existing callers, that is. New callers have to provide
 798  * their own locking.)
 799  */
 800 static int osd_inode_unlinked(const struct inode *inode)
 801 {
 802         return inode->i_nlink == 0;
 803 }
 804
 805 enum {
 806         OSD_TXN_OI_DELETE_CREDITS    = 20,
 807         OSD_TXN_INODE_DELETE_CREDITS = 20
 808 };
 809
 810 /*
 811  * Journal
 812  */
 813
 814 #if OSD_THANDLE_STATS
 815 /**
 816  * Set time when the handle is allocated
 817  */
 818 static void osd_th_alloced(struct osd_thandle *oth)
 819 {
 820         oth->oth_alloced = cfs_time_current();
 821 }
 822
 823 /**
 824  * Set time when the handle started
 825  */
 826 static void osd_th_started(struct osd_thandle *oth)
 827 {
 828         oth->oth_started = cfs_time_current();
 829 }
 830
 831 /**
 832  * Helper function to convert time interval to microseconds packed in
 833  * long int.
 834  */
 835 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 836 {
 837         struct timeval val;
 838
 839         cfs_duration_usec(cfs_time_sub(end, start), &val);
 840         return val.tv_sec * 1000000 + val.tv_usec;
 841 }
 842
 843 /**
 844  * Check whether the we deal with this handle for too long.
 845  */
 846 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 847                                 cfs_time_t alloced, cfs_time_t started,
 848                                 cfs_time_t closed)
 849 {
 850         cfs_time_t now = cfs_time_current();
 851
 852         LASSERT(dev != NULL);
 853
 854         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 855                             interval_to_usec(alloced, started));
 856         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 857                             interval_to_usec(started, closed));
 858         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 859                             interval_to_usec(closed, now));
 860
 861         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 862                 CWARN("transaction handle %p was open for too long: "
 863                       "now "CFS_TIME_T" ,"
 864                       "alloced "CFS_TIME_T" ,"
 865                       "started "CFS_TIME_T" ,"
 866                       "closed "CFS_TIME_T"\n",
 867                       oth, now, alloced, started, closed);
 868                 libcfs_debug_dumpstack(NULL);
 869         }
 870 }
 871
 872 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 873 {                                                                       \
 874         cfs_time_t __closed = cfs_time_current();                       \
 875         cfs_time_t __alloced = oth->oth_alloced;                        \
 876         cfs_time_t __started = oth->oth_started;                        \
 877                                                                         \
 878         expr;                                                           \
 879         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 880 }
 881
 882 #else /* OSD_THANDLE_STATS */
 883
 884 #define osd_th_alloced(h)                  do {} while(0)
 885 #define osd_th_started(h)                  do {} while(0)
 886 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 887
 888 #endif /* OSD_THANDLE_STATS */
 889
 890 /*
 891  * Concurrency: doesn't access mutable data.
 892  */
 893 static int osd_param_is_not_sane(const struct osd_device *dev,
 894                                  const struct thandle *th)
 895 {
 896         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 897
 898         return oh->ot_credits > osd_transaction_size(dev);
 899 }
 900
 901 /*
 902  * Concurrency: shouldn't matter.
 903  */
 904 static void osd_trans_commit_cb(struct super_block *sb,
 905                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 906 {
 907         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 908         struct thandle     *th  = &oh->ot_super;
 909         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 910         struct dt_txn_commit_cb *dcb, *tmp;
 911
 912         LASSERT(oh->ot_handle == NULL);
 913
 914         if (error)
 915                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 916
 917         dt_txn_hook_commit(th);
 918
 919         /* call per-transaction callbacks if any */
 920         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 921                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 922                          "commit callback entry: magic=%x name='%s'\n",
 923                          dcb->dcb_magic, dcb->dcb_name);
 924                 list_del_init(&dcb->dcb_linkage);
 925                 dcb->dcb_func(NULL, th, dcb, error);
 926         }
 927
 928         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 929         lu_device_put(lud);
 930         th->th_dev = NULL;
 931
 932         lu_context_exit(&th->th_ctx);
 933         lu_context_fini(&th->th_ctx);
 934         thandle_put(th);
 935 }
 936
 937 static struct thandle *osd_trans_create(const struct lu_env *env,
 938                                         struct dt_device *d)
 939 {
 940         struct osd_thread_info  *oti = osd_oti_get(env);
 941         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 942         struct osd_thandle      *oh;
 943         struct thandle          *th;
 944         ENTRY;
 945
 946         /* on pending IO in this thread should left from prev. request */
 947         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 948
 949         th = ERR_PTR(-ENOMEM);
 950         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 951         if (oh != NULL) {
 952                 oh->ot_quota_trans = &oti->oti_quota_trans;
 953                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 954                 th = &oh->ot_super;
 955                 th->th_dev = d;
 956                 th->th_result = 0;
 957                 th->th_tags = LCT_TX_HANDLE;
 958                 oh->ot_credits = 0;
 959                 atomic_set(&th->th_refc, 1);
 960                 th->th_alloc_size = sizeof(*oh);
 961                 oti->oti_dev = osd_dt_dev(d);
 962                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 963                 osd_th_alloced(oh);
 964
 965                 memset(oti->oti_declare_ops, 0,
 966                                         sizeof(oti->oti_declare_ops));
 967                 memset(oti->oti_declare_ops_rb, 0,
 968                                         sizeof(oti->oti_declare_ops_rb));
 969                 memset(oti->oti_declare_ops_cred, 0,
 970                                         sizeof(oti->oti_declare_ops_cred));
 971                 oti->oti_rollback = false;
 972         }
 973         RETURN(th);
 974 }
 975
 976 /*
 977  * Concurrency: shouldn't matter.
 978  */
 979 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 980                            struct thandle *th)
 981 {
 982         struct osd_thread_info *oti = osd_oti_get(env);
 983         struct osd_device  *dev = osd_dt_dev(d);
 984         handle_t           *jh;
 985         struct osd_thandle *oh;
 986         int rc;
 987
 988         ENTRY;
 989
 990         LASSERT(current->journal_info == NULL);
 991
 992         oh = container_of0(th, struct osd_thandle, ot_super);
 993         LASSERT(oh != NULL);
 994         LASSERT(oh->ot_handle == NULL);
 995
 996         rc = dt_txn_hook_start(env, d, th);
 997         if (rc != 0)
 998                 GOTO(out, rc);
 999
1000         if (unlikely(osd_param_is_not_sane(dev, th))) {
1001                 static unsigned long last_printed;
1002                 static int last_credits;
1003
1004                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1005                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1006                       oh->ot_credits,
1007                       osd_journal(dev)->j_max_transaction_buffers);
1008                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_CREATE],
1010                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1011                       oti->oti_declare_ops[OSD_OT_DESTROY],
1012                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1013                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1015                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1016                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1017                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1018                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1019                       oti->oti_declare_ops[OSD_OT_WRITE],
1020                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1021                       oti->oti_declare_ops[OSD_OT_PUNCH],
1022                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1023                       oti->oti_declare_ops[OSD_OT_QUOTA],
1024                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1025                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_INSERT],
1027                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1028                       oti->oti_declare_ops[OSD_OT_DELETE],
1029                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1030                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1031                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1032                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1033                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1034                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1035
1036                 if (last_credits != oh->ot_credits &&
1037                     time_after(jiffies, last_printed +
1038                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1039                         libcfs_debug_dumpstack(NULL);
1040                         last_credits = oh->ot_credits;
1041                         last_printed = jiffies;
1042                 }
1043                 /* XXX Limit the credits to 'max_transaction_buffers', and
1044                  *     let the underlying filesystem to catch the error if
1045                  *     we really need so many credits.
1046                  *
1047                  *     This should be removed when we can calculate the
1048                  *     credits precisely. */
1049                 oh->ot_credits = osd_transaction_size(dev);
1050         }
1051
1052         /*
1053          * XXX temporary stuff. Some abstraction layer should
1054          * be used.
1055          */
1056         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1057         osd_th_started(oh);
1058         if (!IS_ERR(jh)) {
1059                 oh->ot_handle = jh;
1060                 LASSERT(oti->oti_txns == 0);
1061                 lu_context_init(&th->th_ctx, th->th_tags);
1062                 lu_context_enter(&th->th_ctx);
1063
1064                 lu_device_get(&d->dd_lu_dev);
1065                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1066                               "osd-tx", th);
1067                 oti->oti_txns++;
1068                 rc = 0;
1069         } else {
1070                 rc = PTR_ERR(jh);
1071         }
1072 out:
1073         RETURN(rc);
1074 }
1075
1076 static int osd_seq_exists(const struct lu_env *env,
1077                           struct osd_device *osd, u64 seq)
1078 {
1079         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1080         struct seq_server_site  *ss = osd_seq_site(osd);
1081         int                     rc;
1082         ENTRY;
1083
1084         LASSERT(ss != NULL);
1085         LASSERT(ss->ss_server_fld != NULL);
1086
1087         rc = osd_fld_lookup(env, osd, seq, range);
1088         if (rc != 0) {
1089                 if (rc != -ENOENT)
1090                         CERROR("%s: can't lookup FLD sequence "LPX64
1091                                ": rc = %d\n", osd_name(osd), seq, rc);
1092                 RETURN(0);
1093         }
1094
1095         RETURN(ss->ss_node_id == range->lsr_index);
1096 }
1097
1098 /*
1099  * Concurrency: shouldn't matter.
1100  */
1101 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1102                           struct thandle *th)
1103 {
1104         int                     rc = 0;
1105         struct osd_thandle     *oh;
1106         struct osd_thread_info *oti = osd_oti_get(env);
1107         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1108         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1109         struct lquota_trans    *qtrans;
1110         ENTRY;
1111
1112         oh = container_of0(th, struct osd_thandle, ot_super);
1113
1114         qtrans = oh->ot_quota_trans;
1115         oh->ot_quota_trans = NULL;
1116
1117         if (oh->ot_handle != NULL) {
1118                 handle_t *hdl = oh->ot_handle;
1119
1120                 /*
1121                  * add commit callback
1122                  * notice we don't do this in osd_trans_start()
1123                  * as underlying transaction can change during truncate
1124                  */
1125                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1126                                          &oh->ot_jcb);
1127
1128                 LASSERT(oti->oti_txns == 1);
1129                 oti->oti_txns--;
1130                 rc = dt_txn_hook_stop(env, th);
1131                 if (rc != 0)
1132                         CERROR("Failure in transaction hook: %d\n", rc);
1133
1134                 /* hook functions might modify th_sync */
1135                 hdl->h_sync = th->th_sync;
1136
1137                 oh->ot_handle = NULL;
1138                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1139                                   rc = ldiskfs_journal_stop(hdl));
1140                 if (rc != 0)
1141                         CERROR("Failure to stop transaction: %d\n", rc);
1142         } else {
1143                 thandle_put(&oh->ot_super);
1144         }
1145
1146         /* inform the quota slave device that the transaction is stopping */
1147         qsd_op_end(env, qsd, qtrans);
1148
1149         /* as we want IO to journal and data IO be concurrent, we don't block
1150          * awaiting data IO completion in osd_do_bio(), instead we wait here
1151          * once transaction is submitted to the journal. all reqular requests
1152          * don't do direct IO (except read/write), thus this wait_event becomes
1153          * no-op for them.
1154          *
1155          * IMPORTANT: we have to wait till any IO submited by the thread is
1156          * completed otherwise iobuf may be corrupted by different request
1157          */
1158         wait_event(iobuf->dr_wait,
1159                        atomic_read(&iobuf->dr_numreqs) == 0);
1160         osd_fini_iobuf(oti->oti_dev, iobuf);
1161         if (!rc)
1162                 rc = iobuf->dr_error;
1163
1164         RETURN(rc);
1165 }
1166
1167 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1168 {
1169         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1170                                                ot_super);
1171
1172         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1173         LASSERT(&dcb->dcb_func != NULL);
1174         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1175
1176         return 0;
1177 }
1178
1179 /*
1180  * Called just before object is freed. Releases all resources except for
1181  * object itself (that is released by osd_object_free()).
1182  *
1183  * Concurrency: no concurrent access is possible that late in object
1184  * life-cycle.
1185  */
1186 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1187 {
1188         struct osd_object *obj   = osd_obj(l);
1189         struct inode      *inode = obj->oo_inode;
1190
1191         LINVRNT(osd_invariant(obj));
1192
1193         /*
1194          * If object is unlinked remove fid->ino mapping from object index.
1195          */
1196
1197         osd_index_fini(obj);
1198         if (inode != NULL) {
1199                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1200                 qid_t                    uid = i_uid_read(inode);
1201                 qid_t                    gid = i_gid_read(inode);
1202
1203                 iput(inode);
1204                 obj->oo_inode = NULL;
1205
1206                 if (qsd != NULL) {
1207                         struct osd_thread_info  *info = osd_oti_get(env);
1208                         struct lquota_id_info   *qi = &info->oti_qi;
1209
1210                         /* Release granted quota to master if necessary */
1211                         qi->lqi_id.qid_uid = uid;
1212                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1213
1214                         qi->lqi_id.qid_uid = gid;
1215                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1216                 }
1217         }
1218 }
1219
1220 /*
1221  * Concurrency: ->loo_object_release() is called under site spin-lock.
1222  */
1223 static void osd_object_release(const struct lu_env *env,
1224                                struct lu_object *l)
1225 {
1226 }
1227
1228 /*
1229  * Concurrency: shouldn't matter.
1230  */
1231 static int osd_object_print(const struct lu_env *env, void *cookie,
1232                             lu_printer_t p, const struct lu_object *l)
1233 {
1234         struct osd_object *o = osd_obj(l);
1235         struct iam_descr  *d;
1236
1237         if (o->oo_dir != NULL)
1238                 d = o->oo_dir->od_container.ic_descr;
1239         else
1240                 d = NULL;
1241         return (*p)(env, cookie,
1242                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1243                     o, o->oo_inode,
1244                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1245                     o->oo_inode ? o->oo_inode->i_generation : 0,
1246                     d ? d->id_ops->id_name : "plain");
1247 }
1248
1249 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1250
1251 /*
1252  * Concurrency: shouldn't matter.
1253  */
1254 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1255                struct obd_statfs *sfs)
1256 {
1257         struct osd_device  *osd = osd_dt_dev(d);
1258         struct super_block *sb = osd_sb(osd);
1259         struct kstatfs     *ksfs;
1260         int result = 0;
1261
1262         if (unlikely(osd->od_mnt == NULL))
1263                 return -EINPROGRESS;
1264
1265         /* osd_lproc.c call this without env, allocate ksfs for that case */
1266         if (unlikely(env == NULL)) {
1267                 OBD_ALLOC_PTR(ksfs);
1268                 if (ksfs == NULL)
1269                         return -ENOMEM;
1270         } else {
1271                 ksfs = &osd_oti_get(env)->oti_ksfs;
1272         }
1273
1274         spin_lock(&osd->od_osfs_lock);
1275         result = sb->s_op->statfs(sb->s_root, ksfs);
1276         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1277                 statfs_pack(sfs, ksfs);
1278                 if (sb->s_flags & MS_RDONLY)
1279                         sfs->os_state = OS_STATE_READONLY;
1280         }
1281
1282         spin_unlock(&osd->od_osfs_lock);
1283
1284         if (unlikely(env == NULL))
1285                 OBD_FREE_PTR(ksfs);
1286
1287         /* Reserve a small amount of space for local objects like last_rcvd,
1288          * llog, quota files, ... */
1289         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1290                 sfs->os_bavail = 0;
1291         } else {
1292                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1293                 /** Take out metadata overhead for indirect blocks */
1294                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1295         }
1296
1297         return result;
1298 }
1299
1300 /**
1301  * Estimate space needed for file creations. We assume the largest filename
1302  * which is 2^64 - 1, hence a filename of 20 chars.
1303  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1304  */
1305 #ifdef __LDISKFS_DIR_REC_LEN
1306 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1307 #else
1308 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1309 #endif
1310
1311 /*
1312  * Concurrency: doesn't access mutable data.
1313  */
1314 static void osd_conf_get(const struct lu_env *env,
1315                          const struct dt_device *dev,
1316                          struct dt_device_param *param)
1317 {
1318         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1319         int                ea_overhead;
1320
1321         /*
1322          * XXX should be taken from not-yet-existing fs abstraction layer.
1323          */
1324         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1325         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1326         param->ddp_block_shift  = sb->s_blocksize_bits;
1327         param->ddp_mount_type     = LDD_MT_LDISKFS;
1328         param->ddp_maxbytes       = sb->s_maxbytes;
1329         /* Overhead estimate should be fairly accurate, so we really take a tiny
1330          * error margin which also avoids fragmenting the filesystem too much */
1331         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1332         /* inode are statically allocated, so per-inode space consumption
1333          * is the space consumed by the directory entry */
1334         param->ddp_inodespace     = PER_OBJ_USAGE;
1335         /* per-fragment overhead to be used by the client code */
1336         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1337         param->ddp_mntopts      = 0;
1338         if (test_opt(sb, XATTR_USER))
1339                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1340         if (test_opt(sb, POSIX_ACL))
1341                 param->ddp_mntopts |= MNTOPT_ACL;
1342
1343         /* LOD might calculate the max stripe count based on max_ea_size,
1344          * so we need take account in the overhead as well,
1345          * xattr_header + magic + xattr_entry_head */
1346         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1347                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1348
1349 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1350         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1351                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1352                                                                 ea_overhead;
1353         else
1354 #endif
1355                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1356 }
1357
1358 /*
1359  * Concurrency: shouldn't matter.
1360  */
1361 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1362 {
1363         int rc;
1364
1365         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1366
1367         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1368
1369         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1370                LUSTRE_OSD_LDISKFS_NAME, rc);
1371
1372         return rc;
1373 }
1374
1375 /**
1376  * Start commit for OSD device.
1377  *
1378  * An implementation of dt_commit_async method for OSD device.
1379  * Asychronously starts underlayng fs sync and thereby a transaction
1380  * commit.
1381  *
1382  * \param env environment
1383  * \param d dt device
1384  *
1385  * \see dt_device_operations
1386  */
1387 static int osd_commit_async(const struct lu_env *env,
1388                             struct dt_device *d)
1389 {
1390         struct super_block *s = osd_sb(osd_dt_dev(d));
1391         ENTRY;
1392
1393         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1394         RETURN(s->s_op->sync_fs(s, 0));
1395 }
1396
1397 /*
1398  * Concurrency: shouldn't matter.
1399  */
1400
1401 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1402 {
1403         struct super_block *sb = osd_sb(osd_dt_dev(d));
1404         struct block_device *dev = sb->s_bdev;
1405 #ifdef HAVE_DEV_SET_RDONLY
1406         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1407         int rc = 0;
1408 #else
1409         int rc = -EOPNOTSUPP;
1410 #endif
1411         ENTRY;
1412
1413 #ifdef HAVE_DEV_SET_RDONLY
1414         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1415
1416         if (jdev && (jdev != dev)) {
1417                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1418                        (long)jdev);
1419                 dev_set_rdonly(jdev);
1420         }
1421         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1422         dev_set_rdonly(dev);
1423 #else
1424         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1425                osd_dt_dev(d)->od_svname, (long)dev, rc);
1426 #endif
1427         RETURN(rc);
1428 }
1429
1430 /*
1431  * Concurrency: serialization provided by callers.
1432  */
1433 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1434                               int mode, unsigned long timeout, __u32 alg,
1435                               struct lustre_capa_key *keys)
1436 {
1437         struct osd_device *dev = osd_dt_dev(d);
1438         ENTRY;
1439
1440         dev->od_fl_capa = mode;
1441         dev->od_capa_timeout = timeout;
1442         dev->od_capa_alg = alg;
1443         dev->od_capa_keys = keys;
1444         RETURN(0);
1445 }
1446
1447 /**
1448  * Note: we do not count into QUOTA here.
1449  * If we mount with --data_journal we may need more.
1450  */
1451 const int osd_dto_credits_noquota[DTO_NR] = {
1452         /**
1453          * Insert.
1454          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1455          * SINGLEDATA_TRANS_BLOCKS(8)
1456          * XXX Note: maybe iam need more, since iam have more level than
1457          *           EXT3 htree.
1458          */
1459         [DTO_INDEX_INSERT]  = 16,
1460         /**
1461          * Delete
1462          * just modify a single entry, probably merge few within a block
1463          */
1464         [DTO_INDEX_DELETE]  = 1,
1465         /**
1466          * Used for OI scrub
1467          */
1468         [DTO_INDEX_UPDATE]  = 16,
1469         /**
1470          * 4(inode, inode bits, groups, GDT)
1471          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1472          */
1473         [DTO_OBJECT_CREATE] = 4,
1474         /**
1475          * 4(inode, inode bits, groups, GDT)
1476          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1477          */
1478         [DTO_OBJECT_DELETE] = 4,
1479         /**
1480          * Attr set credits (inode)
1481          */
1482         [DTO_ATTR_SET_BASE] = 1,
1483         /**
1484          * Xattr set. The same as xattr of EXT3.
1485          * DATA_TRANS_BLOCKS(14)
1486          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1487          * are also counted in. Do not know why?
1488          */
1489         [DTO_XATTR_SET]     = 14,
1490         /**
1491          * credits for inode change during write.
1492          */
1493         [DTO_WRITE_BASE]    = 3,
1494         /**
1495          * credits for single block write.
1496          */
1497         [DTO_WRITE_BLOCK]   = 14,
1498         /**
1499          * Attr set credits for chown.
1500          * This is extra credits for setattr, and it is null without quota
1501          */
1502         [DTO_ATTR_SET_CHOWN] = 0
1503 };
1504
1505 static const struct dt_device_operations osd_dt_ops = {
1506         .dt_root_get       = osd_root_get,
1507         .dt_statfs         = osd_statfs,
1508         .dt_trans_create   = osd_trans_create,
1509         .dt_trans_start    = osd_trans_start,
1510         .dt_trans_stop     = osd_trans_stop,
1511         .dt_trans_cb_add   = osd_trans_cb_add,
1512         .dt_conf_get       = osd_conf_get,
1513         .dt_sync           = osd_sync,
1514         .dt_ro             = osd_ro,
1515         .dt_commit_async   = osd_commit_async,
1516         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1517 };
1518
1519 static void osd_object_read_lock(const struct lu_env *env,
1520                                  struct dt_object *dt, unsigned role)
1521 {
1522         struct osd_object *obj = osd_dt_obj(dt);
1523         struct osd_thread_info *oti = osd_oti_get(env);
1524
1525         LINVRNT(osd_invariant(obj));
1526
1527         LASSERT(obj->oo_owner != env);
1528         down_read_nested(&obj->oo_sem, role);
1529
1530         LASSERT(obj->oo_owner == NULL);
1531         oti->oti_r_locks++;
1532 }
1533
1534 static void osd_object_write_lock(const struct lu_env *env,
1535                                   struct dt_object *dt, unsigned role)
1536 {
1537         struct osd_object *obj = osd_dt_obj(dt);
1538         struct osd_thread_info *oti = osd_oti_get(env);
1539
1540         LINVRNT(osd_invariant(obj));
1541
1542         LASSERT(obj->oo_owner != env);
1543         down_write_nested(&obj->oo_sem, role);
1544
1545         LASSERT(obj->oo_owner == NULL);
1546         obj->oo_owner = env;
1547         oti->oti_w_locks++;
1548 }
1549
1550 static void osd_object_read_unlock(const struct lu_env *env,
1551                                    struct dt_object *dt)
1552 {
1553         struct osd_object *obj = osd_dt_obj(dt);
1554         struct osd_thread_info *oti = osd_oti_get(env);
1555
1556         LINVRNT(osd_invariant(obj));
1557
1558         LASSERT(oti->oti_r_locks > 0);
1559         oti->oti_r_locks--;
1560         up_read(&obj->oo_sem);
1561 }
1562
1563 static void osd_object_write_unlock(const struct lu_env *env,
1564                                     struct dt_object *dt)
1565 {
1566         struct osd_object *obj = osd_dt_obj(dt);
1567         struct osd_thread_info *oti = osd_oti_get(env);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         LASSERT(obj->oo_owner == env);
1572         LASSERT(oti->oti_w_locks > 0);
1573         oti->oti_w_locks--;
1574         obj->oo_owner = NULL;
1575         up_write(&obj->oo_sem);
1576 }
1577
1578 static int osd_object_write_locked(const struct lu_env *env,
1579                                    struct dt_object *dt)
1580 {
1581         struct osd_object *obj = osd_dt_obj(dt);
1582
1583         LINVRNT(osd_invariant(obj));
1584
1585         return obj->oo_owner == env;
1586 }
1587
1588 static int capa_is_sane(const struct lu_env *env,
1589                         struct osd_device *dev,
1590                         struct lustre_capa *capa,
1591                         struct lustre_capa_key *keys)
1592 {
1593         struct osd_thread_info *oti = osd_oti_get(env);
1594         struct lustre_capa *tcapa = &oti->oti_capa;
1595         struct obd_capa *oc;
1596         int i, rc = 0;
1597         ENTRY;
1598
1599         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1600         if (oc) {
1601                 if (capa_is_expired(oc)) {
1602                         DEBUG_CAPA(D_ERROR, capa, "expired");
1603                         rc = -ESTALE;
1604                 }
1605                 capa_put(oc);
1606                 RETURN(rc);
1607         }
1608
1609         if (capa_is_expired_sec(capa)) {
1610                 DEBUG_CAPA(D_ERROR, capa, "expired");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         spin_lock(&capa_lock);
1615         for (i = 0; i < 2; i++) {
1616                 if (keys[i].lk_keyid == capa->lc_keyid) {
1617                         oti->oti_capa_key = keys[i];
1618                         break;
1619                 }
1620         }
1621         spin_unlock(&capa_lock);
1622
1623         if (i == 2) {
1624                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1625                 RETURN(-ESTALE);
1626         }
1627
1628         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1629         if (rc)
1630                 RETURN(rc);
1631
1632         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1633                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1634                 RETURN(-EACCES);
1635         }
1636
1637         oc = capa_add(dev->od_capa_hash, capa);
1638         capa_put(oc);
1639
1640         RETURN(0);
1641 }
1642
1643 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1644                     struct lustre_capa *capa, __u64 opc)
1645 {
1646         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1647         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1648         struct lu_capainfo *lci;
1649         int rc;
1650
1651         if (!osd->od_fl_capa)
1652                 return 0;
1653
1654         if (capa == BYPASS_CAPA)
1655                 return 0;
1656
1657         lci = lu_capainfo_get(env);
1658         if (unlikely(lci == NULL))
1659                 return 0;
1660
1661         if (lci->lci_auth == LC_ID_NONE)
1662                 return 0;
1663
1664         if (capa == NULL) {
1665                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1666                        osd_name(osd), PFID(fid), -EACCES);
1667                 return -EACCES;
1668         }
1669
1670         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1671                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1672                            PFID(fid));
1673                 return -EACCES;
1674         }
1675
1676         if (!capa_opc_supported(capa, opc)) {
1677                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1678                 return -EACCES;
1679         }
1680
1681         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1682         if (rc != 0) {
1683                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1684                 return -EACCES;
1685         }
1686
1687         return 0;
1688 }
1689
1690 static struct timespec *osd_inode_time(const struct lu_env *env,
1691                                        struct inode *inode, __u64 seconds)
1692 {
1693         struct osd_thread_info  *oti = osd_oti_get(env);
1694         struct timespec         *t   = &oti->oti_time;
1695
1696         t->tv_sec = seconds;
1697         t->tv_nsec = 0;
1698         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1699         return t;
1700 }
1701
1702
1703 static void osd_inode_getattr(const struct lu_env *env,
1704                               struct inode *inode, struct lu_attr *attr)
1705 {
1706         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1707                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1708                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1709                            LA_TYPE;
1710
1711         attr->la_atime   = LTIME_S(inode->i_atime);
1712         attr->la_mtime   = LTIME_S(inode->i_mtime);
1713         attr->la_ctime   = LTIME_S(inode->i_ctime);
1714         attr->la_mode    = inode->i_mode;
1715         attr->la_size    = i_size_read(inode);
1716         attr->la_blocks  = inode->i_blocks;
1717         attr->la_uid     = i_uid_read(inode);
1718         attr->la_gid     = i_gid_read(inode);
1719         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1720         attr->la_nlink   = inode->i_nlink;
1721         attr->la_rdev    = inode->i_rdev;
1722         attr->la_blksize = 1 << inode->i_blkbits;
1723         attr->la_blkbits = inode->i_blkbits;
1724 }
1725
1726 static int osd_attr_get(const struct lu_env *env,
1727                         struct dt_object *dt,
1728                         struct lu_attr *attr,
1729                         struct lustre_capa *capa)
1730 {
1731         struct osd_object *obj = osd_dt_obj(dt);
1732
1733         LASSERT(dt_object_exists(dt));
1734         LASSERT(!dt_object_remote(dt));
1735         LINVRNT(osd_invariant(obj));
1736
1737         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1738                 return -EACCES;
1739
1740         spin_lock(&obj->oo_guard);
1741         osd_inode_getattr(env, obj->oo_inode, attr);
1742         spin_unlock(&obj->oo_guard);
1743         return 0;
1744 }
1745
1746 static int osd_declare_attr_set(const struct lu_env *env,
1747                                 struct dt_object *dt,
1748                                 const struct lu_attr *attr,
1749                                 struct thandle *handle)
1750 {
1751         struct osd_thandle     *oh;
1752         struct osd_object      *obj;
1753         struct osd_thread_info *info = osd_oti_get(env);
1754         struct lquota_id_info  *qi = &info->oti_qi;
1755         qid_t                   uid;
1756         qid_t                   gid;
1757         long long               bspace;
1758         int                     rc = 0;
1759         bool                    enforce;
1760         ENTRY;
1761
1762         LASSERT(dt != NULL);
1763         LASSERT(handle != NULL);
1764
1765         obj = osd_dt_obj(dt);
1766         LASSERT(osd_invariant(obj));
1767
1768         oh = container_of0(handle, struct osd_thandle, ot_super);
1769         LASSERT(oh->ot_handle == NULL);
1770
1771         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1772                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1773
1774         if (attr == NULL || obj->oo_inode == NULL)
1775                 RETURN(rc);
1776
1777         bspace   = obj->oo_inode->i_blocks;
1778         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1779         bspace   = toqb(bspace);
1780
1781         /* Changing ownership is always preformed by super user, it should not
1782          * fail with EDQUOT.
1783          *
1784          * We still need to call the osd_declare_qid() to calculate the journal
1785          * credits for updating quota accounting files and to trigger quota
1786          * space adjustment once the operation is completed.*/
1787         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1788                 /* USERQUOTA */
1789                 uid = i_uid_read(obj->oo_inode);
1790                 qi->lqi_type = USRQUOTA;
1791                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1792                 /* inode accounting */
1793                 qi->lqi_is_blk = false;
1794
1795                 /* one more inode for the new uid ... */
1796                 qi->lqi_id.qid_uid = attr->la_uid;
1797                 qi->lqi_space      = 1;
1798                 /* Reserve credits for the new uid */
1799                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1800                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1801                         rc = 0;
1802                 if (rc)
1803                         RETURN(rc);
1804
1805                 /* and one less inode for the current uid */
1806                 qi->lqi_id.qid_uid = uid;
1807                 qi->lqi_space      = -1;
1808                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1809                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1810                         rc = 0;
1811                 if (rc)
1812                         RETURN(rc);
1813
1814                 /* block accounting */
1815                 qi->lqi_is_blk = true;
1816
1817                 /* more blocks for the new uid ... */
1818                 qi->lqi_id.qid_uid = attr->la_uid;
1819                 qi->lqi_space      = bspace;
1820                 /*
1821                  * Credits for the new uid has been reserved, re-use "obj"
1822                  * to save credit reservation.
1823                  */
1824                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1825                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1826                         rc = 0;
1827                 if (rc)
1828                         RETURN(rc);
1829
1830                 /* and finally less blocks for the current uid */
1831                 qi->lqi_id.qid_uid = uid;
1832                 qi->lqi_space      = -bspace;
1833                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1834                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1835                         rc = 0;
1836                 if (rc)
1837                         RETURN(rc);
1838
1839                 /* GROUP QUOTA */
1840                 gid = i_gid_read(obj->oo_inode);
1841                 qi->lqi_type = GRPQUOTA;
1842                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1843
1844                 /* inode accounting */
1845                 qi->lqi_is_blk = false;
1846
1847                 /* one more inode for the new gid ... */
1848                 qi->lqi_id.qid_gid = attr->la_gid;
1849                 qi->lqi_space      = 1;
1850                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1851                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1852                         rc = 0;
1853                 if (rc)
1854                         RETURN(rc);
1855
1856                 /* and one less inode for the current gid */
1857                 qi->lqi_id.qid_gid = gid;
1858                 qi->lqi_space      = -1;
1859                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1860                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1861                         rc = 0;
1862                 if (rc)
1863                         RETURN(rc);
1864
1865                 /* block accounting */
1866                 qi->lqi_is_blk = true;
1867
1868                 /* more blocks for the new gid ... */
1869                 qi->lqi_id.qid_gid = attr->la_gid;
1870                 qi->lqi_space      = bspace;
1871                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1872                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1873                         rc = 0;
1874                 if (rc)
1875                         RETURN(rc);
1876
1877                 /* and finally less blocks for the current gid */
1878                 qi->lqi_id.qid_gid = gid;
1879                 qi->lqi_space      = -bspace;
1880                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1881                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1882                         rc = 0;
1883                 if (rc)
1884                         RETURN(rc);
1885         }
1886
1887         RETURN(rc);
1888 }
1889
1890 static int osd_inode_setattr(const struct lu_env *env,
1891                              struct inode *inode, const struct lu_attr *attr)
1892 {
1893         __u64 bits = attr->la_valid;
1894
1895         /* Only allow set size for regular file */
1896         if (!S_ISREG(inode->i_mode))
1897                 bits &= ~(LA_SIZE | LA_BLOCKS);
1898
1899         if (bits == 0)
1900                 return 0;
1901
1902         if (bits & LA_ATIME)
1903                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1904         if (bits & LA_CTIME)
1905                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1906         if (bits & LA_MTIME)
1907                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1908         if (bits & LA_SIZE) {
1909                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1910                 i_size_write(inode, attr->la_size);
1911         }
1912
1913 #if 0
1914         /* OSD should not change "i_blocks" which is used by quota.
1915          * "i_blocks" should be changed by ldiskfs only. */
1916         if (bits & LA_BLOCKS)
1917                 inode->i_blocks = attr->la_blocks;
1918 #endif
1919         if (bits & LA_MODE)
1920                 inode->i_mode = (inode->i_mode & S_IFMT) |
1921                                 (attr->la_mode & ~S_IFMT);
1922         if (bits & LA_UID)
1923                 i_uid_write(inode, attr->la_uid);
1924         if (bits & LA_GID)
1925                 i_gid_write(inode, attr->la_gid);
1926         if (bits & LA_NLINK)
1927                 set_nlink(inode, attr->la_nlink);
1928         if (bits & LA_RDEV)
1929                 inode->i_rdev = attr->la_rdev;
1930
1931         if (bits & LA_FLAGS) {
1932                 /* always keep S_NOCMTIME */
1933                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1934                                  S_NOCMTIME;
1935         }
1936         return 0;
1937 }
1938
1939 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1940 {
1941         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1942             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1943                 struct iattr    iattr;
1944                 int             rc;
1945
1946                 ll_vfs_dq_init(inode);
1947                 iattr.ia_valid = 0;
1948                 if (attr->la_valid & LA_UID)
1949                         iattr.ia_valid |= ATTR_UID;
1950                 if (attr->la_valid & LA_GID)
1951                         iattr.ia_valid |= ATTR_GID;
1952                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1953                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1954
1955                 rc = ll_vfs_dq_transfer(inode, &iattr);
1956                 if (rc) {
1957                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1958                                "enforcement enabled on the ldiskfs "
1959                                "filesystem?\n", inode->i_sb->s_id, rc);
1960                         return rc;
1961                 }
1962         }
1963         return 0;
1964 }
1965
1966 static int osd_attr_set(const struct lu_env *env,
1967                         struct dt_object *dt,
1968                         const struct lu_attr *attr,
1969                         struct thandle *handle,
1970                         struct lustre_capa *capa)
1971 {
1972         struct osd_object *obj = osd_dt_obj(dt);
1973         struct inode      *inode;
1974         int rc;
1975
1976         LASSERT(handle != NULL);
1977         LASSERT(dt_object_exists(dt));
1978         LASSERT(!dt_object_remote(dt));
1979         LASSERT(osd_invariant(obj));
1980
1981         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1982                 return -EACCES;
1983
1984         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1985
1986         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1987                 struct osd_thread_info  *oti  = osd_oti_get(env);
1988                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1989                 struct lu_fid           *fid1 = &oti->oti_fid;
1990                 struct osd_inode_id     *id   = &oti->oti_id;
1991                 struct iam_path_descr   *ipd;
1992                 struct iam_container    *bag;
1993                 struct osd_thandle      *oh;
1994                 int                      rc;
1995
1996                 fid_cpu_to_be(fid1, fid0);
1997                 memset(id, 1, sizeof(*id));
1998                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1999                                   fid0)->oi_dir.od_container;
2000                 ipd = osd_idx_ipd_get(env, bag);
2001                 if (unlikely(ipd == NULL))
2002                         RETURN(-ENOMEM);
2003
2004                 oh = container_of0(handle, struct osd_thandle, ot_super);
2005                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2006                                 (const struct iam_rec *)id, ipd);
2007                 osd_ipd_put(env, bag, ipd);
2008                 return(rc > 0 ? 0 : rc);
2009         }
2010
2011         inode = obj->oo_inode;
2012
2013         rc = osd_quota_transfer(inode, attr);
2014         if (rc)
2015                 return rc;
2016
2017         spin_lock(&obj->oo_guard);
2018         rc = osd_inode_setattr(env, inode, attr);
2019         spin_unlock(&obj->oo_guard);
2020
2021         if (!rc)
2022                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2023         return rc;
2024 }
2025
2026 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2027                                            struct osd_object *obj,
2028                                            const char *name, const int namelen)
2029 {
2030         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2031 }
2032
2033 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2034                       umode_t mode, struct dt_allocation_hint *hint,
2035                       struct thandle *th)
2036 {
2037         int result;
2038         struct osd_device  *osd = osd_obj2dev(obj);
2039         struct osd_thandle *oth;
2040         struct dt_object   *parent = NULL;
2041         struct inode       *inode;
2042
2043         LINVRNT(osd_invariant(obj));
2044         LASSERT(obj->oo_inode == NULL);
2045         LASSERT(obj->oo_hl_head == NULL);
2046
2047         if (S_ISDIR(mode) && ldiskfs_pdo) {
2048                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2049                 if (obj->oo_hl_head == NULL)
2050                         return -ENOMEM;
2051         }
2052
2053         oth = container_of(th, struct osd_thandle, ot_super);
2054         LASSERT(oth->ot_handle->h_transaction != NULL);
2055
2056         if (hint && hint->dah_parent)
2057                 parent = hint->dah_parent;
2058
2059         inode = ldiskfs_create_inode(oth->ot_handle,
2060                                      parent ? osd_dt_obj(parent)->oo_inode :
2061                                               osd_sb(osd)->s_root->d_inode,
2062                                      mode);
2063         if (!IS_ERR(inode)) {
2064                 /* Do not update file c/mtime in ldiskfs.
2065                  * NB: don't need any lock because no contention at this
2066                  * early stage */
2067                 inode->i_flags |= S_NOCMTIME;
2068
2069                 /* For new created object, it must be consistent,
2070                  * and it is unnecessary to scrub against it. */
2071                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2072                 obj->oo_inode = inode;
2073                 result = 0;
2074         } else {
2075                 if (obj->oo_hl_head != NULL) {
2076                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2077                         obj->oo_hl_head = NULL;
2078                 }
2079                 result = PTR_ERR(inode);
2080         }
2081         LINVRNT(osd_invariant(obj));
2082         return result;
2083 }
2084
2085 enum {
2086         OSD_NAME_LEN = 255
2087 };
2088
2089 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2090                      struct lu_attr *attr,
2091                      struct dt_allocation_hint *hint,
2092                      struct dt_object_format *dof,
2093                      struct thandle *th)
2094 {
2095         int result;
2096         struct osd_thandle *oth;
2097         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2098
2099         LASSERT(S_ISDIR(attr->la_mode));
2100
2101         oth = container_of(th, struct osd_thandle, ot_super);
2102         LASSERT(oth->ot_handle->h_transaction != NULL);
2103         result = osd_mkfile(info, obj, mode, hint, th);
2104
2105         return result;
2106 }
2107
2108 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2109                         struct lu_attr *attr,
2110                         struct dt_allocation_hint *hint,
2111                         struct dt_object_format *dof,
2112                         struct thandle *th)
2113 {
2114         int result;
2115         struct osd_thandle *oth;
2116         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2117
2118         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2119
2120         LASSERT(S_ISREG(attr->la_mode));
2121
2122         oth = container_of(th, struct osd_thandle, ot_super);
2123         LASSERT(oth->ot_handle->h_transaction != NULL);
2124
2125         result = osd_mkfile(info, obj, mode, hint, th);
2126         if (result == 0) {
2127                 LASSERT(obj->oo_inode != NULL);
2128                 if (feat->dif_flags & DT_IND_VARKEY)
2129                         result = iam_lvar_create(obj->oo_inode,
2130                                                  feat->dif_keysize_max,
2131                                                  feat->dif_ptrsize,
2132                                                  feat->dif_recsize_max,
2133                                                  oth->ot_handle);
2134                 else
2135                         result = iam_lfix_create(obj->oo_inode,
2136                                                  feat->dif_keysize_max,
2137                                                  feat->dif_ptrsize,
2138                                                  feat->dif_recsize_max,
2139                                                  oth->ot_handle);
2140
2141         }
2142         return result;
2143 }
2144
2145 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2146                      struct lu_attr *attr,
2147                      struct dt_allocation_hint *hint,
2148                      struct dt_object_format *dof,
2149                      struct thandle *th)
2150 {
2151         LASSERT(S_ISREG(attr->la_mode));
2152         return osd_mkfile(info, obj, (attr->la_mode &
2153                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2154 }
2155
2156 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2157                      struct lu_attr *attr,
2158                      struct dt_allocation_hint *hint,
2159                      struct dt_object_format *dof,
2160                      struct thandle *th)
2161 {
2162         LASSERT(S_ISLNK(attr->la_mode));
2163         return osd_mkfile(info, obj, (attr->la_mode &
2164                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2165 }
2166
2167 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2168                      struct lu_attr *attr,
2169                      struct dt_allocation_hint *hint,
2170                      struct dt_object_format *dof,
2171                      struct thandle *th)
2172 {
2173         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2174         int result;
2175
2176         LINVRNT(osd_invariant(obj));
2177         LASSERT(obj->oo_inode == NULL);
2178         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2179                 S_ISFIFO(mode) || S_ISSOCK(mode));
2180
2181         result = osd_mkfile(info, obj, mode, hint, th);
2182         if (result == 0) {
2183                 LASSERT(obj->oo_inode != NULL);
2184                 /*
2185                  * This inode should be marked dirty for i_rdev.  Currently
2186                  * that is done in the osd_attr_init().
2187                  */
2188                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2189                                    attr->la_rdev);
2190         }
2191         LINVRNT(osd_invariant(obj));
2192         return result;
2193 }
2194
2195 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2196                               struct lu_attr *,
2197                               struct dt_allocation_hint *hint,
2198                               struct dt_object_format *dof,
2199                               struct thandle *);
2200
2201 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2202 {
2203         osd_obj_type_f result;
2204
2205         switch (type) {
2206         case DFT_DIR:
2207                 result = osd_mkdir;
2208                 break;
2209         case DFT_REGULAR:
2210                 result = osd_mkreg;
2211                 break;
2212         case DFT_SYM:
2213                 result = osd_mksym;
2214                 break;
2215         case DFT_NODE:
2216                 result = osd_mknod;
2217                 break;
2218         case DFT_INDEX:
2219                 result = osd_mk_index;
2220                 break;
2221
2222         default:
2223                 LBUG();
2224                 break;
2225         }
2226         return result;
2227 }
2228
2229
2230 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2231                         struct dt_object *parent, struct dt_object *child,
2232                         umode_t child_mode)
2233 {
2234         LASSERT(ah);
2235
2236         ah->dah_parent = parent;
2237         ah->dah_mode = child_mode;
2238 }
2239
2240 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2241                           struct lu_attr *attr, struct dt_object_format *dof)
2242 {
2243         struct inode   *inode = obj->oo_inode;
2244         __u64           valid = attr->la_valid;
2245         int             result;
2246
2247         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2248
2249         if (dof->dof_type != DFT_NODE)
2250                 attr->la_valid &= ~LA_RDEV;
2251         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2252                 attr->la_valid &= ~LA_ATIME;
2253         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2254                 attr->la_valid &= ~LA_CTIME;
2255         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2256                 attr->la_valid &= ~LA_MTIME;
2257
2258         result = osd_quota_transfer(inode, attr);
2259         if (result)
2260                 return;
2261
2262         if (attr->la_valid != 0) {
2263                 result = osd_inode_setattr(info->oti_env, inode, attr);
2264                 /*
2265                  * The osd_inode_setattr() should always succeed here.  The
2266                  * only error that could be returned is EDQUOT when we are
2267                  * trying to change the UID or GID of the inode. However, this
2268                  * should not happen since quota enforcement is no longer
2269                  * enabled on ldiskfs (lquota takes care of it).
2270                  */
2271                 LASSERTF(result == 0, "%d\n", result);
2272                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2273         }
2274
2275         attr->la_valid = valid;
2276 }
2277
2278 /**
2279  * Helper function for osd_object_create()
2280  *
2281  * \retval 0, on success
2282  */
2283 static int __osd_object_create(struct osd_thread_info *info,
2284                                struct osd_object *obj, struct lu_attr *attr,
2285                                struct dt_allocation_hint *hint,
2286                                struct dt_object_format *dof,
2287                                struct thandle *th)
2288 {
2289         int     result;
2290         __u32   umask;
2291
2292         /* we drop umask so that permissions we pass are not affected */
2293         umask = current->fs->umask;
2294         current->fs->umask = 0;
2295
2296         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2297                                                   th);
2298         if (result == 0) {
2299                 osd_attr_init(info, obj, attr, dof);
2300                 osd_object_init0(obj);
2301                 /* bz 24037 */
2302                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2303                         unlock_new_inode(obj->oo_inode);
2304         }
2305
2306         /* restore previous umask value */
2307         current->fs->umask = umask;
2308
2309         return result;
2310 }
2311
2312 /**
2313  * Helper function for osd_object_create()
2314  *
2315  * \retval 0, on success
2316  */
2317 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2318                            const struct lu_fid *fid, struct thandle *th)
2319 {
2320         struct osd_thread_info *info = osd_oti_get(env);
2321         struct osd_inode_id    *id   = &info->oti_id;
2322         struct osd_device      *osd  = osd_obj2dev(obj);
2323         struct osd_thandle     *oh;
2324
2325         LASSERT(obj->oo_inode != NULL);
2326
2327         oh = container_of0(th, struct osd_thandle, ot_super);
2328         LASSERT(oh->ot_handle);
2329
2330         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2331         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2332 }
2333
2334 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2335                    u64 seq, struct lu_seq_range *range)
2336 {
2337         struct seq_server_site  *ss = osd_seq_site(osd);
2338
2339         if (fid_seq_is_idif(seq)) {
2340                 fld_range_set_ost(range);
2341                 range->lsr_index = idif_ost_idx(seq);
2342                 return 0;
2343         }
2344
2345         if (!fid_seq_in_fldb(seq)) {
2346                 fld_range_set_mdt(range);
2347                 if (ss != NULL)
2348                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2349                          * at all */
2350                         range->lsr_index = ss->ss_node_id;
2351                 return 0;
2352         }
2353
2354         LASSERT(ss != NULL);
2355         fld_range_set_any(range);
2356         /* OSD will only do local fld lookup */
2357         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2358 }
2359
2360 /*
2361  * Concurrency: no external locking is necessary.
2362  */
2363 static int osd_declare_object_create(const struct lu_env *env,
2364                                      struct dt_object *dt,
2365                                      struct lu_attr *attr,
2366                                      struct dt_allocation_hint *hint,
2367                                      struct dt_object_format *dof,
2368                                      struct thandle *handle)
2369 {
2370         struct osd_thandle      *oh;
2371         int                      rc;
2372         ENTRY;
2373
2374         LASSERT(handle != NULL);
2375
2376         oh = container_of0(handle, struct osd_thandle, ot_super);
2377         LASSERT(oh->ot_handle == NULL);
2378
2379         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2380                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2381         /* Reuse idle OI block may cause additional one OI block
2382          * to be changed. */
2383         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2384                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2385
2386         /* If this is directory, then we expect . and .. to be inserted as
2387          * well. The one directory block always needs to be created for the
2388          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2389          * block), there is no danger of needing a tree for the first block.
2390          */
2391         if (attr && S_ISDIR(attr->la_mode)) {
2392                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2393                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2394                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2395         }
2396
2397         if (!attr)
2398                 RETURN(0);
2399
2400         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2401                                    osd_dt_obj(dt), false, NULL, false);
2402         if (rc != 0)
2403                 RETURN(rc);
2404
2405         RETURN(rc);
2406 }
2407
2408 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2409                              struct lu_attr *attr,
2410                              struct dt_allocation_hint *hint,
2411                              struct dt_object_format *dof, struct thandle *th)
2412 {
2413         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2414         struct osd_object       *obj    = osd_dt_obj(dt);
2415         struct osd_thread_info  *info   = osd_oti_get(env);
2416         int result;
2417         ENTRY;
2418
2419         LINVRNT(osd_invariant(obj));
2420         LASSERT(!dt_object_exists(dt));
2421         LASSERT(!dt_object_remote(dt));
2422         LASSERT(osd_write_locked(env, obj));
2423         LASSERT(th != NULL);
2424
2425         if (unlikely(fid_is_acct(fid)))
2426                 /* Quota files can't be created from the kernel any more,
2427                  * 'tune2fs -O quota' will take care of creating them */
2428                 RETURN(-EPERM);
2429
2430         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2431         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2432
2433         result = __osd_object_create(info, obj, attr, hint, dof, th);
2434         if (result == 0)
2435                 result = __osd_oi_insert(env, obj, fid, th);
2436
2437         LASSERT(ergo(result == 0,
2438                      dt_object_exists(dt) && !dt_object_remote(dt)));
2439
2440         LASSERT(osd_invariant(obj));
2441         RETURN(result);
2442 }
2443
2444 /**
2445  * Called to destroy on-disk representation of the object
2446  *
2447  * Concurrency: must be locked
2448  */
2449 static int osd_declare_object_destroy(const struct lu_env *env,
2450                                       struct dt_object *dt,
2451                                       struct thandle *th)
2452 {
2453         struct osd_object  *obj = osd_dt_obj(dt);
2454         struct inode       *inode = obj->oo_inode;
2455         struct osd_thandle *oh;
2456         int                 rc;
2457         ENTRY;
2458
2459         oh = container_of0(th, struct osd_thandle, ot_super);
2460         LASSERT(oh->ot_handle == NULL);
2461         LASSERT(inode);
2462
2463         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2464                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2465         /* Recycle idle OI leaf may cause additional three OI blocks
2466          * to be changed. */
2467         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2468                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2469         /* one less inode */
2470         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2471                                    -1, oh, obj, false, NULL, false);
2472         if (rc)
2473                 RETURN(rc);
2474         /* data to be truncated */
2475         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2476                                    0, oh, obj, true, NULL, false);
2477         RETURN(rc);
2478 }
2479
2480 static int osd_object_destroy(const struct lu_env *env,
2481                               struct dt_object *dt,
2482                               struct thandle *th)
2483 {
2484         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2485         struct osd_object      *obj = osd_dt_obj(dt);
2486         struct inode           *inode = obj->oo_inode;
2487         struct osd_device      *osd = osd_obj2dev(obj);
2488         struct osd_thandle     *oh;
2489         int                     result;
2490         ENTRY;
2491
2492         oh = container_of0(th, struct osd_thandle, ot_super);
2493         LASSERT(oh->ot_handle);
2494         LASSERT(inode);
2495         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2496
2497         if (unlikely(fid_is_acct(fid)))
2498                 RETURN(-EPERM);
2499
2500         if (S_ISDIR(inode->i_mode)) {
2501                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2502                         inode->i_nlink == 2);
2503                 /* it will check/delete the inode from remote parent,
2504                  * how to optimize it? unlink performance impaction XXX */
2505                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2506                 if (result != 0 && result != -ENOENT) {
2507                         CERROR("%s: delete inode "DFID": rc = %d\n",
2508                                osd_name(osd), PFID(fid), result);
2509                 }
2510                 spin_lock(&obj->oo_guard);
2511                 clear_nlink(inode);
2512                 spin_unlock(&obj->oo_guard);
2513                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2514         }
2515
2516         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2517
2518         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2519                                OI_CHECK_FLD);
2520
2521         /* XXX: add to ext3 orphan list */
2522         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2523
2524         /* not needed in the cache anymore */
2525         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2526
2527         RETURN(0);
2528 }
2529
2530 /**
2531  * Put the fid into lustre_mdt_attrs, and then place the structure
2532  * inode's ea. This fid should not be altered during the life time
2533  * of the inode.
2534  *
2535  * \retval +ve, on success
2536  * \retval -ve, on error
2537  *
2538  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2539  */
2540 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2541                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2542 {
2543         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2544         int                      rc;
2545         ENTRY;
2546
2547         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2548                 RETURN(0);
2549
2550         lustre_lma_init(lma, fid, compat, incompat);
2551         lustre_lma_swab(lma);
2552
2553         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2554                              XATTR_CREATE);
2555         /* LMA may already exist, but we need to check that all the
2556          * desired compat/incompat flags have been added. */
2557         if (unlikely(rc == -EEXIST)) {
2558                 if (compat == 0 && incompat == 0)
2559                         RETURN(0);
2560
2561                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2562                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2563                                      LMA_OLD_SIZE);
2564                 if (rc <= 0)
2565                         RETURN(-EINVAL);
2566
2567                 lustre_lma_swab(lma);
2568                 if (!(~lma->lma_compat & compat) &&
2569                     !(~lma->lma_incompat & incompat))
2570                         RETURN(0);
2571
2572                 lma->lma_compat |= compat;
2573                 lma->lma_incompat |= incompat;
2574                 lustre_lma_swab(lma);
2575                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2576                                      sizeof(*lma), XATTR_REPLACE);
2577         }
2578
2579         RETURN(rc);
2580 }
2581
2582 /**
2583  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2584  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2585  * To have compatilibility with 1.8 ldiskfs driver we need to have
2586  * magic number at start of fid data.
2587  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2588  * its inmemory API.
2589  */
2590 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2591                                          const struct lu_fid *fid)
2592 {
2593         if (!fid_is_namespace_visible(fid) ||
2594             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2595                 param->edp_magic = 0;
2596                 return;
2597         }
2598
2599         param->edp_magic = LDISKFS_LUFID_MAGIC;
2600         param->edp_len =  sizeof(struct lu_fid) + 1;
2601         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2602 }
2603
2604 /**
2605  * Try to read the fid from inode ea into dt_rec.
2606  *
2607  * \param fid object fid.
2608  *
2609  * \retval 0 on success
2610  */
2611 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2612                           __u32 ino, struct lu_fid *fid,
2613                           struct osd_inode_id *id)
2614 {
2615         struct osd_thread_info *info  = osd_oti_get(env);
2616         struct inode           *inode;
2617         ENTRY;
2618
2619         osd_id_gen(id, ino, OSD_OII_NOGEN);
2620         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2621         if (IS_ERR(inode))
2622                 RETURN(PTR_ERR(inode));
2623
2624         iput(inode);
2625         RETURN(0);
2626 }
2627
2628 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2629                                         struct inode *dir,
2630                                         struct inode *parent_dir,
2631                                         const struct lu_fid *dot_fid,
2632                                         const struct lu_fid *dot_dot_fid,
2633                                         struct osd_thandle *oth)
2634 {
2635         struct ldiskfs_dentry_param *dot_ldp;
2636         struct ldiskfs_dentry_param *dot_dot_ldp;
2637
2638         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2639         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2640
2641         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2642         dot_ldp->edp_magic = 0;
2643         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2644                                         dir, dot_ldp, dot_dot_ldp);
2645 }
2646
2647 /**
2648  * Create an local agent inode for remote entry
2649  */
2650 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2651                                                   struct osd_device *osd,
2652                                                   struct osd_object *pobj,
2653                                                   const struct lu_fid *fid,
2654                                                   __u32 type,
2655                                                   struct thandle *th)
2656 {
2657         struct osd_thread_info  *info = osd_oti_get(env);
2658         struct inode            *local;
2659         struct osd_thandle      *oh;
2660         int                     rc;
2661         ENTRY;
2662
2663         LASSERT(th);
2664         oh = container_of(th, struct osd_thandle, ot_super);
2665         LASSERT(oh->ot_handle->h_transaction != NULL);
2666
2667         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2668         if (IS_ERR(local)) {
2669                 CERROR("%s: create local error %d\n", osd_name(osd),
2670                        (int)PTR_ERR(local));
2671                 RETURN(local);
2672         }
2673
2674         /* Set special LMA flag for local agent inode */
2675         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2676         if (rc != 0) {
2677                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2678                        osd_name(osd), PFID(fid), rc);
2679                 RETURN(ERR_PTR(rc));
2680         }
2681
2682         if (!S_ISDIR(type))
2683                 RETURN(local);
2684
2685         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2686                                          lu_object_fid(&pobj->oo_dt.do_lu),
2687                                          fid, oh);
2688         if (rc != 0) {
2689                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2690                         osd_name(osd), PFID(fid), rc);
2691                 RETURN(ERR_PTR(rc));
2692         }
2693
2694         RETURN(local);
2695 }
2696
2697 /**
2698  * Delete local agent inode for remote entry
2699  */
2700 static int osd_delete_local_agent_inode(const struct lu_env *env,
2701                                         struct osd_device *osd,
2702                                         const struct lu_fid *fid,
2703                                         __u32 ino, struct osd_thandle *oh)
2704 {
2705         struct osd_thread_info  *oti = osd_oti_get(env);
2706         struct osd_inode_id     *id = &oti->oti_id;
2707         struct inode            *inode;
2708         ENTRY;
2709
2710         id->oii_ino = le32_to_cpu(ino);
2711         id->oii_gen = OSD_OII_NOGEN;
2712         inode = osd_iget(oti, osd, id);
2713         if (IS_ERR(inode)) {
2714                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2715                        PFID(fid), id->oii_ino, id->oii_gen);
2716                 RETURN(PTR_ERR(inode));
2717         }
2718
2719         clear_nlink(inode);
2720         mark_inode_dirty(inode);
2721         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2722                 osd_name(osd), PFID(fid), inode->i_ino);
2723         iput(inode);
2724         RETURN(0);
2725 }
2726
2727 /**
2728  * OSD layer object create function for interoperability mode (b11826).
2729  * This is mostly similar to osd_object_create(). Only difference being, fid is
2730  * inserted into inode ea here.
2731  *
2732  * \retval   0, on success
2733  * \retval -ve, on error
2734  */
2735 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2736                                 struct lu_attr *attr,
2737                                 struct dt_allocation_hint *hint,
2738                                 struct dt_object_format *dof,
2739                                 struct thandle *th)
2740 {
2741         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2742         struct osd_object       *obj    = osd_dt_obj(dt);
2743         struct osd_thread_info  *info   = osd_oti_get(env);
2744         int                      result;
2745
2746         ENTRY;
2747
2748         LASSERT(osd_invariant(obj));
2749         LASSERT(!dt_object_exists(dt));
2750         LASSERT(!dt_object_remote(dt));
2751         LASSERT(osd_write_locked(env, obj));
2752         LASSERT(th != NULL);
2753
2754         if (unlikely(fid_is_acct(fid)))
2755                 /* Quota files can't be created from the kernel any more,
2756                  * 'tune2fs -O quota' will take care of creating them */
2757                 RETURN(-EPERM);
2758
2759         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2760         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2761
2762         result = __osd_object_create(info, obj, attr, hint, dof, th);
2763         if (result == 0)
2764                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2765                                 fid_is_on_ost(info, osd_obj2dev(obj),
2766                                               fid, OI_CHECK_FLD) ?
2767                                 LMAC_FID_ON_OST : 0, 0);
2768
2769         if (result == 0)
2770                 result = __osd_oi_insert(env, obj, fid, th);
2771
2772         LASSERT(ergo(result == 0,
2773                      dt_object_exists(dt) && !dt_object_remote(dt)));
2774         LINVRNT(osd_invariant(obj));
2775         RETURN(result);
2776 }
2777
2778 static int osd_declare_object_ref_add(const struct lu_env *env,
2779                                       struct dt_object *dt,
2780                                       struct thandle *handle)
2781 {
2782         struct osd_thandle       *oh;
2783
2784         /* it's possible that object doesn't exist yet */
2785         LASSERT(handle != NULL);
2786
2787         oh = container_of0(handle, struct osd_thandle, ot_super);
2788         LASSERT(oh->ot_handle == NULL);
2789
2790         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2791                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2792
2793         return 0;
2794 }
2795
2796 /*
2797  * Concurrency: @dt is write locked.
2798  */
2799 static int osd_object_ref_add(const struct lu_env *env,
2800                               struct dt_object *dt, struct thandle *th)
2801 {
2802         struct osd_object  *obj = osd_dt_obj(dt);
2803         struct inode       *inode = obj->oo_inode;
2804         struct osd_thandle *oh;
2805         int                 rc = 0;
2806
2807         LINVRNT(osd_invariant(obj));
2808         LASSERT(dt_object_exists(dt));
2809         LASSERT(!dt_object_remote(dt));
2810         LASSERT(osd_write_locked(env, obj));
2811         LASSERT(th != NULL);
2812
2813         oh = container_of0(th, struct osd_thandle, ot_super);
2814         LASSERT(oh->ot_handle != NULL);
2815
2816         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2817
2818         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2819                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2820         /*
2821          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2822          * (65000) subdirectories by storing "1" in i_nlink if the link count
2823          * would otherwise overflow. Directory tranversal tools understand
2824          * that (st_nlink == 1) indicates that the filesystem dose not track
2825          * hard links count on the directory, and will not abort subdirectory
2826          * scanning early once (st_nlink - 2) subdirs have been found.
2827          *
2828          * This also has to properly handle the case of inodes with nlink == 0
2829          * in case they are being linked into the PENDING directory
2830          */
2831         spin_lock(&obj->oo_guard);
2832         if (unlikely(inode->i_nlink == 0))
2833                 /* inc_nlink from 0 may cause WARN_ON */
2834                 set_nlink(inode, 1);
2835         else {
2836                 ldiskfs_inc_count(oh->ot_handle, inode);
2837                 if (!S_ISDIR(inode->i_mode))
2838                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2839         }
2840         spin_unlock(&obj->oo_guard);
2841
2842         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2843         LINVRNT(osd_invariant(obj));
2844
2845         return rc;
2846 }
2847
2848 static int osd_declare_object_ref_del(const struct lu_env *env,
2849                                       struct dt_object *dt,
2850                                       struct thandle *handle)
2851 {
2852         struct osd_thandle *oh;
2853
2854         LASSERT(dt_object_exists(dt));
2855         LASSERT(!dt_object_remote(dt));
2856         LASSERT(handle != NULL);
2857
2858         oh = container_of0(handle, struct osd_thandle, ot_super);
2859         LASSERT(oh->ot_handle == NULL);
2860
2861         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2862                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2863
2864         return 0;
2865 }
2866
2867 /*
2868  * Concurrency: @dt is write locked.
2869  */
2870 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2871                               struct thandle *th)
2872 {
2873         struct osd_object       *obj = osd_dt_obj(dt);
2874         struct inode            *inode = obj->oo_inode;
2875         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2876         struct osd_thandle      *oh;
2877
2878         LINVRNT(osd_invariant(obj));
2879         LASSERT(dt_object_exists(dt));
2880         LASSERT(!dt_object_remote(dt));
2881         LASSERT(osd_write_locked(env, obj));
2882         LASSERT(th != NULL);
2883
2884         oh = container_of0(th, struct osd_thandle, ot_super);
2885         LASSERT(oh->ot_handle != NULL);
2886
2887         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2888
2889         spin_lock(&obj->oo_guard);
2890         /* That can be result of upgrade from old Lustre version and
2891          * applied only to local files.  Just skip this ref_del call.
2892          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2893         if (inode->i_nlink == 0) {
2894                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2895                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2896                              ", maybe an upgraded file? (LU-3915)\n",
2897                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2898                 spin_unlock(&obj->oo_guard);
2899                 return 0;
2900         }
2901
2902         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2903                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2904
2905         ldiskfs_dec_count(oh->ot_handle, inode);
2906         spin_unlock(&obj->oo_guard);
2907
2908         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2909         LINVRNT(osd_invariant(obj));
2910
2911         return 0;
2912 }
2913
2914 /*
2915  * Get the 64-bit version for an inode.
2916  */
2917 static int osd_object_version_get(const struct lu_env *env,
2918                                   struct dt_object *dt, dt_obj_version_t *ver)
2919 {
2920         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2921
2922         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2923                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2924         *ver = LDISKFS_I(inode)->i_fs_version;
2925         return 0;
2926 }
2927
2928 /*
2929  * Concurrency: @dt is read locked.
2930  */
2931 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2932                          struct lu_buf *buf, const char *name,
2933                          struct lustre_capa *capa)
2934 {
2935         struct osd_object      *obj    = osd_dt_obj(dt);
2936         struct inode           *inode  = obj->oo_inode;
2937         struct osd_thread_info *info   = osd_oti_get(env);
2938         struct dentry          *dentry = &info->oti_obj_dentry;
2939
2940         /* version get is not real XATTR but uses xattr API */
2941         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2942                 /* for version we are just using xattr API but change inode
2943                  * field instead */
2944                 if (buf->lb_len == 0)
2945                         return sizeof(dt_obj_version_t);
2946
2947                 if (buf->lb_len < sizeof(dt_obj_version_t))
2948                         return -ERANGE;
2949
2950                 osd_object_version_get(env, dt, buf->lb_buf);
2951
2952                 return sizeof(dt_obj_version_t);
2953         }
2954
2955         LASSERT(dt_object_exists(dt));
2956         LASSERT(!dt_object_remote(dt));
2957         LASSERT(inode->i_op != NULL);
2958         LASSERT(inode->i_op->getxattr != NULL);
2959
2960         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2961                 return -EACCES;
2962
2963         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2964 }
2965
2966
2967 static int osd_declare_xattr_set(const struct lu_env *env,
2968                                  struct dt_object *dt,
2969                                  const struct lu_buf *buf, const char *name,
2970                                  int fl, struct thandle *handle)
2971 {
2972         struct osd_thandle *oh;
2973         int credits;
2974         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2975
2976         LASSERT(handle != NULL);
2977
2978         oh = container_of0(handle, struct osd_thandle, ot_super);
2979         LASSERT(oh->ot_handle == NULL);
2980
2981         /* optimistic optimization: LMA is set first and usually fit inode */
2982         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2983                 if (dt_object_exists(dt))
2984                         credits = 0;
2985                 else
2986                         credits = 1;
2987         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2988                 credits = 1;
2989         } else {
2990                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2991                 if (buf && buf->lb_len > sb->s_blocksize) {
2992                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2993                                         sb->s_blocksize_bits;
2994                 }
2995                 /*
2996                  * xattr set may involve inode quota change, reserve credits for
2997                  * dquot_initialize()
2998                  */
2999                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3000         }
3001
3002         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3003
3004         return 0;
3005 }
3006
3007 /*
3008  * Set the 64-bit version for object
3009  */
3010 static void osd_object_version_set(const struct lu_env *env,
3011                                    struct dt_object *dt,
3012                                    dt_obj_version_t *new_version)
3013 {
3014         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3015
3016         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3017                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3018
3019         LDISKFS_I(inode)->i_fs_version = *new_version;
3020         /** Version is set after all inode operations are finished,
3021          *  so we should mark it dirty here */
3022         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3023 }
3024
3025 /*
3026  * Concurrency: @dt is write locked.
3027  */
3028 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3029                          const struct lu_buf *buf, const char *name, int fl,
3030                          struct thandle *handle, struct lustre_capa *capa)
3031 {
3032         struct osd_object      *obj      = osd_dt_obj(dt);
3033         struct inode           *inode    = obj->oo_inode;
3034         struct osd_thread_info *info     = osd_oti_get(env);
3035         int                     fs_flags = 0;
3036         ENTRY;
3037
3038         LASSERT(handle != NULL);
3039
3040         /* version set is not real XATTR */
3041         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3042                 /* for version we are just using xattr API but change inode
3043                  * field instead */
3044                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3045                 osd_object_version_set(env, dt, buf->lb_buf);
3046                 return sizeof(dt_obj_version_t);
3047         }
3048
3049         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3050                 return -EACCES;
3051
3052         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3053                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3054
3055         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3056         if (fl & LU_XATTR_REPLACE)
3057                 fs_flags |= XATTR_REPLACE;
3058
3059         if (fl & LU_XATTR_CREATE)
3060                 fs_flags |= XATTR_CREATE;
3061
3062         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3063                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3064                 int                      rc;
3065
3066                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3067                 if (rc != 0)
3068                         RETURN(rc);
3069
3070                 lma->lma_incompat |= LMAI_STRIPED;
3071                 lustre_lma_swab(lma);
3072                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3073                                      sizeof(*lma), XATTR_REPLACE);
3074                 if (rc != 0)
3075                         RETURN(rc);
3076         }
3077
3078         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3079             strcmp(name, XATTR_NAME_LINK) == 0)
3080                 return -ENOSPC;
3081
3082         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3083                                fs_flags);
3084 }
3085
3086 /*
3087  * Concurrency: @dt is read locked.
3088  */
3089 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3090                           const struct lu_buf *buf, struct lustre_capa *capa)
3091 {
3092         struct osd_object      *obj    = osd_dt_obj(dt);
3093         struct inode           *inode  = obj->oo_inode;
3094         struct osd_thread_info *info   = osd_oti_get(env);
3095         struct dentry          *dentry = &info->oti_obj_dentry;
3096
3097         LASSERT(dt_object_exists(dt));
3098         LASSERT(!dt_object_remote(dt));
3099         LASSERT(inode->i_op != NULL);
3100         LASSERT(inode->i_op->listxattr != NULL);
3101
3102         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3103                 return -EACCES;
3104
3105         dentry->d_inode = inode;
3106         dentry->d_sb = inode->i_sb;
3107         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3108 }
3109
3110 static int osd_declare_xattr_del(const struct lu_env *env,
3111                                  struct dt_object *dt, const char *name,
3112                                  struct thandle *handle)
3113 {
3114         struct osd_thandle *oh;
3115         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3116
3117         LASSERT(dt_object_exists(dt));
3118         LASSERT(!dt_object_remote(dt));
3119         LASSERT(handle != NULL);
3120
3121         oh = container_of0(handle, struct osd_thandle, ot_super);
3122         LASSERT(oh->ot_handle == NULL);
3123
3124         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3125                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3126         /*
3127          * xattr del may involve inode quota change, reserve credits for
3128          * dquot_initialize()
3129          */
3130         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3131
3132         return 0;
3133 }
3134
3135 /*
3136  * Concurrency: @dt is write locked.
3137  */
3138 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3139                          const char *name, struct thandle *handle,
3140                          struct lustre_capa *capa)
3141 {
3142         struct osd_object      *obj    = osd_dt_obj(dt);
3143         struct inode           *inode  = obj->oo_inode;
3144         struct osd_thread_info *info   = osd_oti_get(env);
3145         struct dentry          *dentry = &info->oti_obj_dentry;
3146         int                     rc;
3147
3148         LASSERT(dt_object_exists(dt));
3149         LASSERT(!dt_object_remote(dt));
3150         LASSERT(inode->i_op != NULL);
3151         LASSERT(inode->i_op->removexattr != NULL);
3152         LASSERT(handle != NULL);
3153
3154         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3155                 return -EACCES;
3156
3157         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3158
3159         ll_vfs_dq_init(inode);
3160         dentry->d_inode = inode;
3161         dentry->d_sb = inode->i_sb;
3162         rc = inode->i_op->removexattr(dentry, name);
3163         return rc;
3164 }
3165
3166 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3167                                      struct dt_object *dt,
3168                                      struct lustre_capa *old, __u64 opc)
3169 {
3170         struct osd_thread_info *info = osd_oti_get(env);
3171         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3172         struct osd_object *obj = osd_dt_obj(dt);
3173         struct osd_device *osd = osd_obj2dev(obj);
3174         struct lustre_capa_key *key = &info->oti_capa_key;
3175         struct lustre_capa *capa = &info->oti_capa;
3176         struct obd_capa *oc;
3177         struct lu_capainfo *lci;
3178         int rc;
3179         ENTRY;
3180
3181         if (!osd->od_fl_capa)
3182                 RETURN(ERR_PTR(-ENOENT));
3183
3184         LASSERT(dt_object_exists(dt));
3185         LASSERT(!dt_object_remote(dt));
3186         LINVRNT(osd_invariant(obj));
3187
3188         /* renewal sanity check */
3189         if (old && osd_object_auth(env, dt, old, opc))
3190                 RETURN(ERR_PTR(-EACCES));
3191
3192         lci = lu_capainfo_get(env);
3193         if (unlikely(lci == NULL))
3194                 RETURN(ERR_PTR(-ENOENT));
3195
3196         switch (lci->lci_auth) {
3197         case LC_ID_NONE:
3198                 RETURN(NULL);
3199         case LC_ID_PLAIN:
3200                 capa->lc_uid = i_uid_read(obj->oo_inode);
3201                 capa->lc_gid = i_gid_read(obj->oo_inode);
3202                 capa->lc_flags = LC_ID_PLAIN;
3203                 break;
3204         case LC_ID_CONVERT: {
3205                 __u32 d[4], s[4];
3206
3207                 s[0] = i_uid_read(obj->oo_inode);
3208                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3209                 s[2] = i_uid_read(obj->oo_inode);
3210                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3211                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3212                 if (unlikely(rc))
3213                         RETURN(ERR_PTR(rc));
3214
3215                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3216                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3217                 capa->lc_flags = LC_ID_CONVERT;
3218                 break;
3219         }
3220         default:
3221                 RETURN(ERR_PTR(-EINVAL));
3222         }
3223
3224         capa->lc_fid = *fid;
3225         capa->lc_opc = opc;
3226         capa->lc_flags |= osd->od_capa_alg << 24;
3227         capa->lc_timeout = osd->od_capa_timeout;
3228         capa->lc_expiry = 0;
3229
3230         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3231         if (oc) {
3232                 LASSERT(!capa_is_expired(oc));
3233                 RETURN(oc);
3234         }
3235
3236         spin_lock(&capa_lock);
3237         *key = osd->od_capa_keys[1];
3238         spin_unlock(&capa_lock);
3239
3240         capa->lc_keyid = key->lk_keyid;
3241         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3242
3243         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3244         if (rc) {
3245                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3246                 RETURN(ERR_PTR(rc));
3247         }
3248
3249         oc = capa_add(osd->od_capa_hash, capa);
3250         RETURN(oc);
3251 }
3252
3253 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3254                            __u64 start, __u64 end)
3255 {
3256         struct osd_object       *obj    = osd_dt_obj(dt);
3257         struct inode            *inode  = obj->oo_inode;
3258         struct osd_thread_info  *info   = osd_oti_get(env);
3259         struct dentry           *dentry = &info->oti_obj_dentry;
3260         struct file             *file   = &info->oti_file;
3261         int                     rc;
3262
3263         ENTRY;
3264
3265         dentry->d_inode = inode;
3266         dentry->d_sb = inode->i_sb;
3267         file->f_dentry = dentry;
3268         file->f_mapping = inode->i_mapping;
3269         file->f_op = inode->i_fop;
3270         set_file_inode(file, inode);
3271
3272         rc = ll_vfs_fsync_range(file, start, end, 0);
3273
3274         RETURN(rc);
3275 }
3276
3277 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3278                         void **data)
3279 {
3280         struct osd_object *obj = osd_dt_obj(dt);
3281         ENTRY;
3282
3283         *data = (void *)obj->oo_inode;
3284         RETURN(0);
3285 }
3286
3287 /*
3288  * Index operations.
3289  */
3290
3291 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3292                            const struct dt_index_features *feat)
3293 {
3294         struct iam_descr *descr;
3295
3296         if (osd_object_is_root(o))
3297                 return feat == &dt_directory_features;
3298
3299         LASSERT(o->oo_dir != NULL);
3300
3301         descr = o->oo_dir->od_container.ic_descr;
3302         if (feat == &dt_directory_features) {
3303                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3304                         return 1;
3305                 else
3306                         return 0;
3307         } else {
3308                 return
3309                         feat->dif_keysize_min <= descr->id_key_size &&
3310                         descr->id_key_size <= feat->dif_keysize_max &&
3311                         feat->dif_recsize_min <= descr->id_rec_size &&
3312                         descr->id_rec_size <= feat->dif_recsize_max &&
3313                         !(feat->dif_flags & (DT_IND_VARKEY |
3314                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3315                         ergo(feat->dif_flags & DT_IND_UPDATE,
3316                              1 /* XXX check that object (and file system) is
3317                                 * writable */);
3318         }
3319 }
3320
3321 static int osd_iam_container_init(const struct lu_env *env,
3322                                   struct osd_object *obj,
3323                                   struct osd_directory *dir)
3324 {
3325         struct iam_container *bag = &dir->od_container;
3326         int result;
3327
3328         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3329         if (result != 0)
3330                 return result;
3331
3332         result = iam_container_setup(bag);
3333         if (result == 0)
3334                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3335         else
3336                 iam_container_fini(bag);
3337
3338         return result;
3339 }
3340
3341
3342 /*
3343  * Concurrency: no external locking is necessary.
3344  */
3345 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3346                          const struct dt_index_features *feat)
3347 {
3348         int                      result;
3349         int                      skip_iam = 0;
3350         struct osd_object       *obj = osd_dt_obj(dt);
3351
3352         LINVRNT(osd_invariant(obj));
3353
3354         if (osd_object_is_root(obj)) {
3355                 dt->do_index_ops = &osd_index_ea_ops;
3356                 result = 0;
3357         } else if (feat == &dt_directory_features) {
3358                 dt->do_index_ops = &osd_index_ea_ops;
3359                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3360                         result = 0;
3361                 else
3362                         result = -ENOTDIR;
3363                 skip_iam = 1;
3364         } else if (unlikely(feat == &dt_otable_features)) {
3365                 dt->do_index_ops = &osd_otable_ops;
3366                 return 0;
3367         } else if (unlikely(feat == &dt_acct_features)) {
3368                 dt->do_index_ops = &osd_acct_index_ops;
3369                 result = 0;
3370                 skip_iam = 1;
3371         } else if (!osd_has_index(obj)) {
3372                 struct osd_directory *dir;
3373
3374                 OBD_ALLOC_PTR(dir);
3375                 if (dir != NULL) {
3376
3377                         spin_lock(&obj->oo_guard);
3378                         if (obj->oo_dir == NULL)
3379                                 obj->oo_dir = dir;
3380                         else
3381                                 /*
3382                                  * Concurrent thread allocated container data.
3383                                  */
3384                                 OBD_FREE_PTR(dir);
3385                         spin_unlock(&obj->oo_guard);
3386                         /*
3387                          * Now, that we have container data, serialize its
3388                          * initialization.
3389                          */
3390                         down_write(&obj->oo_ext_idx_sem);
3391                         /*
3392                          * recheck under lock.
3393                          */
3394                         if (!osd_has_index(obj))
3395                                 result = osd_iam_container_init(env, obj,
3396                                                                 obj->oo_dir);
3397                         else
3398                                 result = 0;
3399                         up_write(&obj->oo_ext_idx_sem);
3400                 } else {
3401                         result = -ENOMEM;
3402                 }
3403         } else {
3404                 result = 0;
3405         }
3406
3407         if (result == 0 && skip_iam == 0) {
3408                 if (!osd_iam_index_probe(env, obj, feat))
3409                         result = -ENOTDIR;
3410         }
3411         LINVRNT(osd_invariant(obj));
3412
3413         if (result == 0 && is_quota_glb_feat(feat) &&
3414             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3415                 result = osd_quota_migration(env, dt, feat);
3416
3417         return result;
3418 }
3419
3420 static int osd_otable_it_attr_get(const struct lu_env *env,
3421                                  struct dt_object *dt,
3422                                  struct lu_attr *attr,
3423                                  struct lustre_capa *capa)
3424 {
3425         attr->la_valid = 0;
3426         return 0;
3427 }
3428
3429 static const struct dt_object_operations osd_obj_ops = {
3430         .do_read_lock         = osd_object_read_lock,
3431         .do_write_lock        = osd_object_write_lock,
3432         .do_read_unlock       = osd_object_read_unlock,
3433         .do_write_unlock      = osd_object_write_unlock,
3434         .do_write_locked      = osd_object_write_locked,
3435         .do_attr_get          = osd_attr_get,
3436         .do_declare_attr_set  = osd_declare_attr_set,
3437         .do_attr_set          = osd_attr_set,
3438         .do_ah_init           = osd_ah_init,
3439         .do_declare_create    = osd_declare_object_create,
3440         .do_create            = osd_object_create,
3441         .do_declare_destroy   = osd_declare_object_destroy,
3442         .do_destroy           = osd_object_destroy,
3443         .do_index_try         = osd_index_try,
3444         .do_declare_ref_add   = osd_declare_object_ref_add,
3445         .do_ref_add           = osd_object_ref_add,
3446         .do_declare_ref_del   = osd_declare_object_ref_del,
3447         .do_ref_del           = osd_object_ref_del,
3448         .do_xattr_get         = osd_xattr_get,
3449         .do_declare_xattr_set = osd_declare_xattr_set,
3450         .do_xattr_set         = osd_xattr_set,
3451         .do_declare_xattr_del = osd_declare_xattr_del,
3452         .do_xattr_del         = osd_xattr_del,
3453         .do_xattr_list        = osd_xattr_list,
3454         .do_capa_get          = osd_capa_get,
3455         .do_object_sync       = osd_object_sync,
3456         .do_data_get          = osd_data_get,
3457 };
3458
3459 /**
3460  * dt_object_operations for interoperability mode
3461  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3462  */
3463 static const struct dt_object_operations osd_obj_ea_ops = {
3464         .do_read_lock         = osd_object_read_lock,
3465         .do_write_lock        = osd_object_write_lock,
3466         .do_read_unlock       = osd_object_read_unlock,
3467         .do_write_unlock      = osd_object_write_unlock,
3468         .do_write_locked      = osd_object_write_locked,
3469         .do_attr_get          = osd_attr_get,
3470         .do_declare_attr_set  = osd_declare_attr_set,
3471         .do_attr_set          = osd_attr_set,
3472         .do_ah_init           = osd_ah_init,
3473         .do_declare_create    = osd_declare_object_create,
3474         .do_create            = osd_object_ea_create,
3475         .do_declare_destroy   = osd_declare_object_destroy,
3476         .do_destroy           = osd_object_destroy,
3477         .do_index_try         = osd_index_try,
3478         .do_declare_ref_add   = osd_declare_object_ref_add,
3479         .do_ref_add           = osd_object_ref_add,
3480         .do_declare_ref_del   = osd_declare_object_ref_del,
3481         .do_ref_del           = osd_object_ref_del,
3482         .do_xattr_get         = osd_xattr_get,
3483         .do_declare_xattr_set = osd_declare_xattr_set,
3484         .do_xattr_set         = osd_xattr_set,
3485         .do_declare_xattr_del = osd_declare_xattr_del,
3486         .do_xattr_del         = osd_xattr_del,
3487         .do_xattr_list        = osd_xattr_list,
3488         .do_capa_get          = osd_capa_get,
3489         .do_object_sync       = osd_object_sync,
3490         .do_data_get          = osd_data_get,
3491 };
3492
3493 static const struct dt_object_operations osd_obj_otable_it_ops = {
3494         .do_attr_get    = osd_otable_it_attr_get,
3495         .do_index_try   = osd_index_try,
3496 };
3497
3498 static int osd_index_declare_iam_delete(const struct lu_env *env,
3499                                         struct dt_object *dt,
3500                                         const struct dt_key *key,
3501                                         struct thandle *handle)
3502 {
3503         struct osd_thandle    *oh;
3504
3505         oh = container_of0(handle, struct osd_thandle, ot_super);
3506         LASSERT(oh->ot_handle == NULL);
3507
3508         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3509                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3510
3511         return 0;
3512 }
3513
3514 /**
3515  *      delete a (key, value) pair from index \a dt specified by \a key
3516  *
3517  *      \param  dt      osd index object
3518  *      \param  key     key for index
3519  *      \param  rec     record reference
3520  *      \param  handle  transaction handler
3521  *
3522  *      \retval  0  success
3523  *      \retval -ve   failure
3524  */
3525 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3526                                 const struct dt_key *key,
3527                                 struct thandle *handle,
3528                                 struct lustre_capa *capa)
3529 {
3530         struct osd_thread_info *oti = osd_oti_get(env);
3531         struct osd_object      *obj = osd_dt_obj(dt);
3532         struct osd_thandle     *oh;
3533         struct iam_path_descr  *ipd;
3534         struct iam_container   *bag = &obj->oo_dir->od_container;
3535         int                     rc;
3536         ENTRY;
3537
3538         LINVRNT(osd_invariant(obj));
3539         LASSERT(dt_object_exists(dt));
3540         LASSERT(!dt_object_remote(dt));
3541         LASSERT(bag->ic_object == obj->oo_inode);
3542         LASSERT(handle != NULL);
3543
3544         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3545                 RETURN(-EACCES);
3546
3547         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3548
3549         ipd = osd_idx_ipd_get(env, bag);
3550         if (unlikely(ipd == NULL))
3551                 RETURN(-ENOMEM);
3552
3553         oh = container_of0(handle, struct osd_thandle, ot_super);
3554         LASSERT(oh->ot_handle != NULL);
3555         LASSERT(oh->ot_handle->h_transaction != NULL);
3556
3557         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3558                 /* swab quota uid/gid provided by caller */
3559                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3560                 key = (const struct dt_key *)&oti->oti_quota_id;
3561         }
3562
3563         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3564         osd_ipd_put(env, bag, ipd);
3565         LINVRNT(osd_invariant(obj));
3566         RETURN(rc);
3567 }
3568
3569 static int osd_index_declare_ea_delete(const struct lu_env *env,
3570                                        struct dt_object *dt,
3571                                        const struct dt_key *key,
3572                                        struct thandle *handle)
3573 {
3574         struct osd_thandle *oh;
3575         struct inode       *inode;
3576         int                 rc;
3577         ENTRY;
3578
3579         LASSERT(dt_object_exists(dt));
3580         LASSERT(!dt_object_remote(dt));
3581         LASSERT(handle != NULL);
3582
3583         oh = container_of0(handle, struct osd_thandle, ot_super);
3584         LASSERT(oh->ot_handle == NULL);
3585
3586         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3587                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3588
3589         inode = osd_dt_obj(dt)->oo_inode;
3590         LASSERT(inode);
3591
3592         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3593                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3594         RETURN(rc);
3595 }
3596
3597 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3598                                           struct dt_rec *fid)
3599 {
3600         struct osd_fid_pack *rec;
3601         int                  rc = -ENODATA;
3602
3603         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3604                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3605                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3606         }
3607         return rc;
3608 }
3609
3610 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3611                           const struct lu_fid *fid)
3612 {
3613         struct seq_server_site  *ss = osd_seq_site(osd);
3614         ENTRY;
3615
3616         /* FID seqs not in FLDB, must be local seq */
3617         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3618                 RETURN(0);
3619
3620         /* If FLD is not being initialized yet, it only happens during the
3621          * initialization, likely during mgs initialization, and we assume
3622          * this is local FID. */
3623         if (ss == NULL || ss->ss_server_fld == NULL)
3624                 RETURN(0);
3625
3626         /* Only check the local FLDB here */
3627         if (osd_seq_exists(env, osd, fid_seq(fid)))
3628                 RETURN(0);
3629
3630         RETURN(1);
3631 }
3632
3633 /**
3634  * Index delete function for interoperability mode (b11826).
3635  * It will remove the directory entry added by osd_index_ea_insert().
3636  * This entry is needed to maintain name->fid mapping.
3637  *
3638  * \param key,  key i.e. file entry to be deleted
3639  *
3640  * \retval   0, on success
3641  * \retval -ve, on error
3642  */
3643 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3644                                const struct dt_key *key, struct thandle *handle,
3645                                struct lustre_capa *capa)
3646 {
3647         struct osd_object          *obj = osd_dt_obj(dt);
3648         struct inode               *dir = obj->oo_inode;
3649         struct dentry              *dentry;
3650         struct osd_thandle         *oh;
3651         struct ldiskfs_dir_entry_2 *de = NULL;
3652         struct buffer_head         *bh;
3653         struct htree_lock          *hlock = NULL;
3654         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3655         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3656         int                        rc;
3657         ENTRY;
3658
3659         LINVRNT(osd_invariant(obj));
3660         LASSERT(dt_object_exists(dt));
3661         LASSERT(!dt_object_remote(dt));
3662         LASSERT(handle != NULL);
3663
3664         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3665
3666         oh = container_of(handle, struct osd_thandle, ot_super);
3667         LASSERT(oh->ot_handle != NULL);
3668         LASSERT(oh->ot_handle->h_transaction != NULL);
3669
3670         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3671                 RETURN(-EACCES);
3672
3673         ll_vfs_dq_init(dir);
3674         dentry = osd_child_dentry_get(env, obj,
3675                                       (char *)key, strlen((char *)key));
3676
3677         if (obj->oo_hl_head != NULL) {
3678                 hlock = osd_oti_get(env)->oti_hlock;
3679                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3680                                    dir, LDISKFS_HLOCK_DEL);
3681         } else {
3682                 down_write(&obj->oo_ext_idx_sem);
3683         }
3684
3685         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3686         if (bh) {
3687                 __u32 ino = 0;
3688
3689                 /* If this is not the ".." entry, it might be a remote DNE
3690                  * entry and  we need to check if the FID is for a remote
3691                  * MDT.  If the FID is  not in the directory entry (e.g.
3692                  * upgraded 1.8 filesystem without dirdata enabled) then
3693                  * we need to get the FID from the LMA. For a remote directory
3694                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3695                  * case.
3696                  *
3697                  * Delete the entry before the agent inode in order to
3698                  * simplify error handling.  At worst an error after deleting
3699                  * the entry first might leak the agent inode afterward. The
3700                  * reverse would need filesystem abort in case of error deleting
3701                  * the entry after the agent had been removed, or leave a
3702                  * dangling entry pointing at a random inode. */
3703                 if (strcmp((char *)key, dotdot) != 0) {
3704                         LASSERT(de != NULL);
3705                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3706                         /* If Fid is not in dentry, try to get it from LMA */
3707                         if (rc == -ENODATA) {
3708                                 struct osd_inode_id *id;
3709                                 struct inode *inode;
3710
3711                                 /* Before trying to get fid from the inode,
3712                                  * check whether the inode is valid.
3713                                  *
3714                                  * If the inode has been deleted, do not go
3715                                  * ahead to do osd_ea_fid_get, which will set
3716                                  * the inode to bad inode, which might cause
3717                                  * the inode to be deleted uncorrectly */
3718                                 inode = ldiskfs_iget(osd_sb(osd),
3719                                                      le32_to_cpu(de->inode));
3720                                 if (IS_ERR(inode)) {
3721                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3722                                                "error.\n", osd_name(osd),
3723                                                PFID(fid));
3724                                         rc = PTR_ERR(inode);
3725                                 } else {
3726                                         if (likely(inode->i_nlink != 0)) {
3727                                                 id = &osd_oti_get(env)->oti_id;
3728                                                 rc = osd_ea_fid_get(env, obj,
3729                                                         le32_to_cpu(de->inode),
3730                                                                     fid, id);
3731                                         } else {
3732                                                 CDEBUG(D_INFO, "%s: %u "DFID
3733                                                        "deleted.\n",
3734                                                        osd_name(osd),
3735                                                        le32_to_cpu(de->inode),
3736                                                        PFID(fid));
3737                                                 rc = -ESTALE;
3738                                         }
3739                                         iput(inode);
3740                                 }
3741                         }
3742                         if (rc == 0 &&
3743                             unlikely(osd_remote_fid(env, osd, fid)))
3744                                 /* Need to delete agent inode */
3745                                 ino = le32_to_cpu(de->inode);
3746                 }
3747                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3748                 brelse(bh);
3749                 if (rc == 0 && unlikely(ino != 0)) {
3750                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3751                                                           oh);
3752                         if (rc != 0)
3753                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3754                                        osd_name(osd), PFID(fid), rc);
3755                 }
3756         } else {
3757                 rc = -ENOENT;
3758         }
3759         if (hlock != NULL)
3760                 ldiskfs_htree_unlock(hlock);
3761         else
3762                 up_write(&obj->oo_ext_idx_sem);
3763
3764         if (rc != 0)
3765                 GOTO(out, rc);
3766
3767         /* For inode on the remote MDT, .. will point to
3768          * /Agent directory, Check whether it needs to delete
3769          * from agent directory */
3770         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3771                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3772                                                    oh);
3773                 if (rc != 0 && rc != -ENOENT) {
3774                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3775                                osd_name(osd), PFID(fid), rc);
3776                 }
3777
3778                 if (rc == -ENOENT)
3779                         rc = 0;
3780
3781                 GOTO(out, rc);
3782         }
3783 out:
3784
3785         LASSERT(osd_invariant(obj));
3786         RETURN(rc);
3787 }
3788
3789 /**
3790  *      Lookup index for \a key and copy record to \a rec.
3791  *
3792  *      \param  dt      osd index object
3793  *      \param  key     key for index
3794  *      \param  rec     record reference
3795  *
3796  *      \retval  +ve  success : exact mach
3797  *      \retval  0    return record with key not greater than \a key
3798  *      \retval -ve   failure
3799  */
3800 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3801                                 struct dt_rec *rec, const struct dt_key *key,
3802                                 struct lustre_capa *capa)
3803 {
3804         struct osd_object      *obj = osd_dt_obj(dt);
3805         struct iam_path_descr  *ipd;
3806         struct iam_container   *bag = &obj->oo_dir->od_container;
3807         struct osd_thread_info *oti = osd_oti_get(env);
3808         struct iam_iterator    *it = &oti->oti_idx_it;
3809         struct iam_rec         *iam_rec;
3810         int                     rc;
3811         ENTRY;
3812
3813         LASSERT(osd_invariant(obj));
3814         LASSERT(dt_object_exists(dt));
3815         LASSERT(!dt_object_remote(dt));
3816         LASSERT(bag->ic_object == obj->oo_inode);
3817
3818         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3819                 RETURN(-EACCES);
3820
3821         ipd = osd_idx_ipd_get(env, bag);
3822         if (IS_ERR(ipd))
3823                 RETURN(-ENOMEM);
3824
3825         /* got ipd now we can start iterator. */
3826         iam_it_init(it, bag, 0, ipd);
3827
3828         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3829                 /* swab quota uid/gid provided by caller */
3830                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3831                 key = (const struct dt_key *)&oti->oti_quota_id;
3832         }
3833
3834         rc = iam_it_get(it, (struct iam_key *)key);
3835         if (rc >= 0) {
3836                 if (S_ISDIR(obj->oo_inode->i_mode))
3837                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3838                 else
3839                         iam_rec = (struct iam_rec *) rec;
3840
3841                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3842
3843                 if (S_ISDIR(obj->oo_inode->i_mode))
3844                         osd_fid_unpack((struct lu_fid *) rec,
3845                                        (struct osd_fid_pack *)iam_rec);
3846                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3847                         osd_quota_unpack(obj, rec);
3848         }
3849
3850         iam_it_put(it);
3851         iam_it_fini(it);
3852         osd_ipd_put(env, bag, ipd);
3853
3854         LINVRNT(osd_invariant(obj));
3855
3856         RETURN(rc);
3857 }
3858
3859 static int osd_index_declare_iam_insert(const struct lu_env *env,
3860                                         struct dt_object *dt,
3861                                         const struct dt_rec *rec,
3862                                         const struct dt_key *key,
3863                                         struct thandle *handle)
3864 {
3865         struct osd_thandle *oh;
3866
3867         LASSERT(handle != NULL);
3868
3869         oh = container_of0(handle, struct osd_thandle, ot_super);
3870         LASSERT(oh->ot_handle == NULL);
3871
3872         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3873                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3874
3875         return 0;
3876 }
3877
3878 /**
3879  *      Inserts (key, value) pair in \a dt index object.
3880  *
3881  *      \param  dt      osd index object
3882  *      \param  key     key for index
3883  *      \param  rec     record reference
3884  *      \param  th      transaction handler
3885  *
3886  *      \retval  0  success
3887  *      \retval -ve failure
3888  */
3889 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3890                                 const struct dt_rec *rec,
3891                                 const struct dt_key *key, struct thandle *th,
3892                                 struct lustre_capa *capa, int ignore_quota)
3893 {
3894         struct osd_object     *obj = osd_dt_obj(dt);
3895         struct iam_path_descr *ipd;
3896         struct osd_thandle    *oh;
3897         struct iam_container  *bag = &obj->oo_dir->od_container;
3898         struct osd_thread_info *oti = osd_oti_get(env);
3899         struct iam_rec         *iam_rec;
3900         int                     rc;
3901         ENTRY;
3902
3903         LINVRNT(osd_invariant(obj));
3904         LASSERT(dt_object_exists(dt));
3905         LASSERT(!dt_object_remote(dt));
3906         LASSERT(bag->ic_object == obj->oo_inode);
3907         LASSERT(th != NULL);
3908
3909         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3910                 RETURN(-EACCES);
3911
3912         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3913
3914         ipd = osd_idx_ipd_get(env, bag);
3915         if (unlikely(ipd == NULL))
3916                 RETURN(-ENOMEM);
3917
3918         oh = container_of0(th, struct osd_thandle, ot_super);
3919         LASSERT(oh->ot_handle != NULL);
3920         LASSERT(oh->ot_handle->h_transaction != NULL);
3921         if (S_ISDIR(obj->oo_inode->i_mode)) {
3922                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3923                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3924         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3925                 /* pack quota uid/gid */
3926                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3927                 key = (const struct dt_key *)&oti->oti_quota_id;
3928                 /* pack quota record */
3929                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3930                 iam_rec = (struct iam_rec *)rec;
3931         } else {
3932                 iam_rec = (struct iam_rec *)rec;
3933         }
3934
3935         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3936                         iam_rec, ipd);
3937         osd_ipd_put(env, bag, ipd);
3938         LINVRNT(osd_invariant(obj));
3939         RETURN(rc);
3940 }
3941
3942 /**
3943  * Calls ldiskfs_add_entry() to add directory entry
3944  * into the directory. This is required for
3945  * interoperability mode (b11826)
3946  *
3947  * \retval   0, on success
3948  * \retval -ve, on error
3949  */
3950 static int __osd_ea_add_rec(struct osd_thread_info *info,
3951                             struct osd_object *pobj, struct inode  *cinode,
3952                             const char *name, const struct lu_fid *fid,
3953                             struct htree_lock *hlock, struct thandle *th)
3954 {
3955         struct ldiskfs_dentry_param *ldp;
3956         struct dentry               *child;
3957         struct osd_thandle          *oth;
3958         int                          rc;
3959
3960         oth = container_of(th, struct osd_thandle, ot_super);
3961         LASSERT(oth->ot_handle != NULL);
3962         LASSERT(oth->ot_handle->h_transaction != NULL);
3963         LASSERT(pobj->oo_inode);
3964
3965         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3966         if (unlikely(pobj->oo_inode ==
3967                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3968                 ldp->edp_magic = 0;
3969         else
3970                 osd_get_ldiskfs_dirent_param(ldp, fid);
3971         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3972         child->d_fsdata = (void *)ldp;
3973         ll_vfs_dq_init(pobj->oo_inode);
3974         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3975         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3976                 struct ldiskfs_dir_entry_2      *de;
3977                 struct buffer_head              *bh;
3978                 int                              rc1;
3979
3980                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3981                                             NULL, hlock);
3982                 if (bh != NULL) {
3983                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3984                                                                bh);
3985                         if (rc1 == 0) {
3986                                 if (S_ISDIR(cinode->i_mode))
3987                                         de->file_type = LDISKFS_DIRENT_LUFID |
3988                                                         LDISKFS_FT_REG_FILE;
3989                                 else
3990                                         de->file_type = LDISKFS_DIRENT_LUFID |
3991                                                         LDISKFS_FT_DIR;
3992                                 ldiskfs_journal_dirty_metadata(oth->ot_handle,
3993                                                                bh);
3994                                 brelse(bh);
3995                         }
3996                 }
3997         }
3998
3999         RETURN(rc);
4000 }
4001
4002 /**
4003  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4004  * into the directory.Also sets flags into osd object to
4005  * indicate dot and dotdot are created. This is required for
4006  * interoperability mode (b11826)
4007  *
4008  * \param dir   directory for dot and dotdot fixup.
4009  * \param obj   child object for linking
4010  *
4011  * \retval   0, on success
4012  * \retval -ve, on error
4013  */
4014 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4015                               struct osd_object *dir,
4016                               struct inode *parent_dir, const char *name,
4017                               const struct lu_fid *dot_fid,
4018                               const struct lu_fid *dot_dot_fid,
4019                               struct thandle *th)
4020 {
4021         struct inode                *inode = dir->oo_inode;
4022         struct osd_thandle          *oth;
4023         int result = 0;
4024
4025         oth = container_of(th, struct osd_thandle, ot_super);
4026         LASSERT(oth->ot_handle->h_transaction != NULL);
4027         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4028
4029         if (strcmp(name, dot) == 0) {
4030                 if (dir->oo_compat_dot_created) {
4031                         result = -EEXIST;
4032                 } else {
4033                         LASSERT(inode == parent_dir);
4034                         dir->oo_compat_dot_created = 1;
4035                         result = 0;
4036                 }
4037         } else if (strcmp(name, dotdot) == 0) {
4038                 if (!dir->oo_compat_dot_created)
4039                         return -EINVAL;
4040                 /* in case of rename, dotdot is already created */
4041                 if (dir->oo_compat_dotdot_created) {
4042                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4043                                                 dot_dot_fid, NULL, th);
4044                 }
4045
4046                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4047                         struct lu_fid tfid = *dot_dot_fid;
4048
4049                         tfid.f_oid--;
4050                         result = osd_add_dot_dotdot_internal(info,
4051                                         dir->oo_inode, parent_dir, dot_fid,
4052                                         &tfid, oth);
4053                 } else {
4054                         result = osd_add_dot_dotdot_internal(info,
4055                                         dir->oo_inode, parent_dir, dot_fid,
4056                                         dot_dot_fid, oth);
4057                 }
4058
4059                 if (result == 0)
4060                         dir->oo_compat_dotdot_created = 1;
4061         }
4062
4063         return result;
4064 }
4065
4066
4067 /**
4068  * It will call the appropriate osd_add* function and return the
4069  * value, return by respective functions.
4070  */
4071 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4072                           struct inode *cinode, const char *name,
4073                           const struct lu_fid *fid, struct thandle *th)
4074 {
4075         struct osd_thread_info *info   = osd_oti_get(env);
4076         struct htree_lock      *hlock;
4077         int                     rc;
4078
4079         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4080
4081         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4082                                                    name[2] =='\0'))) {
4083                 if (hlock != NULL) {
4084                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4085                                            pobj->oo_inode, 0);
4086                 } else {
4087                         down_write(&pobj->oo_ext_idx_sem);
4088                 }
4089
4090                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4091                                         lu_object_fid(&pobj->oo_dt.do_lu),
4092                                         fid, th);
4093         } else {
4094                 if (hlock != NULL) {
4095                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4096                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4097                 } else {
4098                         down_write(&pobj->oo_ext_idx_sem);
4099                 }
4100
4101                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4102                         struct lu_fid *tfid = &info->oti_fid;
4103
4104                         *tfid = *fid;
4105                         tfid->f_ver = ~0;
4106                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4107                                               tfid, hlock, th);
4108                 } else {
4109                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4110                                               hlock, th);
4111                 }
4112         }
4113         if (hlock != NULL)
4114                 ldiskfs_htree_unlock(hlock);
4115         else
4116                 up_write(&pobj->oo_ext_idx_sem);
4117
4118         return rc;
4119 }
4120
4121 static void
4122 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4123                       struct osd_idmap_cache *oic)
4124 {
4125         struct osd_scrub    *scrub = &dev->od_scrub;
4126         struct lu_fid       *fid   = &oic->oic_fid;
4127         struct osd_inode_id *id    = &oti->oti_id;
4128         int                  once  = 0;
4129         int                  rc;
4130         ENTRY;
4131
4132         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4133                 RETURN_EXIT;
4134
4135         if (scrub->os_pos_current > id->oii_ino)
4136                 RETURN_EXIT;
4137
4138 again:
4139         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4140         if (rc != 0 && rc != -ENOENT)
4141                 RETURN_EXIT;
4142
4143         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4144                 RETURN_EXIT;
4145
4146         if (thread_is_running(&scrub->os_thread)) {
4147                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4148                 /* There is race condition between osd_oi_lookup and OI scrub.
4149                  * The OI scrub finished just after osd_oi_lookup() failure.
4150                  * Under such case, it is unnecessary to trigger OI scrub again,
4151                  * but try to call osd_oi_lookup() again. */
4152                 if (unlikely(rc == -EAGAIN))
4153                         goto again;
4154
4155                 RETURN_EXIT;
4156         }
4157
4158         if (!dev->od_noscrub && ++once == 1) {
4159                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4160                                      SS_CLEAR_FAILOUT);
4161                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4162                               ", rc = %d [2]\n",
4163                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4164                               PFID(fid), rc);
4165                 if (rc == 0 || rc == -EALREADY)
4166                         goto again;
4167         }
4168
4169         EXIT;
4170 }
4171
4172 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4173                                struct osd_device *dev,
4174                                struct osd_idmap_cache *oic,
4175                                struct lu_fid *fid, __u32 ino)
4176 {
4177         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4178         struct inode            *inode;
4179         int                      rc;
4180
4181         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4182         inode = osd_iget(oti, dev, &oic->oic_lid);
4183         if (IS_ERR(inode)) {
4184                 fid_zero(&oic->oic_fid);
4185                 return PTR_ERR(inode);
4186         }
4187
4188         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4189         iput(inode);
4190         if (rc != 0)
4191                 fid_zero(&oic->oic_fid);
4192         else
4193                 *fid = oic->oic_fid = lma->lma_self_fid;
4194         return rc;
4195 }
4196
4197 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4198                       struct osd_inode_id *id, const struct lu_fid *fid)
4199 {
4200         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4201                id->oii_ino, id->oii_gen, info);
4202         info->oti_cache.oic_lid = *id;
4203         info->oti_cache.oic_fid = *fid;
4204         info->oti_cache.oic_dev = osd;
4205 }
4206
4207 /**
4208  * Get parent FID from the linkEA.
4209  *
4210  * For a directory which parent resides on remote MDT, to satisfy the
4211  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4212  * the other hand, to make the lookup(..) on the directory can return
4213  * the real parent FID, we append the real parent FID after its ".."
4214  * name entry in the /REMOTE_PARENT_DIR.
4215  *
4216  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4217  * backup. So after the restore, we cannot get the right parent FID from
4218  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4219  * we have stored the real parent FID in the directory object's linkEA,
4220  * we can parse the linkEA for the real parent FID.
4221  *
4222  * \param[in] env       pointer to the thread context
4223  * \param[in] obj       pointer to the object to be handled
4224  * \param[out]fid       pointer to the buffer to hold the parent FID
4225  *
4226  * \retval              0 for getting the real parent FID successfully
4227  * \retval              negative error number on failure
4228  */
4229 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4230                                     struct osd_object *obj,
4231                                     struct lu_fid *fid)
4232 {
4233         struct osd_thread_info  *oti    = osd_oti_get(env);
4234         struct lu_buf           *buf    = &oti->oti_big_buf;
4235         struct dentry           *dentry = &oti->oti_obj_dentry;
4236         struct inode            *inode  = obj->oo_inode;
4237         struct linkea_data       ldata  = { 0 };
4238         int                      rc;
4239         ENTRY;
4240
4241         fid_zero(fid);
4242         if (!S_ISDIR(inode->i_mode))
4243                 RETURN(-EIO);
4244
4245 again:
4246         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4247                              buf->lb_buf, buf->lb_len);
4248         if (rc == -ERANGE) {
4249                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4250                                      NULL, 0);
4251                 if (rc > 0) {
4252                         lu_buf_realloc(buf, rc);
4253                         if (buf->lb_buf == NULL)
4254                                 RETURN(-ENOMEM);
4255
4256                         goto again;
4257                 }
4258         }
4259
4260         if (unlikely(rc == 0))
4261                 RETURN(-ENODATA);
4262
4263         if (rc < 0)
4264                 RETURN(rc);
4265
4266         if (unlikely(buf->lb_buf == NULL)) {
4267                 lu_buf_realloc(buf, rc);
4268                 if (buf->lb_buf == NULL)
4269                         RETURN(-ENOMEM);
4270
4271                 goto again;
4272         }
4273
4274         ldata.ld_buf = buf;
4275         rc = linkea_init(&ldata);
4276         if (rc == 0) {
4277                 linkea_first_entry(&ldata);
4278                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4279         }
4280
4281         RETURN(rc);
4282 }
4283
4284 /**
4285  * Calls ->lookup() to find dentry. From dentry get inode and
4286  * read inode's ea to get fid. This is required for  interoperability
4287  * mode (b11826)
4288  *
4289  * \retval   0, on success
4290  * \retval -ve, on error
4291  */
4292 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4293                              struct dt_rec *rec, const struct dt_key *key)
4294 {
4295         struct inode                    *dir    = obj->oo_inode;
4296         struct dentry                   *dentry;
4297         struct ldiskfs_dir_entry_2      *de;
4298         struct buffer_head              *bh;
4299         struct lu_fid                   *fid = (struct lu_fid *) rec;
4300         struct htree_lock               *hlock = NULL;
4301         int                             ino;
4302         int                             rc;
4303         ENTRY;
4304
4305         LASSERT(dir->i_op != NULL);
4306         LASSERT(dir->i_op->lookup != NULL);
4307
4308         dentry = osd_child_dentry_get(env, obj,
4309                                       (char *)key, strlen((char *)key));
4310
4311         if (obj->oo_hl_head != NULL) {
4312                 hlock = osd_oti_get(env)->oti_hlock;
4313                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4314                                    dir, LDISKFS_HLOCK_LOOKUP);
4315         } else {
4316                 down_read(&obj->oo_ext_idx_sem);
4317         }
4318
4319         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4320         if (bh) {
4321                 struct osd_thread_info *oti = osd_oti_get(env);
4322                 struct osd_inode_id *id = &oti->oti_id;
4323                 struct osd_idmap_cache *oic = &oti->oti_cache;
4324                 struct osd_device *dev = osd_obj2dev(obj);
4325
4326                 ino = le32_to_cpu(de->inode);
4327                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4328                         brelse(bh);
4329                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4330                         GOTO(out, rc);
4331                 }
4332
4333                 rc = osd_get_fid_from_dentry(de, rec);
4334
4335                 /* done with de, release bh */
4336                 brelse(bh);
4337                 if (rc != 0) {
4338                         if (unlikely(ino == osd_remote_parent_ino(dev)))
4339                                 /* If the parent is on remote MDT, and there
4340                                  * is no FID-in-dirent, then we have to get
4341                                  * the parent FID from the linkEA.  */
4342                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
4343                         else
4344                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4345                 } else {
4346                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4347                 }
4348
4349                 if (rc != 0) {
4350                         fid_zero(&oic->oic_fid);
4351                         GOTO(out, rc);
4352                 }
4353
4354                 if (osd_remote_fid(env, dev, fid))
4355                         GOTO(out, rc = 0);
4356
4357                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4358                 osd_consistency_check(oti, dev, oic);
4359         } else {
4360                 rc = -ENOENT;
4361         }
4362
4363         GOTO(out, rc);
4364
4365 out:
4366         if (hlock != NULL)
4367                 ldiskfs_htree_unlock(hlock);
4368         else
4369                 up_read(&obj->oo_ext_idx_sem);
4370         return rc;
4371 }
4372
4373 /**
4374  * Find the osd object for given fid.
4375  *
4376  * \param fid need to find the osd object having this fid
4377  *
4378  * \retval osd_object on success
4379  * \retval        -ve on error
4380  */
4381 static struct osd_object *osd_object_find(const struct lu_env *env,
4382                                           struct dt_object *dt,
4383                                           const struct lu_fid *fid)
4384 {
4385         struct lu_device  *ludev = dt->do_lu.lo_dev;
4386         struct osd_object *child = NULL;
4387         struct lu_object  *luch;
4388         struct lu_object  *lo;
4389
4390         /*
4391          * at this point topdev might not exist yet
4392          * (i.e. MGS is preparing profiles). so we can
4393          * not rely on topdev and instead lookup with
4394          * our device passed as topdev. this can't work
4395          * if the object isn't cached yet (as osd doesn't
4396          * allocate lu_header). IOW, the object must be
4397          * in the cache, otherwise lu_object_alloc() crashes
4398          * -bzzz
4399          */
4400         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4401                                  ludev : ludev->ld_site->ls_top_dev,
4402                                  fid, NULL);
4403         if (!IS_ERR(luch)) {
4404                 if (lu_object_exists(luch)) {
4405                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4406                         if (lo != NULL)
4407                                 child = osd_obj(lo);
4408                         else
4409                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4410                                                 "lu_object can't be located"
4411                                                 DFID"\n", PFID(fid));
4412
4413                         if (child == NULL) {
4414                                 lu_object_put(env, luch);
4415                                 CERROR("Unable to get osd_object\n");
4416                                 child = ERR_PTR(-ENOENT);
4417                         }
4418                 } else {
4419                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4420                                         "lu_object does not exists "DFID"\n",
4421                                         PFID(fid));
4422                         lu_object_put(env, luch);
4423                         child = ERR_PTR(-ENOENT);
4424                 }
4425         } else {
4426                 child = ERR_CAST(luch);
4427         }
4428
4429         return child;
4430 }
4431
4432 /**
4433  * Put the osd object once done with it.
4434  *
4435  * \param obj osd object that needs to be put
4436  */
4437 static inline void osd_object_put(const struct lu_env *env,
4438                                   struct osd_object *obj)
4439 {
4440         lu_object_put(env, &obj->oo_dt.do_lu);
4441 }
4442
4443 static int osd_index_declare_ea_insert(const struct lu_env *env,
4444                                        struct dt_object *dt,
4445                                        const struct dt_rec *rec,
4446                                        const struct dt_key *key,
4447                                        struct thandle *handle)
4448 {
4449         struct osd_thandle      *oh;
4450         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4451         struct lu_fid           *fid = (struct lu_fid *)rec;
4452         int                     rc;
4453         ENTRY;
4454
4455         LASSERT(!dt_object_remote(dt));
4456         LASSERT(handle != NULL);
4457
4458         oh = container_of0(handle, struct osd_thandle, ot_super);
4459         LASSERT(oh->ot_handle == NULL);
4460
4461         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4462                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4463
4464         if (osd_dt_obj(dt)->oo_inode != NULL) {
4465                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4466
4467                 /* We ignore block quota on meta pool (MDTs), so needn't
4468                  * calculate how many blocks will be consumed by this index
4469                  * insert */
4470                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4471                                            i_gid_read(inode), 0, oh,
4472                                            osd_dt_obj(dt), true, NULL, false);
4473         }
4474
4475         if (fid == NULL)
4476                 RETURN(0);
4477
4478         rc = osd_remote_fid(env, osd, fid);
4479         if (rc <= 0)
4480                 RETURN(rc);
4481
4482         rc = 0;
4483
4484         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4485                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4486         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4487                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4488         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4489                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4490
4491         RETURN(rc);
4492 }
4493
4494 /**
4495  * Index add function for interoperability mode (b11826).
4496  * It will add the directory entry.This entry is needed to
4497  * maintain name->fid mapping.
4498  *
4499  * \param key it is key i.e. file entry to be inserted
4500  * \param rec it is value of given key i.e. fid
4501  *
4502  * \retval   0, on success
4503  * \retval -ve, on error
4504  */
4505 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4506                                const struct dt_rec *rec,
4507                                const struct dt_key *key, struct thandle *th,
4508                                struct lustre_capa *capa, int ignore_quota)
4509 {
4510         struct osd_object       *obj = osd_dt_obj(dt);
4511         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4512         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4513         const struct lu_fid     *fid    = rec1->rec_fid;
4514         const char              *name = (const char *)key;
4515         struct osd_thread_info  *oti   = osd_oti_get(env);
4516         struct osd_inode_id     *id    = &oti->oti_id;
4517         struct inode            *child_inode = NULL;
4518         struct osd_object       *child = NULL;
4519         int                     rc;
4520         ENTRY;
4521
4522         LASSERT(osd_invariant(obj));
4523         LASSERT(dt_object_exists(dt));
4524         LASSERT(!dt_object_remote(dt));
4525         LASSERT(th != NULL);
4526
4527         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4528
4529         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4530                 RETURN(-EACCES);
4531
4532         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4533
4534         rc = osd_remote_fid(env, osd, fid);
4535         if (rc < 0) {
4536                 CERROR("%s: Can not find object "DFID" rc %d\n",
4537                        osd_name(osd), PFID(fid), rc);
4538                 RETURN(rc);
4539         }
4540
4541         if (rc == 1) {
4542                 /* Insert remote entry */
4543                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4544                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4545                         struct osd_thandle      *oh;
4546
4547                         /* If parent on remote MDT, we need put this object
4548                          * under AGENT */
4549                         oh = container_of(th, typeof(*oh), ot_super);
4550                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4551                         if (rc != 0) {
4552                                 CERROR("%s: add "DFID" error: rc = %d\n",
4553                                        osd_name(osd),
4554                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4555                                 RETURN(rc);
4556                         }
4557
4558                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4559                 } else {
4560                         child_inode = osd_create_local_agent_inode(env, osd,
4561                                         obj, fid, rec1->rec_type & S_IFMT, th);
4562                         if (IS_ERR(child_inode))
4563                                 RETURN(PTR_ERR(child_inode));
4564                 }
4565         } else {
4566                 /* Insert local entry */
4567                 child = osd_object_find(env, dt, fid);
4568                 if (IS_ERR(child)) {
4569                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4570                                osd_name(osd), PFID(fid),
4571                                id->oii_ino, id->oii_gen,
4572                                (int)PTR_ERR(child));
4573                         RETURN(PTR_ERR(child));
4574                 }
4575                 child_inode = igrab(child->oo_inode);
4576         }
4577
4578         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4579
4580         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4581                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4582
4583         iput(child_inode);
4584         if (child != NULL)
4585                 osd_object_put(env, child);
4586         LASSERT(osd_invariant(obj));
4587         RETURN(rc);
4588 }
4589
4590 /**
4591  *  Initialize osd Iterator for given osd index object.
4592  *
4593  *  \param  dt      osd index object
4594  */
4595
4596 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4597                                      struct dt_object *dt,
4598                                      __u32 unused,
4599                                      struct lustre_capa *capa)
4600 {
4601         struct osd_it_iam      *it;
4602         struct osd_thread_info *oti = osd_oti_get(env);
4603         struct osd_object      *obj = osd_dt_obj(dt);
4604         struct lu_object       *lo  = &dt->do_lu;
4605         struct iam_path_descr  *ipd;
4606         struct iam_container   *bag = &obj->oo_dir->od_container;
4607
4608         LASSERT(lu_object_exists(lo));
4609
4610         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4611                 return ERR_PTR(-EACCES);
4612
4613         if (oti->oti_it_inline) {
4614                 OBD_ALLOC_PTR(it);
4615                 if (it == NULL)
4616                         return ERR_PTR(-ENOMEM);
4617         } else {
4618                 it = &oti->oti_it;
4619                 oti->oti_it_inline = 1;
4620         }
4621
4622         ipd = osd_it_ipd_get(env, bag);
4623         if (likely(ipd != NULL)) {
4624                 it->oi_obj = obj;
4625                 it->oi_ipd = ipd;
4626                 lu_object_get(lo);
4627                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4628                 return (struct dt_it *)it;
4629         } else {
4630                 if (it != &oti->oti_it)
4631                         OBD_FREE_PTR(it);
4632                 else
4633                         oti->oti_it_inline = 0;
4634
4635                 return ERR_PTR(-ENOMEM);
4636         }
4637 }
4638
4639 /**
4640  * free given Iterator.
4641  */
4642
4643 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4644 {
4645         struct osd_thread_info  *oti = osd_oti_get(env);
4646         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4647         struct osd_object       *obj = it->oi_obj;
4648
4649         iam_it_fini(&it->oi_it);
4650         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4651         lu_object_put(env, &obj->oo_dt.do_lu);
4652         if (it != &oti->oti_it)
4653                 OBD_FREE_PTR(it);
4654         else
4655                 oti->oti_it_inline = 0;
4656 }
4657
4658 /**
4659  *  Move Iterator to record specified by \a key
4660  *
4661  *  \param  di      osd iterator
4662  *  \param  key     key for index
4663  *
4664  *  \retval +ve  di points to record with least key not larger than key
4665  *  \retval  0   di points to exact matched key
4666  *  \retval -ve  failure
4667  */
4668
4669 static int osd_it_iam_get(const struct lu_env *env,
4670                           struct dt_it *di, const struct dt_key *key)
4671 {
4672         struct osd_thread_info  *oti = osd_oti_get(env);
4673         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4674
4675         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4676                 /* swab quota uid/gid */
4677                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4678                 key = (struct dt_key *)&oti->oti_quota_id;
4679         }
4680
4681         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4682 }
4683
4684 /**
4685  *  Release Iterator
4686  *
4687  *  \param  di      osd iterator
4688  */
4689 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4690 {
4691         struct osd_it_iam *it = (struct osd_it_iam *)di;
4692
4693         iam_it_put(&it->oi_it);
4694 }
4695
4696 /**
4697  *  Move iterator by one record
4698  *
4699  *  \param  di      osd iterator
4700  *
4701  *  \retval +1   end of container reached
4702  *  \retval  0   success
4703  *  \retval -ve  failure
4704  */
4705
4706 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4707 {
4708         struct osd_it_iam *it = (struct osd_it_iam *)di;
4709
4710         return iam_it_next(&it->oi_it);
4711 }
4712
4713 /**
4714  * Return pointer to the key under iterator.
4715  */
4716
4717 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4718                                  const struct dt_it *di)
4719 {
4720         struct osd_thread_info *oti = osd_oti_get(env);
4721         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4722         struct osd_object      *obj = it->oi_obj;
4723         struct dt_key          *key;
4724
4725         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4726
4727         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4728                 /* swab quota uid/gid */
4729                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4730                 key = (struct dt_key *)&oti->oti_quota_id;
4731         }
4732
4733         return key;
4734 }
4735
4736 /**
4737  * Return size of key under iterator (in bytes)
4738  */
4739
4740 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4741 {
4742         struct osd_it_iam *it = (struct osd_it_iam *)di;
4743
4744         return iam_it_key_size(&it->oi_it);
4745 }
4746
4747 static inline void
4748 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4749 {
4750         /* check if file type is required */
4751         if (ent->lde_attrs & LUDA_TYPE) {
4752                 struct luda_type *lt;
4753                 int align = sizeof(*lt) - 1;
4754
4755                 len = (len + align) & ~align;
4756                 lt = (struct luda_type *)(ent->lde_name + len);
4757                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4758         }
4759
4760         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4761 }
4762
4763 /**
4764  * build lu direct from backend fs dirent.
4765  */
4766
4767 static inline void
4768 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4769                    char *name, __u16 namelen, __u16 type, __u32 attr)
4770 {
4771         ent->lde_attrs = attr | LUDA_FID;
4772         fid_cpu_to_le(&ent->lde_fid, fid);
4773
4774         ent->lde_hash = cpu_to_le64(offset);
4775         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4776
4777         strncpy(ent->lde_name, name, namelen);
4778         ent->lde_name[namelen] = '\0';
4779         ent->lde_namelen = cpu_to_le16(namelen);
4780
4781         /* append lustre attributes */
4782         osd_it_append_attrs(ent, namelen, type);
4783 }
4784
4785 /**
4786  * Return pointer to the record under iterator.
4787  */
4788 static int osd_it_iam_rec(const struct lu_env *env,
4789                           const struct dt_it *di,
4790                           struct dt_rec *dtrec, __u32 attr)
4791 {
4792         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4793         struct osd_thread_info *info = osd_oti_get(env);
4794         ENTRY;
4795
4796         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4797                 const struct osd_fid_pack *rec;
4798                 struct lu_fid             *fid = &info->oti_fid;
4799                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4800                 char                      *name;
4801                 int                        namelen;
4802                 __u64                      hash;
4803                 int                        rc;
4804
4805                 name = (char *)iam_it_key_get(&it->oi_it);
4806                 if (IS_ERR(name))
4807                         RETURN(PTR_ERR(name));
4808
4809                 namelen = iam_it_key_size(&it->oi_it);
4810
4811                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4812                 if (IS_ERR(rec))
4813                         RETURN(PTR_ERR(rec));
4814
4815                 rc = osd_fid_unpack(fid, rec);
4816                 if (rc)
4817                         RETURN(rc);
4818
4819                 hash = iam_it_store(&it->oi_it);
4820
4821                 /* IAM does not store object type in IAM index (dir) */
4822                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4823                                    0, LUDA_FID);
4824         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4825                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4826                            (struct iam_rec *)dtrec);
4827                 osd_quota_unpack(it->oi_obj, dtrec);
4828         } else {
4829                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4830                            (struct iam_rec *)dtrec);
4831         }
4832
4833         RETURN(0);
4834 }
4835
4836 /**
4837  * Returns cookie for current Iterator position.
4838  */
4839 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4840 {
4841         struct osd_it_iam *it = (struct osd_it_iam *)di;
4842
4843         return iam_it_store(&it->oi_it);
4844 }
4845
4846 /**
4847  * Restore iterator from cookie.
4848  *
4849  * \param  di      osd iterator
4850  * \param  hash    Iterator location cookie
4851  *
4852  * \retval +ve  di points to record with least key not larger than key.
4853  * \retval  0   di points to exact matched key
4854  * \retval -ve  failure
4855  */
4856
4857 static int osd_it_iam_load(const struct lu_env *env,
4858                            const struct dt_it *di, __u64 hash)
4859 {
4860         struct osd_it_iam *it = (struct osd_it_iam *)di;
4861
4862         return iam_it_load(&it->oi_it, hash);
4863 }
4864
4865 static const struct dt_index_operations osd_index_iam_ops = {
4866         .dio_lookup         = osd_index_iam_lookup,
4867         .dio_declare_insert = osd_index_declare_iam_insert,
4868         .dio_insert         = osd_index_iam_insert,
4869         .dio_declare_delete = osd_index_declare_iam_delete,
4870         .dio_delete         = osd_index_iam_delete,
4871         .dio_it     = {
4872                 .init     = osd_it_iam_init,
4873                 .fini     = osd_it_iam_fini,
4874                 .get      = osd_it_iam_get,
4875                 .put      = osd_it_iam_put,
4876                 .next     = osd_it_iam_next,
4877                 .key      = osd_it_iam_key,
4878                 .key_size = osd_it_iam_key_size,
4879                 .rec      = osd_it_iam_rec,
4880                 .store    = osd_it_iam_store,
4881                 .load     = osd_it_iam_load
4882         }
4883 };
4884
4885
4886 /**
4887  * Creates or initializes iterator context.
4888  *
4889  * \retval struct osd_it_ea, iterator structure on success
4890  *
4891  */
4892 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4893                                     struct dt_object *dt,
4894                                     __u32 attr,
4895                                     struct lustre_capa *capa)
4896 {
4897         struct osd_object       *obj  = osd_dt_obj(dt);
4898         struct osd_thread_info  *info = osd_oti_get(env);
4899         struct osd_it_ea        *it;
4900         struct file             *file;
4901         struct lu_object        *lo   = &dt->do_lu;
4902         struct dentry           *obj_dentry = &info->oti_it_dentry;
4903         ENTRY;
4904
4905         LASSERT(lu_object_exists(lo));
4906
4907         if (info->oti_it_inline) {
4908                 OBD_ALLOC_PTR(it);
4909                 if (it == NULL)
4910                         RETURN(ERR_PTR(-ENOMEM));
4911         } else {
4912                 it = &info->oti_it_ea;
4913                 info->oti_it_inline = 1;
4914         }
4915
4916         obj_dentry->d_inode = obj->oo_inode;
4917         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4918         obj_dentry->d_name.hash = 0;
4919
4920         it->oie_rd_dirent       = 0;
4921         it->oie_it_dirent       = 0;
4922         it->oie_dirent          = NULL;
4923         it->oie_buf             = info->oti_it_ea_buf;
4924         it->oie_obj             = obj;
4925
4926         file = &it->oie_file;
4927         /* Reset the "file" totally to avoid to reuse any old value from
4928          * former readdir handling, the "file->f_pos" should be zero. */
4929         memset(file, 0, sizeof(*file));
4930         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4931         if (attr & LUDA_64BITHASH)
4932                 file->f_mode    = FMODE_64BITHASH;
4933         else
4934                 file->f_mode    = FMODE_32BITHASH;
4935         file->f_dentry          = obj_dentry;
4936         file->f_mapping         = obj->oo_inode->i_mapping;
4937         file->f_op              = obj->oo_inode->i_fop;
4938         set_file_inode(file, obj->oo_inode);
4939
4940         lu_object_get(lo);
4941         RETURN((struct dt_it *) it);
4942 }
4943
4944 /**
4945  * Destroy or finishes iterator context.
4946  *
4947  * \param di iterator structure to be destroyed
4948  */
4949 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4950 {
4951         struct osd_thread_info  *info   = osd_oti_get(env);
4952         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4953         struct osd_object       *obj    = it->oie_obj;
4954         struct inode            *inode  = obj->oo_inode;
4955
4956         ENTRY;
4957         it->oie_file.f_op->release(inode, &it->oie_file);
4958         lu_object_put(env, &obj->oo_dt.do_lu);
4959         if (it != &info->oti_it_ea)
4960                 OBD_FREE_PTR(it);
4961         else
4962                 info->oti_it_inline = 0;
4963         EXIT;
4964 }
4965
4966 /**
4967  * It position the iterator at given key, so that next lookup continues from
4968  * that key Or it is similar to dio_it->load() but based on a key,
4969  * rather than file position.
4970  *
4971  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4972  * to the beginning.
4973  *
4974  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4975  */
4976 static int osd_it_ea_get(const struct lu_env *env,
4977                          struct dt_it *di, const struct dt_key *key)
4978 {
4979         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4980
4981         ENTRY;
4982         LASSERT(((const char *)key)[0] == '\0');
4983         it->oie_file.f_pos      = 0;
4984         it->oie_rd_dirent       = 0;
4985         it->oie_it_dirent       = 0;
4986         it->oie_dirent          = NULL;
4987
4988         RETURN(+1);
4989 }
4990
4991 /**
4992  * Does nothing
4993  */
4994 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4995 {
4996 }
4997
4998 struct osd_filldir_cbs {
4999 #ifdef HAVE_DIR_CONTEXT
5000         struct dir_context ctx;
5001 #endif
5002         struct osd_it_ea  *it;
5003 };
5004 /**
5005  * It is called internally by ->readdir(). It fills the
5006  * iterator's in-memory data structure with required
5007  * information i.e. name, namelen, rec_size etc.
5008  *
5009  * \param buf in which information to be filled in.
5010  * \param name name of the file in given dir
5011  *
5012  * \retval 0 on success
5013  * \retval 1 on buffer full
5014  */
5015 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5016                                loff_t offset, __u64 ino,
5017                                unsigned d_type)
5018 {
5019         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5020         struct osd_object       *obj  = it->oie_obj;
5021         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5022         struct lu_fid           *fid  = &ent->oied_fid;
5023         struct osd_fid_pack     *rec;
5024         ENTRY;
5025
5026         /* this should never happen */
5027         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5028                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5029                 RETURN(-EIO);
5030         }
5031
5032         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5033             OSD_IT_EA_BUFSIZE)
5034                 RETURN(1);
5035
5036         /* "." is just the object itself. */
5037         if (namelen == 1 && name[0] == '.') {
5038                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5039         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5040                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5041                 if (osd_fid_unpack(fid, rec) != 0)
5042                         fid_zero(fid);
5043         } else {
5044                 fid_zero(fid);
5045         }
5046         d_type &= ~LDISKFS_DIRENT_LUFID;
5047
5048         /* NOT export local root. */
5049         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5050                 ino = obj->oo_inode->i_ino;
5051                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5052         }
5053
5054         ent->oied_ino     = ino;
5055         ent->oied_off     = offset;
5056         ent->oied_namelen = namelen;
5057         ent->oied_type    = d_type;
5058
5059         memcpy(ent->oied_name, name, namelen);
5060
5061         it->oie_rd_dirent++;
5062         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5063         RETURN(0);
5064 }
5065
5066 /**
5067  * Calls ->readdir() to load a directory entry at a time
5068  * and stored it in iterator's in-memory data structure.
5069  *
5070  * \param di iterator's in memory structure
5071  *
5072  * \retval   0 on success
5073  * \retval -ve on error
5074  * \retval +1 reach the end of entry
5075  */
5076 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5077                                const struct dt_it *di)
5078 {
5079         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5080         struct osd_object  *obj   = it->oie_obj;
5081         struct inode       *inode = obj->oo_inode;
5082         struct htree_lock  *hlock = NULL;
5083         struct file        *filp  = &it->oie_file;
5084         int                 rc = 0;
5085         struct osd_filldir_cbs buf = {
5086 #ifdef HAVE_DIR_CONTEXT
5087                 .ctx.actor = osd_ldiskfs_filldir,
5088 #endif
5089                 .it = it
5090         };
5091
5092         ENTRY;
5093         it->oie_dirent = it->oie_buf;
5094         it->oie_rd_dirent = 0;
5095
5096         if (obj->oo_hl_head != NULL) {
5097                 hlock = osd_oti_get(env)->oti_hlock;
5098                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5099                                    inode, LDISKFS_HLOCK_READDIR);
5100         } else {
5101                 down_read(&obj->oo_ext_idx_sem);
5102         }
5103
5104 #ifdef HAVE_DIR_CONTEXT
5105         buf.ctx.pos = filp->f_pos;
5106         rc = inode->i_fop->iterate(filp, &buf.ctx);
5107         filp->f_pos = buf.ctx.pos;
5108 #else
5109         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5110 #endif
5111
5112         if (hlock != NULL)
5113                 ldiskfs_htree_unlock(hlock);
5114         else
5115                 up_read(&obj->oo_ext_idx_sem);
5116
5117         if (it->oie_rd_dirent == 0) {
5118                 /*If it does not get any dirent, it means it has been reached
5119                  *to the end of the dir */
5120                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5121                 if (rc == 0)
5122                         rc = 1;
5123         } else {
5124                 it->oie_dirent = it->oie_buf;
5125                 it->oie_it_dirent = 1;
5126         }
5127
5128         RETURN(rc);
5129 }
5130
5131 /**
5132  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5133  * to load a directory entry at a time and stored it in
5134  * iterator's in-memory data structure.
5135  *
5136  * \param di iterator's in memory structure
5137  *
5138  * \retval +ve iterator reached to end
5139  * \retval   0 iterator not reached to end
5140  * \retval -ve on error
5141  */
5142 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5143 {
5144         struct osd_it_ea *it = (struct osd_it_ea *)di;
5145         int rc;
5146
5147         ENTRY;
5148
5149         if (it->oie_it_dirent < it->oie_rd_dirent) {
5150                 it->oie_dirent =
5151                         (void *) it->oie_dirent +
5152                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5153                                        it->oie_dirent->oied_namelen);
5154                 it->oie_it_dirent++;
5155                 RETURN(0);
5156         } else {
5157                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5158                         rc = +1;
5159                 else
5160                         rc = osd_ldiskfs_it_fill(env, di);
5161         }
5162
5163         RETURN(rc);
5164 }
5165
5166 /**
5167  * Returns the key at current position from iterator's in memory structure.
5168  *
5169  * \param di iterator's in memory structure
5170  *
5171  * \retval key i.e. struct dt_key on success
5172  */
5173 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5174                                     const struct dt_it *di)
5175 {
5176         struct osd_it_ea *it = (struct osd_it_ea *)di;
5177
5178         return (struct dt_key *)it->oie_dirent->oied_name;
5179 }
5180
5181 /**
5182  * Returns the key's size at current position from iterator's in memory structure.
5183  *
5184  * \param di iterator's in memory structure
5185  *
5186  * \retval key_size i.e. struct dt_key on success
5187  */
5188 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5189 {
5190         struct osd_it_ea *it = (struct osd_it_ea *)di;
5191
5192         return it->oie_dirent->oied_namelen;
5193 }
5194
5195 static int
5196 osd_dirent_update(handle_t *jh, struct super_block *sb,
5197                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5198                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5199 {
5200         struct osd_fid_pack *rec;
5201         int                  rc;
5202         ENTRY;
5203
5204         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5205         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5206
5207         rc = ldiskfs_journal_get_write_access(jh, bh);
5208         if (rc != 0)
5209                 RETURN(rc);
5210
5211         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5212         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5213         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5214
5215         RETURN(rc);
5216 }
5217
5218 static inline int
5219 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5220 {
5221         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5222             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5223                 return 1;
5224         else
5225                 return 0;
5226 }
5227
5228 static inline int
5229 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5230 {
5231         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5232                  "dot_dotdot = %d\n", dot_dotdot);
5233
5234         if (LDISKFS_DIR_REC_LEN(de) >=
5235             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5236                 return 1;
5237         else
5238                 return 0;
5239 }
5240
5241 static int
5242 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5243                     struct inode *dir, struct inode *inode,
5244                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5245                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5246                     struct htree_lock *hlock)
5247 {
5248         struct dentry               *dentry;
5249         struct osd_fid_pack         *rec;
5250         struct ldiskfs_dentry_param *ldp;
5251         int                          rc;
5252         ENTRY;
5253
5254         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5255                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5256                 RETURN(0);
5257
5258         /* There is enough space to hold the FID-in-dirent. */
5259         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5260                                  dir->i_sb->s_blocksize)) {
5261                 rc = ldiskfs_journal_get_write_access(jh, bh);
5262                 if (rc != 0)
5263                         RETURN(rc);
5264
5265                 de->name[de->name_len] = 0;
5266                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5267                 rec->fp_len = sizeof(struct lu_fid) + 1;
5268                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5269                 de->file_type |= LDISKFS_DIRENT_LUFID;
5270
5271                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5272
5273                 RETURN(rc);
5274         }
5275
5276         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5277         if (rc != 0)
5278                 RETURN(rc);
5279
5280         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5281                                            ent->oied_namelen);
5282         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5283         osd_get_ldiskfs_dirent_param(ldp, fid);
5284         dentry->d_fsdata = (void *)ldp;
5285         ll_vfs_dq_init(dir);
5286         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5287         /* It is too bad, we cannot reinsert the name entry back.
5288          * That means we lose it! */
5289         if (rc != 0)
5290                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5291                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5292                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5293                        dir->i_ino, dir->i_generation,
5294                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5295
5296         RETURN(rc);
5297 }
5298
5299 static int
5300 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5301                         struct osd_it_ea *it, struct lu_fid *fid,
5302                         struct osd_inode_id *id, __u32 *attr)
5303 {
5304         struct osd_thread_info     *info        = osd_oti_get(env);
5305         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5306         struct osd_device          *dev         = osd_obj2dev(obj);
5307         struct super_block         *sb          = osd_sb(dev);
5308         const char                 *devname     =
5309                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5310         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5311         struct inode               *dir         = obj->oo_inode;
5312         struct htree_lock          *hlock       = NULL;
5313         struct buffer_head         *bh          = NULL;
5314         handle_t                   *jh          = NULL;
5315         struct ldiskfs_dir_entry_2 *de;
5316         struct dentry              *dentry;
5317         struct inode               *inode;
5318         int                         credits;
5319         int                         rc;
5320         int                         dot_dotdot  = 0;
5321         bool                        dirty       = false;
5322         ENTRY;
5323
5324         if (ent->oied_name[0] == '.') {
5325                 if (ent->oied_namelen == 1)
5326                         dot_dotdot = 1;
5327                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5328                         dot_dotdot = 2;
5329         }
5330
5331         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5332                                       ent->oied_namelen);
5333
5334         /* We need to ensure that the name entry is still valid.
5335          * Because it may be removed or renamed by other already.
5336          *
5337          * The unlink or rename operation will start journal before PDO lock,
5338          * so to avoid deadlock, here we need to start journal handle before
5339          * related PDO lock also. But because we do not know whether there
5340          * will be something to be repaired before PDO lock, we just start
5341          * journal without conditions.
5342          *
5343          * We may need to remove the name entry firstly, then insert back.
5344          * One credit is for user quota file update.
5345          * One credit is for group quota file update.
5346          * Two credits are for dirty inode. */
5347         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5348                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5349
5350 again:
5351         if (dev->od_dirent_journal) {
5352                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5353                 if (IS_ERR(jh)) {
5354                         rc = PTR_ERR(jh);
5355                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5356                                "check_repair, dir = %lu/%u, credits = %d, "
5357                                "name = %.*s: rc = %d\n",
5358                                devname, dir->i_ino, dir->i_generation, credits,
5359                                ent->oied_namelen, ent->oied_name, rc);
5360                         RETURN(rc);
5361                 }
5362
5363                 if (obj->oo_hl_head != NULL) {
5364                         hlock = osd_oti_get(env)->oti_hlock;
5365                         /* "0" means exclusive lock for the whole directory.
5366                          * We need to prevent others access such name entry
5367                          * during the delete + insert. Neither HLOCK_ADD nor
5368                          * HLOCK_DEL cannot guarantee the atomicity. */
5369                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5370                 } else {
5371                         down_write(&obj->oo_ext_idx_sem);
5372                 }
5373         } else {
5374                 if (obj->oo_hl_head != NULL) {
5375                         hlock = osd_oti_get(env)->oti_hlock;
5376                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5377                                            LDISKFS_HLOCK_LOOKUP);
5378                 } else {
5379                         down_read(&obj->oo_ext_idx_sem);
5380                 }
5381         }
5382
5383         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5384         /* For dot/dotdot entry, if there is not enough space to hold the
5385          * FID-in-dirent, just keep them there. It only happens when the
5386          * device upgraded from 1.8 or restored from MDT file-level backup.
5387          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5388          * and needs to get FID from LMA when readdir, it will not affect the
5389          * performance much. */
5390         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5391             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5392                 *attr |= LUDA_IGNORE;
5393                 GOTO(out_journal, rc = 0);
5394         }
5395
5396         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5397         inode = osd_iget(info, dev, id);
5398         if (IS_ERR(inode)) {
5399                 rc = PTR_ERR(inode);
5400                 if (rc == -ENOENT || rc == -ESTALE)
5401                         rc = 1;
5402                 else
5403                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5404                                "check_repair, dir = %lu/%u, name = %.*s: "
5405                                "rc = %d\n",
5406                                devname, dir->i_ino, dir->i_generation,
5407                                ent->oied_namelen, ent->oied_name, rc);
5408
5409                 GOTO(out_journal, rc);
5410         }
5411
5412         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5413         if (rc == 0) {
5414                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5415
5416                 if (fid_is_sane(fid)) {
5417                         /* FID-in-dirent is valid. */
5418                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5419                                 GOTO(out_inode, rc = 0);
5420
5421                         /* Do not repair under dryrun mode. */
5422                         if (*attr & LUDA_VERIFY_DRYRUN) {
5423                                 *attr |= LUDA_REPAIR;
5424                                 GOTO(out_inode, rc = 0);
5425                         }
5426
5427                         if (!dev->od_dirent_journal) {
5428                                 iput(inode);
5429                                 brelse(bh);
5430                                 if (hlock != NULL)
5431                                         ldiskfs_htree_unlock(hlock);
5432                                 else
5433                                         up_read(&obj->oo_ext_idx_sem);
5434                                 dev->od_dirent_journal = 1;
5435                                 goto again;
5436                         }
5437
5438                         *fid = lma->lma_self_fid;
5439                         dirty = true;
5440                         /* Update the FID-in-dirent. */
5441                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5442                         if (rc == 0)
5443                                 *attr |= LUDA_REPAIR;
5444                         else
5445                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5446                                        "in the dirent, dir = %lu/%u, "
5447                                        "name = %.*s, "DFID": rc = %d\n",
5448                                        devname, dir->i_ino, dir->i_generation,
5449                                        ent->oied_namelen, ent->oied_name,
5450                                        PFID(fid), rc);
5451                 } else {
5452                         /* Do not repair under dryrun mode. */
5453                         if (*attr & LUDA_VERIFY_DRYRUN) {
5454                                 *fid = lma->lma_self_fid;
5455                                 *attr |= LUDA_REPAIR;
5456                                 GOTO(out_inode, rc = 0);
5457                         }
5458
5459                         if (!dev->od_dirent_journal) {
5460                                 iput(inode);
5461                                 brelse(bh);
5462                                 if (hlock != NULL)
5463                                         ldiskfs_htree_unlock(hlock);
5464                                 else
5465                                         up_read(&obj->oo_ext_idx_sem);
5466                                 dev->od_dirent_journal = 1;
5467                                 goto again;
5468                         }
5469
5470                         *fid = lma->lma_self_fid;
5471                         dirty = true;
5472                         /* Append the FID-in-dirent. */
5473                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5474                                                  fid, bh, de, hlock);
5475                         if (rc == 0)
5476                                 *attr |= LUDA_REPAIR;
5477                         else
5478                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5479                                        "after the dirent, dir = %lu/%u, "
5480                                        "name = %.*s, "DFID": rc = %d\n",
5481                                        devname, dir->i_ino, dir->i_generation,
5482                                        ent->oied_namelen, ent->oied_name,
5483                                        PFID(fid), rc);
5484                 }
5485         } else if (rc == -ENODATA) {
5486                 /* Do not repair under dryrun mode. */
5487                 if (*attr & LUDA_VERIFY_DRYRUN) {
5488                         if (fid_is_sane(fid)) {
5489                                 *attr |= LUDA_REPAIR;
5490                         } else {
5491                                 lu_igif_build(fid, inode->i_ino,
5492                                               inode->i_generation);
5493                                 *attr |= LUDA_UPGRADE;
5494                         }
5495                         GOTO(out_inode, rc = 0);
5496                 }
5497
5498                 if (!dev->od_dirent_journal) {
5499                         iput(inode);
5500                         brelse(bh);
5501                         if (hlock != NULL)
5502                                 ldiskfs_htree_unlock(hlock);
5503                         else
5504                                 up_read(&obj->oo_ext_idx_sem);
5505                         dev->od_dirent_journal = 1;
5506                         goto again;
5507                 }
5508
5509                 dirty = true;
5510                 if (unlikely(fid_is_sane(fid))) {
5511                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5512                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5513                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5514                         if (rc == 0)
5515                                 *attr |= LUDA_REPAIR;
5516                         else
5517                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5518                                        "update dirent, dir = %lu/%u, "
5519                                        "name = %.*s, "DFID": rc = %d\n",
5520                                        devname, dir->i_ino, dir->i_generation,
5521                                        ent->oied_namelen, ent->oied_name,
5522                                        PFID(fid), rc);
5523                 } else {
5524                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5525                         /* It is probably IGIF object. Only aappend the
5526                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5527                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5528                                                  fid, bh, de, hlock);
5529                         if (rc == 0)
5530                                 *attr |= LUDA_UPGRADE;
5531                         else
5532                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5533                                        "after the dirent, dir = %lu/%u, "
5534                                        "name = %.*s, "DFID": rc = %d\n",
5535                                        devname, dir->i_ino, dir->i_generation,
5536                                        ent->oied_namelen, ent->oied_name,
5537                                        PFID(fid), rc);
5538                 }
5539         }
5540
5541         GOTO(out_inode, rc);
5542
5543 out_inode:
5544         iput(inode);
5545
5546 out_journal:
5547         brelse(bh);
5548         if (hlock != NULL) {
5549                 ldiskfs_htree_unlock(hlock);
5550         } else {
5551                 if (dev->od_dirent_journal)
5552                         up_write(&obj->oo_ext_idx_sem);
5553                 else
5554                         up_read(&obj->oo_ext_idx_sem);
5555         }
5556         if (jh != NULL)
5557                 ldiskfs_journal_stop(jh);
5558         if (rc >= 0 && !dirty)
5559                 dev->od_dirent_journal = 0;
5560         return rc;
5561 }
5562
5563 /**
5564  * Returns the value at current position from iterator's in memory structure.
5565  *
5566  * \param di struct osd_it_ea, iterator's in memory structure
5567  * \param attr attr requested for dirent.
5568  * \param lde lustre dirent
5569  *
5570  * \retval   0 no error and \param lde has correct lustre dirent.
5571  * \retval -ve on error
5572  */
5573 static inline int osd_it_ea_rec(const struct lu_env *env,
5574                                 const struct dt_it *di,
5575                                 struct dt_rec *dtrec, __u32 attr)
5576 {
5577         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5578         struct osd_object      *obj   = it->oie_obj;
5579         struct osd_device      *dev   = osd_obj2dev(obj);
5580         struct osd_thread_info *oti   = osd_oti_get(env);
5581         struct osd_inode_id    *id    = &oti->oti_id;
5582         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5583         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5584         __u32                   ino   = it->oie_dirent->oied_ino;
5585         int                     rc    = 0;
5586         ENTRY;
5587
5588         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5589
5590         if (attr & LUDA_VERIFY) {
5591                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5592                         attr |= LUDA_IGNORE;
5593                         /* If the parent is on remote MDT, and there
5594                          * is no FID-in-dirent, then we have to get
5595                          * the parent FID from the linkEA.  */
5596                         osd_get_pfid_from_linkea(env, obj, fid);
5597                 } else {
5598                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5599                                                      &attr);
5600                 }
5601         } else {
5602                 attr &= ~LU_DIRENT_ATTRS_MASK;
5603                 if (!fid_is_sane(fid)) {
5604                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5605                             likely(it->oie_dirent->oied_namelen != 2 ||
5606                                    it->oie_dirent->oied_name[0] != '.' ||
5607                                    it->oie_dirent->oied_name[1] != '.'))
5608                                 RETURN(-ENOENT);
5609
5610                         if (unlikely(ino == osd_remote_parent_ino(dev)))
5611                                 /* If the parent is on remote MDT, and there
5612                                  * is no FID-in-dirent, then we have to get
5613                                  * the parent FID from the linkEA.  */
5614                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5615                         else
5616                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5617                 } else {
5618                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5619                 }
5620         }
5621
5622         /* Pack the entry anyway, at least the offset is right. */
5623         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5624                            it->oie_dirent->oied_name,
5625                            it->oie_dirent->oied_namelen,
5626                            it->oie_dirent->oied_type, attr);
5627
5628         if (rc < 0)
5629                 RETURN(rc);
5630
5631         if (osd_remote_fid(env, dev, fid))
5632                 RETURN(0);
5633
5634         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5635                 osd_add_oi_cache(oti, dev, id, fid);
5636
5637         RETURN(rc > 0 ? 0 : rc);
5638 }
5639
5640 /**
5641  * Returns the record size size at current position.
5642  *
5643  * This function will return record(lu_dirent) size in bytes.
5644  *
5645  * \param[in] env       execution environment
5646  * \param[in] di        iterator's in memory structure
5647  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5648  *                      calculate the lu_dirent size.
5649  *
5650  * \retval      record size(in bytes & in memory) of the current lu_dirent
5651  *              entry.
5652  */
5653 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5654                               __u32 attr)
5655 {
5656         struct osd_it_ea *it = (struct osd_it_ea *)di;
5657
5658         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5659 }
5660
5661 /**
5662  * Returns a cookie for current position of the iterator head, so that
5663  * user can use this cookie to load/start the iterator next time.
5664  *
5665  * \param di iterator's in memory structure
5666  *
5667  * \retval cookie for current position, on success
5668  */
5669 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5670 {
5671         struct osd_it_ea *it = (struct osd_it_ea *)di;
5672
5673         return it->oie_dirent->oied_off;
5674 }
5675
5676 /**
5677  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5678  * to load a directory entry at a time and stored it i inn,
5679  * in iterator's in-memory data structure.
5680  *
5681  * \param di struct osd_it_ea, iterator's in memory structure
5682  *
5683  * \retval +ve on success
5684  * \retval -ve on error
5685  */
5686 static int osd_it_ea_load(const struct lu_env *env,
5687                           const struct dt_it *di, __u64 hash)
5688 {
5689         struct osd_it_ea *it = (struct osd_it_ea *)di;
5690         int rc;
5691
5692         ENTRY;
5693         it->oie_file.f_pos = hash;
5694
5695         rc =  osd_ldiskfs_it_fill(env, di);
5696         if (rc > 0)
5697                 rc = -ENODATA;
5698
5699         if (rc == 0)
5700                 rc = +1;
5701
5702         RETURN(rc);
5703 }
5704
5705 /**
5706  * Index lookup function for interoperability mode (b11826).
5707  *
5708  * \param key,  key i.e. file name to be searched
5709  *
5710  * \retval +ve, on success
5711  * \retval -ve, on error
5712  */
5713 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5714                                struct dt_rec *rec, const struct dt_key *key,
5715                                struct lustre_capa *capa)
5716 {
5717         struct osd_object *obj = osd_dt_obj(dt);
5718         int rc = 0;
5719
5720         ENTRY;
5721
5722         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5723         LINVRNT(osd_invariant(obj));
5724
5725         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5726                 return -EACCES;
5727
5728         rc = osd_ea_lookup_rec(env, obj, rec, key);
5729         if (rc == 0)
5730                 rc = +1;
5731         RETURN(rc);
5732 }
5733
5734 /**
5735  * Index and Iterator operations for interoperability
5736  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5737  */
5738 static const struct dt_index_operations osd_index_ea_ops = {
5739         .dio_lookup         = osd_index_ea_lookup,
5740         .dio_declare_insert = osd_index_declare_ea_insert,
5741         .dio_insert         = osd_index_ea_insert,
5742         .dio_declare_delete = osd_index_declare_ea_delete,
5743         .dio_delete         = osd_index_ea_delete,
5744         .dio_it     = {
5745                 .init     = osd_it_ea_init,
5746                 .fini     = osd_it_ea_fini,
5747                 .get      = osd_it_ea_get,
5748                 .put      = osd_it_ea_put,
5749                 .next     = osd_it_ea_next,
5750                 .key      = osd_it_ea_key,
5751                 .key_size = osd_it_ea_key_size,
5752                 .rec      = osd_it_ea_rec,
5753                 .rec_size = osd_it_ea_rec_size,
5754                 .store    = osd_it_ea_store,
5755                 .load     = osd_it_ea_load
5756         }
5757 };
5758
5759 static void *osd_key_init(const struct lu_context *ctx,
5760                           struct lu_context_key *key)
5761 {
5762         struct osd_thread_info *info;
5763
5764         OBD_ALLOC_PTR(info);
5765         if (info == NULL)
5766                 return ERR_PTR(-ENOMEM);
5767
5768         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5769         if (info->oti_it_ea_buf == NULL)
5770                 goto out_free_info;
5771
5772         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5773
5774         info->oti_hlock = ldiskfs_htree_lock_alloc();
5775         if (info->oti_hlock == NULL)
5776                 goto out_free_ea;
5777
5778         return info;
5779
5780  out_free_ea:
5781         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5782  out_free_info:
5783         OBD_FREE_PTR(info);
5784         return ERR_PTR(-ENOMEM);
5785 }
5786
5787 static void osd_key_fini(const struct lu_context *ctx,
5788                          struct lu_context_key *key, void* data)
5789 {
5790         struct osd_thread_info *info = data;
5791
5792         if (info->oti_hlock != NULL)
5793                 ldiskfs_htree_lock_free(info->oti_hlock);
5794         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5795         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5796         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5797         lu_buf_free(&info->oti_big_buf);
5798         OBD_FREE_PTR(info);
5799 }
5800
5801 static void osd_key_exit(const struct lu_context *ctx,
5802                          struct lu_context_key *key, void *data)
5803 {
5804         struct osd_thread_info *info = data;
5805
5806         LASSERT(info->oti_r_locks == 0);
5807         LASSERT(info->oti_w_locks == 0);
5808         LASSERT(info->oti_txns    == 0);
5809 }
5810
5811 /* type constructor/destructor: osd_type_init, osd_type_fini */
5812 LU_TYPE_INIT_FINI(osd, &osd_key);
5813
5814 struct lu_context_key osd_key = {
5815         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5816         .lct_init = osd_key_init,
5817         .lct_fini = osd_key_fini,
5818         .lct_exit = osd_key_exit
5819 };
5820
5821
5822 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5823                            const char *name, struct lu_device *next)
5824 {
5825         struct osd_device *osd = osd_dev(d);
5826
5827         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5828             >= sizeof(osd->od_svname))
5829                 return -E2BIG;
5830         return osd_procfs_init(osd, name);
5831 }
5832
5833 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5834 {
5835         struct seq_server_site  *ss = osd_seq_site(osd);
5836         int                     rc;
5837         ENTRY;
5838
5839         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5840                 RETURN(0);
5841
5842         if (unlikely(ss == NULL))
5843                 RETURN(-ENODEV);
5844
5845         OBD_ALLOC_PTR(osd->od_cl_seq);
5846         if (osd->od_cl_seq == NULL)
5847                 RETURN(-ENOMEM);
5848
5849         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5850                              osd->od_svname, ss->ss_server_seq);
5851
5852         if (rc != 0) {
5853                 OBD_FREE_PTR(osd->od_cl_seq);
5854                 osd->od_cl_seq = NULL;
5855         }
5856
5857         RETURN(rc);
5858 }
5859
5860 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5861 {
5862         if (osd->od_cl_seq == NULL)
5863                 return;
5864
5865         seq_client_fini(osd->od_cl_seq);
5866         OBD_FREE_PTR(osd->od_cl_seq);
5867         osd->od_cl_seq = NULL;
5868 }
5869
5870 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5871 {
5872         ENTRY;
5873
5874         /* shutdown quota slave instance associated with the device */
5875         if (o->od_quota_slave != NULL) {
5876                 qsd_fini(env, o->od_quota_slave);
5877                 o->od_quota_slave = NULL;
5878         }
5879
5880         osd_fid_fini(env, o);
5881
5882         RETURN(0);
5883 }
5884
5885 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5886 {
5887         ENTRY;
5888
5889         if (o->od_mnt != NULL) {
5890                 shrink_dcache_sb(osd_sb(o));
5891                 osd_sync(env, &o->od_dt_dev);
5892
5893                 mntput(o->od_mnt);
5894                 o->od_mnt = NULL;
5895         }
5896
5897         EXIT;
5898 }
5899
5900 static int osd_mount(const struct lu_env *env,
5901                      struct osd_device *o, struct lustre_cfg *cfg)
5902 {
5903         const char              *name  = lustre_cfg_string(cfg, 0);
5904         const char              *dev  = lustre_cfg_string(cfg, 1);
5905         const char              *opts;
5906         unsigned long            page, s_flags, lmd_flags = 0;
5907         struct page             *__page;
5908         struct file_system_type *type;
5909         char                    *options = NULL;
5910         char                    *str;
5911         struct osd_thread_info  *info = osd_oti_get(env);
5912         struct lu_fid           *fid = &info->oti_fid;
5913         struct inode            *inode;
5914         int                      rc = 0;
5915         ENTRY;
5916
5917         if (o->od_mnt != NULL)
5918                 RETURN(0);
5919
5920         if (strlen(dev) >= sizeof(o->od_mntdev))
5921                 RETURN(-E2BIG);
5922         strcpy(o->od_mntdev, dev);
5923
5924         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5925         if (__page == NULL)
5926                 GOTO(out, rc = -ENOMEM);
5927
5928         str = lustre_cfg_string(cfg, 2);
5929         s_flags = simple_strtoul(str, NULL, 0);
5930         str = strstr(str, ":");
5931         if (str)
5932                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5933         opts = lustre_cfg_string(cfg, 3);
5934         page = (unsigned long)page_address(__page);
5935         options = (char *)page;
5936         *options = '\0';
5937         if (opts == NULL)
5938                 strcat(options, "user_xattr,acl");
5939         else
5940                 strcat(options, opts);
5941
5942         /* Glom up mount options */
5943         if (*options != '\0')
5944                 strcat(options, ",");
5945         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5946
5947         type = get_fs_type("ldiskfs");
5948         if (!type) {
5949                 CERROR("%s: cannot find ldiskfs module\n", name);
5950                 GOTO(out, rc = -ENODEV);
5951         }
5952
5953         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5954         module_put(type->owner);
5955
5956         if (IS_ERR(o->od_mnt)) {
5957                 rc = PTR_ERR(o->od_mnt);
5958                 o->od_mnt = NULL;
5959                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5960                 GOTO(out, rc);
5961         }
5962
5963 #ifdef HAVE_DEV_SET_RDONLY
5964         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5965                 CERROR("%s: underlying device %s is marked as read-only. "
5966                        "Setup failed\n", name, dev);
5967                 GOTO(out_mnt, rc = -EROFS);
5968         }
5969 #endif
5970
5971         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5972                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5973                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5974                 GOTO(out_mnt, rc = -EINVAL);
5975         }
5976
5977 #ifdef LDISKFS_MOUNT_DIRDATA
5978         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5979                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5980                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5981         else if (!o->od_is_ost)
5982                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5983                       "enabling the dirdata feature. If you do not want to "
5984                       "downgrade to Lustre-1.x again, you can enable it via "
5985                       "'tune2fs -O dirdata device'\n", name, dev);
5986 #endif
5987         inode = osd_sb(o)->s_root->d_inode;
5988         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5989         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5990         if (rc != 0) {
5991                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5992                 GOTO(out_mnt, rc);
5993         }
5994
5995         if (lmd_flags & LMD_FLG_NOSCRUB)
5996                 o->od_noscrub = 1;
5997
5998         GOTO(out, rc = 0);
5999
6000 out_mnt:
6001         mntput(o->od_mnt);
6002         o->od_mnt = NULL;
6003
6004 out:
6005         if (__page)
6006                 OBD_PAGE_FREE(__page);
6007
6008         return rc;
6009 }
6010
6011 static struct lu_device *osd_device_fini(const struct lu_env *env,
6012                                          struct lu_device *d)
6013 {
6014         struct osd_device *o = osd_dev(d);
6015         ENTRY;
6016
6017         osd_shutdown(env, o);
6018         osd_procfs_fini(o);
6019         osd_scrub_cleanup(env, o);
6020         osd_obj_map_fini(o);
6021         osd_umount(env, o);
6022
6023         RETURN(NULL);
6024 }
6025
6026 static int osd_device_init0(const struct lu_env *env,
6027                             struct osd_device *o,
6028                             struct lustre_cfg *cfg)
6029 {
6030         struct lu_device        *l = osd2lu_dev(o);
6031         struct osd_thread_info *info;
6032         int                     rc;
6033         int                     cplen = 0;
6034
6035         /* if the module was re-loaded, env can loose its keys */
6036         rc = lu_env_refill((struct lu_env *) env);
6037         if (rc)
6038                 GOTO(out, rc);
6039         info = osd_oti_get(env);
6040         LASSERT(info);
6041
6042         l->ld_ops = &osd_lu_ops;
6043         o->od_dt_dev.dd_ops = &osd_dt_ops;
6044
6045         spin_lock_init(&o->od_osfs_lock);
6046         mutex_init(&o->od_otable_mutex);
6047
6048         o->od_capa_hash = init_capa_hash();
6049         if (o->od_capa_hash == NULL)
6050                 GOTO(out, rc = -ENOMEM);
6051
6052         o->od_read_cache = 1;
6053         o->od_writethrough_cache = 1;
6054         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6055
6056         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6057                         sizeof(o->od_svname));
6058         if (cplen >= sizeof(o->od_svname)) {
6059                 rc = -E2BIG;
6060                 GOTO(out_capa, rc);
6061         }
6062
6063         if (server_name_is_ost(o->od_svname))
6064                 o->od_is_ost = 1;
6065
6066         o->od_full_scrub_ratio = OFSR_DEFAULT;
6067         o->od_full_scrub_speed = FULL_SCRUB_SPEED_DEFULT;
6068         rc = osd_mount(env, o, cfg);
6069         if (rc != 0)
6070                 GOTO(out_capa, rc);
6071
6072         rc = osd_obj_map_init(env, o);
6073         if (rc != 0)
6074                 GOTO(out_mnt, rc);
6075
6076         rc = lu_site_init(&o->od_site, l);
6077         if (rc != 0)
6078                 GOTO(out_compat, rc);
6079         o->od_site.ls_bottom_dev = l;
6080
6081         rc = lu_site_init_finish(&o->od_site);
6082         if (rc != 0)
6083                 GOTO(out_site, rc);
6084
6085         /* self-repair LMA by default */
6086         o->od_lma_self_repair = 1;
6087
6088         INIT_LIST_HEAD(&o->od_ios_list);
6089         /* setup scrub, including OI files initialization */
6090         rc = osd_scrub_setup(env, o);
6091         if (rc < 0)
6092                 GOTO(out_site, rc);
6093
6094         rc = osd_procfs_init(o, o->od_svname);
6095         if (rc != 0) {
6096                 CERROR("%s: can't initialize procfs: rc = %d\n",
6097                        o->od_svname, rc);
6098                 GOTO(out_scrub, rc);
6099         }
6100
6101         LASSERT(l->ld_site->ls_linkage.next != NULL);
6102         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6103
6104         /* initialize quota slave instance */
6105         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6106                                      o->od_proc_entry);
6107         if (IS_ERR(o->od_quota_slave)) {
6108                 rc = PTR_ERR(o->od_quota_slave);
6109                 o->od_quota_slave = NULL;
6110                 GOTO(out_procfs, rc);
6111         }
6112
6113         RETURN(0);
6114
6115 out_procfs:
6116         osd_procfs_fini(o);
6117 out_scrub:
6118         osd_scrub_cleanup(env, o);
6119 out_site:
6120         lu_site_fini(&o->od_site);
6121 out_compat:
6122         osd_obj_map_fini(o);
6123 out_mnt:
6124         osd_umount(env, o);
6125 out_capa:
6126         cleanup_capa_hash(o->od_capa_hash);
6127 out:
6128         return rc;
6129 }
6130
6131 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6132                                           struct lu_device_type *t,
6133                                           struct lustre_cfg *cfg)
6134 {
6135         struct osd_device *o;
6136         int                rc;
6137
6138         OBD_ALLOC_PTR(o);
6139         if (o == NULL)
6140                 return ERR_PTR(-ENOMEM);
6141
6142         rc = dt_device_init(&o->od_dt_dev, t);
6143         if (rc == 0) {
6144                 /* Because the ctx might be revived in dt_device_init,
6145                  * refill the env here */
6146                 lu_env_refill((struct lu_env *)env);
6147                 rc = osd_device_init0(env, o, cfg);
6148                 if (rc)
6149                         dt_device_fini(&o->od_dt_dev);
6150         }
6151
6152         if (unlikely(rc != 0))
6153                 OBD_FREE_PTR(o);
6154
6155         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6156 }
6157
6158 static struct lu_device *osd_device_free(const struct lu_env *env,
6159                                          struct lu_device *d)
6160 {
6161         struct osd_device *o = osd_dev(d);
6162         ENTRY;
6163
6164         cleanup_capa_hash(o->od_capa_hash);
6165         /* XXX: make osd top device in order to release reference */
6166         d->ld_site->ls_top_dev = d;
6167         lu_site_purge(env, d->ld_site, -1);
6168         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6169                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6170                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6171         }
6172         lu_site_fini(&o->od_site);
6173         dt_device_fini(&o->od_dt_dev);
6174         OBD_FREE_PTR(o);
6175         RETURN(NULL);
6176 }
6177
6178 static int osd_process_config(const struct lu_env *env,
6179                               struct lu_device *d, struct lustre_cfg *cfg)
6180 {
6181         struct osd_device               *o = osd_dev(d);
6182         int                             rc;
6183         ENTRY;
6184
6185         switch (cfg->lcfg_command) {
6186         case LCFG_SETUP:
6187                 rc = osd_mount(env, o, cfg);
6188                 break;
6189         case LCFG_CLEANUP:
6190                 lu_dev_del_linkage(d->ld_site, d);
6191                 rc = osd_shutdown(env, o);
6192                 break;
6193         case LCFG_PARAM:
6194                 LASSERT(&o->od_dt_dev);
6195                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6196                                               cfg, &o->od_dt_dev);
6197                 if (rc > 0 || rc == -ENOSYS)
6198                         rc = class_process_proc_param(PARAM_OST,
6199                                                       lprocfs_osd_obd_vars,
6200                                                       cfg, &o->od_dt_dev);
6201                 break;
6202         default:
6203                 rc = -ENOSYS;
6204         }
6205
6206         RETURN(rc);
6207 }
6208
6209 static int osd_recovery_complete(const struct lu_env *env,
6210                                  struct lu_device *d)
6211 {
6212         struct osd_device       *osd = osd_dev(d);
6213         int                      rc = 0;
6214         ENTRY;
6215
6216         if (osd->od_quota_slave == NULL)
6217                 RETURN(0);
6218
6219         /* start qsd instance on recovery completion, this notifies the quota
6220          * slave code that we are about to process new requests now */
6221         rc = qsd_start(env, osd->od_quota_slave);
6222         RETURN(rc);
6223 }
6224
6225 /*
6226  * we use exports to track all osd users
6227  */
6228 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6229                            struct obd_device *obd, struct obd_uuid *cluuid,
6230                            struct obd_connect_data *data, void *localdata)
6231 {
6232         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6233         struct lustre_handle  conn;
6234         int                   rc;
6235         ENTRY;
6236
6237         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6238
6239         rc = class_connect(&conn, obd, cluuid);
6240         if (rc)
6241                 RETURN(rc);
6242
6243         *exp = class_conn2export(&conn);
6244
6245         spin_lock(&osd->od_osfs_lock);
6246         osd->od_connects++;
6247         spin_unlock(&osd->od_osfs_lock);
6248
6249         RETURN(0);
6250 }
6251
6252 /*
6253  * once last export (we don't count self-export) disappeared
6254  * osd can be released
6255  */
6256 static int osd_obd_disconnect(struct obd_export *exp)
6257 {
6258         struct obd_device *obd = exp->exp_obd;
6259         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6260         int                rc, release = 0;
6261         ENTRY;
6262
6263         /* Only disconnect the underlying layers on the final disconnect. */
6264         spin_lock(&osd->od_osfs_lock);
6265         osd->od_connects--;
6266         if (osd->od_connects == 0)
6267                 release = 1;
6268         spin_unlock(&osd->od_osfs_lock);
6269
6270         rc = class_disconnect(exp); /* bz 9811 */
6271
6272         if (rc == 0 && release)
6273                 class_manual_cleanup(obd);
6274         RETURN(rc);
6275 }
6276
6277 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6278                        struct lu_device *dev)
6279 {
6280         struct osd_device *osd = osd_dev(dev);
6281         int                result = 0;
6282         ENTRY;
6283
6284         if (osd->od_quota_slave != NULL) {
6285                 /* set up quota slave objects */
6286                 result = qsd_prepare(env, osd->od_quota_slave);
6287                 if (result != 0)
6288                         RETURN(result);
6289         }
6290
6291         result = osd_fid_init(env, osd);
6292
6293         RETURN(result);
6294 }
6295
6296 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6297                          struct lu_fid *fid, struct md_op_data *op_data)
6298 {
6299         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6300
6301         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6302 }
6303
6304 static const struct lu_object_operations osd_lu_obj_ops = {
6305         .loo_object_init      = osd_object_init,
6306         .loo_object_delete    = osd_object_delete,
6307         .loo_object_release   = osd_object_release,
6308         .loo_object_free      = osd_object_free,
6309         .loo_object_print     = osd_object_print,
6310         .loo_object_invariant = osd_object_invariant
6311 };
6312
6313 const struct lu_device_operations osd_lu_ops = {
6314         .ldo_object_alloc      = osd_object_alloc,
6315         .ldo_process_config    = osd_process_config,
6316         .ldo_recovery_complete = osd_recovery_complete,
6317         .ldo_prepare           = osd_prepare,
6318 };
6319
6320 static const struct lu_device_type_operations osd_device_type_ops = {
6321         .ldto_init = osd_type_init,
6322         .ldto_fini = osd_type_fini,
6323
6324         .ldto_start = osd_type_start,
6325         .ldto_stop  = osd_type_stop,
6326
6327         .ldto_device_alloc = osd_device_alloc,
6328         .ldto_device_free  = osd_device_free,
6329
6330         .ldto_device_init    = osd_device_init,
6331         .ldto_device_fini    = osd_device_fini
6332 };
6333
6334 static struct lu_device_type osd_device_type = {
6335         .ldt_tags     = LU_DEVICE_DT,
6336         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6337         .ldt_ops      = &osd_device_type_ops,
6338         .ldt_ctx_tags = LCT_LOCAL,
6339 };
6340
6341 /*
6342  * lprocfs legacy support.
6343  */
6344 static struct obd_ops osd_obd_device_ops = {
6345         .o_owner = THIS_MODULE,
6346         .o_connect      = osd_obd_connect,
6347         .o_disconnect   = osd_obd_disconnect,
6348         .o_fid_alloc    = osd_fid_alloc,
6349 };
6350
6351 static int __init osd_mod_init(void)
6352 {
6353         int rc;
6354
6355         osd_oi_mod_init();
6356
6357         rc = lu_kmem_init(ldiskfs_caches);
6358         if (rc)
6359                 return rc;
6360
6361         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6362                                  lprocfs_osd_module_vars,
6363                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6364         if (rc)
6365                 lu_kmem_fini(ldiskfs_caches);
6366         return rc;
6367 }
6368
6369 static void __exit osd_mod_exit(void)
6370 {
6371         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6372         lu_kmem_fini(ldiskfs_caches);
6373 }
6374
6375 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6376 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6377 MODULE_LICENSE("GPL");
6378
6379 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);