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, 2014, 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         if (!dt_object_exists(dt))
1734                 return -ENOENT;
1735
1736         LASSERT(!dt_object_remote(dt));
1737         LINVRNT(osd_invariant(obj));
1738
1739         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1740                 return -EACCES;
1741
1742         spin_lock(&obj->oo_guard);
1743         osd_inode_getattr(env, obj->oo_inode, attr);
1744         spin_unlock(&obj->oo_guard);
1745         return 0;
1746 }
1747
1748 static int osd_declare_attr_set(const struct lu_env *env,
1749                                 struct dt_object *dt,
1750                                 const struct lu_attr *attr,
1751                                 struct thandle *handle)
1752 {
1753         struct osd_thandle     *oh;
1754         struct osd_object      *obj;
1755         struct osd_thread_info *info = osd_oti_get(env);
1756         struct lquota_id_info  *qi = &info->oti_qi;
1757         qid_t                   uid;
1758         qid_t                   gid;
1759         long long               bspace;
1760         int                     rc = 0;
1761         bool                    enforce;
1762         ENTRY;
1763
1764         LASSERT(dt != NULL);
1765         LASSERT(handle != NULL);
1766
1767         obj = osd_dt_obj(dt);
1768         LASSERT(osd_invariant(obj));
1769
1770         oh = container_of0(handle, struct osd_thandle, ot_super);
1771         LASSERT(oh->ot_handle == NULL);
1772
1773         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1774                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1775
1776         if (attr == NULL || obj->oo_inode == NULL)
1777                 RETURN(rc);
1778
1779         bspace   = obj->oo_inode->i_blocks;
1780         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1781         bspace   = toqb(bspace);
1782
1783         /* Changing ownership is always preformed by super user, it should not
1784          * fail with EDQUOT.
1785          *
1786          * We still need to call the osd_declare_qid() to calculate the journal
1787          * credits for updating quota accounting files and to trigger quota
1788          * space adjustment once the operation is completed.*/
1789         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1790                 /* USERQUOTA */
1791                 uid = i_uid_read(obj->oo_inode);
1792                 qi->lqi_type = USRQUOTA;
1793                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1794                 /* inode accounting */
1795                 qi->lqi_is_blk = false;
1796
1797                 /* one more inode for the new uid ... */
1798                 qi->lqi_id.qid_uid = attr->la_uid;
1799                 qi->lqi_space      = 1;
1800                 /* Reserve credits for the new uid */
1801                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1802                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1803                         rc = 0;
1804                 if (rc)
1805                         RETURN(rc);
1806
1807                 /* and one less inode for the current uid */
1808                 qi->lqi_id.qid_uid = uid;
1809                 qi->lqi_space      = -1;
1810                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1811                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1812                         rc = 0;
1813                 if (rc)
1814                         RETURN(rc);
1815
1816                 /* block accounting */
1817                 qi->lqi_is_blk = true;
1818
1819                 /* more blocks for the new uid ... */
1820                 qi->lqi_id.qid_uid = attr->la_uid;
1821                 qi->lqi_space      = bspace;
1822                 /*
1823                  * Credits for the new uid has been reserved, re-use "obj"
1824                  * to save credit reservation.
1825                  */
1826                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1827                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1828                         rc = 0;
1829                 if (rc)
1830                         RETURN(rc);
1831
1832                 /* and finally less blocks for the current uid */
1833                 qi->lqi_id.qid_uid = uid;
1834                 qi->lqi_space      = -bspace;
1835                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1836                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1837                         rc = 0;
1838                 if (rc)
1839                         RETURN(rc);
1840
1841                 /* GROUP QUOTA */
1842                 gid = i_gid_read(obj->oo_inode);
1843                 qi->lqi_type = GRPQUOTA;
1844                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1845
1846                 /* inode accounting */
1847                 qi->lqi_is_blk = false;
1848
1849                 /* one more inode for the new gid ... */
1850                 qi->lqi_id.qid_gid = attr->la_gid;
1851                 qi->lqi_space      = 1;
1852                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1853                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1854                         rc = 0;
1855                 if (rc)
1856                         RETURN(rc);
1857
1858                 /* and one less inode for the current gid */
1859                 qi->lqi_id.qid_gid = gid;
1860                 qi->lqi_space      = -1;
1861                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1862                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1863                         rc = 0;
1864                 if (rc)
1865                         RETURN(rc);
1866
1867                 /* block accounting */
1868                 qi->lqi_is_blk = true;
1869
1870                 /* more blocks for the new gid ... */
1871                 qi->lqi_id.qid_gid = attr->la_gid;
1872                 qi->lqi_space      = bspace;
1873                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1874                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1875                         rc = 0;
1876                 if (rc)
1877                         RETURN(rc);
1878
1879                 /* and finally less blocks for the current gid */
1880                 qi->lqi_id.qid_gid = gid;
1881                 qi->lqi_space      = -bspace;
1882                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1883                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1884                         rc = 0;
1885                 if (rc)
1886                         RETURN(rc);
1887         }
1888
1889         RETURN(rc);
1890 }
1891
1892 static int osd_inode_setattr(const struct lu_env *env,
1893                              struct inode *inode, const struct lu_attr *attr)
1894 {
1895         __u64 bits = attr->la_valid;
1896
1897         /* Only allow set size for regular file */
1898         if (!S_ISREG(inode->i_mode))
1899                 bits &= ~(LA_SIZE | LA_BLOCKS);
1900
1901         if (bits == 0)
1902                 return 0;
1903
1904         if (bits & LA_ATIME)
1905                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1906         if (bits & LA_CTIME)
1907                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1908         if (bits & LA_MTIME)
1909                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1910         if (bits & LA_SIZE) {
1911                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1912                 i_size_write(inode, attr->la_size);
1913         }
1914
1915 #if 0
1916         /* OSD should not change "i_blocks" which is used by quota.
1917          * "i_blocks" should be changed by ldiskfs only. */
1918         if (bits & LA_BLOCKS)
1919                 inode->i_blocks = attr->la_blocks;
1920 #endif
1921         if (bits & LA_MODE)
1922                 inode->i_mode = (inode->i_mode & S_IFMT) |
1923                                 (attr->la_mode & ~S_IFMT);
1924         if (bits & LA_UID)
1925                 i_uid_write(inode, attr->la_uid);
1926         if (bits & LA_GID)
1927                 i_gid_write(inode, attr->la_gid);
1928         if (bits & LA_NLINK)
1929                 set_nlink(inode, attr->la_nlink);
1930         if (bits & LA_RDEV)
1931                 inode->i_rdev = attr->la_rdev;
1932
1933         if (bits & LA_FLAGS) {
1934                 /* always keep S_NOCMTIME */
1935                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1936                                  S_NOCMTIME;
1937         }
1938         return 0;
1939 }
1940
1941 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1942 {
1943         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1944             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1945                 struct iattr    iattr;
1946                 int             rc;
1947
1948                 ll_vfs_dq_init(inode);
1949                 iattr.ia_valid = 0;
1950                 if (attr->la_valid & LA_UID)
1951                         iattr.ia_valid |= ATTR_UID;
1952                 if (attr->la_valid & LA_GID)
1953                         iattr.ia_valid |= ATTR_GID;
1954                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1955                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1956
1957                 rc = ll_vfs_dq_transfer(inode, &iattr);
1958                 if (rc) {
1959                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1960                                "enforcement enabled on the ldiskfs "
1961                                "filesystem?\n", inode->i_sb->s_id, rc);
1962                         return rc;
1963                 }
1964         }
1965         return 0;
1966 }
1967
1968 static int osd_attr_set(const struct lu_env *env,
1969                         struct dt_object *dt,
1970                         const struct lu_attr *attr,
1971                         struct thandle *handle,
1972                         struct lustre_capa *capa)
1973 {
1974         struct osd_object *obj = osd_dt_obj(dt);
1975         struct inode      *inode;
1976         int rc;
1977
1978         if (!dt_object_exists(dt))
1979                 return -ENOENT;
1980
1981         LASSERT(handle != NULL);
1982         LASSERT(!dt_object_remote(dt));
1983         LASSERT(osd_invariant(obj));
1984
1985         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1986                 return -EACCES;
1987
1988         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1989
1990         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1991                 struct osd_thread_info  *oti  = osd_oti_get(env);
1992                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1993                 struct lu_fid           *fid1 = &oti->oti_fid;
1994                 struct osd_inode_id     *id   = &oti->oti_id;
1995                 struct iam_path_descr   *ipd;
1996                 struct iam_container    *bag;
1997                 struct osd_thandle      *oh;
1998                 int                      rc;
1999
2000                 fid_cpu_to_be(fid1, fid0);
2001                 memset(id, 1, sizeof(*id));
2002                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2003                                   fid0)->oi_dir.od_container;
2004                 ipd = osd_idx_ipd_get(env, bag);
2005                 if (unlikely(ipd == NULL))
2006                         RETURN(-ENOMEM);
2007
2008                 oh = container_of0(handle, struct osd_thandle, ot_super);
2009                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2010                                 (const struct iam_rec *)id, ipd);
2011                 osd_ipd_put(env, bag, ipd);
2012                 return(rc > 0 ? 0 : rc);
2013         }
2014
2015         inode = obj->oo_inode;
2016
2017         rc = osd_quota_transfer(inode, attr);
2018         if (rc)
2019                 return rc;
2020
2021         spin_lock(&obj->oo_guard);
2022         rc = osd_inode_setattr(env, inode, attr);
2023         spin_unlock(&obj->oo_guard);
2024
2025         if (!rc)
2026                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2027         return rc;
2028 }
2029
2030 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2031                                            struct osd_object *obj,
2032                                            const char *name, const int namelen)
2033 {
2034         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2035 }
2036
2037 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2038                       umode_t mode, struct dt_allocation_hint *hint,
2039                       struct thandle *th)
2040 {
2041         int result;
2042         struct osd_device  *osd = osd_obj2dev(obj);
2043         struct osd_thandle *oth;
2044         struct dt_object   *parent = NULL;
2045         struct inode       *inode;
2046
2047         LINVRNT(osd_invariant(obj));
2048         LASSERT(obj->oo_inode == NULL);
2049         LASSERT(obj->oo_hl_head == NULL);
2050
2051         if (S_ISDIR(mode) && ldiskfs_pdo) {
2052                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2053                 if (obj->oo_hl_head == NULL)
2054                         return -ENOMEM;
2055         }
2056
2057         oth = container_of(th, struct osd_thandle, ot_super);
2058         LASSERT(oth->ot_handle->h_transaction != NULL);
2059
2060         if (hint && hint->dah_parent)
2061                 parent = hint->dah_parent;
2062
2063         inode = ldiskfs_create_inode(oth->ot_handle,
2064                                      parent ? osd_dt_obj(parent)->oo_inode :
2065                                               osd_sb(osd)->s_root->d_inode,
2066                                      mode);
2067         if (!IS_ERR(inode)) {
2068                 /* Do not update file c/mtime in ldiskfs.
2069                  * NB: don't need any lock because no contention at this
2070                  * early stage */
2071                 inode->i_flags |= S_NOCMTIME;
2072
2073                 /* For new created object, it must be consistent,
2074                  * and it is unnecessary to scrub against it. */
2075                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2076                 obj->oo_inode = inode;
2077                 result = 0;
2078         } else {
2079                 if (obj->oo_hl_head != NULL) {
2080                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2081                         obj->oo_hl_head = NULL;
2082                 }
2083                 result = PTR_ERR(inode);
2084         }
2085         LINVRNT(osd_invariant(obj));
2086         return result;
2087 }
2088
2089 enum {
2090         OSD_NAME_LEN = 255
2091 };
2092
2093 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2094                      struct lu_attr *attr,
2095                      struct dt_allocation_hint *hint,
2096                      struct dt_object_format *dof,
2097                      struct thandle *th)
2098 {
2099         int result;
2100         struct osd_thandle *oth;
2101         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2102
2103         LASSERT(S_ISDIR(attr->la_mode));
2104
2105         oth = container_of(th, struct osd_thandle, ot_super);
2106         LASSERT(oth->ot_handle->h_transaction != NULL);
2107         result = osd_mkfile(info, obj, mode, hint, th);
2108
2109         return result;
2110 }
2111
2112 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2113                         struct lu_attr *attr,
2114                         struct dt_allocation_hint *hint,
2115                         struct dt_object_format *dof,
2116                         struct thandle *th)
2117 {
2118         int result;
2119         struct osd_thandle *oth;
2120         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2121
2122         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2123
2124         LASSERT(S_ISREG(attr->la_mode));
2125
2126         oth = container_of(th, struct osd_thandle, ot_super);
2127         LASSERT(oth->ot_handle->h_transaction != NULL);
2128
2129         result = osd_mkfile(info, obj, mode, hint, th);
2130         if (result == 0) {
2131                 LASSERT(obj->oo_inode != NULL);
2132                 if (feat->dif_flags & DT_IND_VARKEY)
2133                         result = iam_lvar_create(obj->oo_inode,
2134                                                  feat->dif_keysize_max,
2135                                                  feat->dif_ptrsize,
2136                                                  feat->dif_recsize_max,
2137                                                  oth->ot_handle);
2138                 else
2139                         result = iam_lfix_create(obj->oo_inode,
2140                                                  feat->dif_keysize_max,
2141                                                  feat->dif_ptrsize,
2142                                                  feat->dif_recsize_max,
2143                                                  oth->ot_handle);
2144
2145         }
2146         return result;
2147 }
2148
2149 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2150                      struct lu_attr *attr,
2151                      struct dt_allocation_hint *hint,
2152                      struct dt_object_format *dof,
2153                      struct thandle *th)
2154 {
2155         LASSERT(S_ISREG(attr->la_mode));
2156         return osd_mkfile(info, obj, (attr->la_mode &
2157                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2158 }
2159
2160 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2161                      struct lu_attr *attr,
2162                      struct dt_allocation_hint *hint,
2163                      struct dt_object_format *dof,
2164                      struct thandle *th)
2165 {
2166         LASSERT(S_ISLNK(attr->la_mode));
2167         return osd_mkfile(info, obj, (attr->la_mode &
2168                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2169 }
2170
2171 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2172                      struct lu_attr *attr,
2173                      struct dt_allocation_hint *hint,
2174                      struct dt_object_format *dof,
2175                      struct thandle *th)
2176 {
2177         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2178         int result;
2179
2180         LINVRNT(osd_invariant(obj));
2181         LASSERT(obj->oo_inode == NULL);
2182         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2183                 S_ISFIFO(mode) || S_ISSOCK(mode));
2184
2185         result = osd_mkfile(info, obj, mode, hint, th);
2186         if (result == 0) {
2187                 LASSERT(obj->oo_inode != NULL);
2188                 /*
2189                  * This inode should be marked dirty for i_rdev.  Currently
2190                  * that is done in the osd_attr_init().
2191                  */
2192                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2193                                    attr->la_rdev);
2194         }
2195         LINVRNT(osd_invariant(obj));
2196         return result;
2197 }
2198
2199 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2200                               struct lu_attr *,
2201                               struct dt_allocation_hint *hint,
2202                               struct dt_object_format *dof,
2203                               struct thandle *);
2204
2205 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2206 {
2207         osd_obj_type_f result;
2208
2209         switch (type) {
2210         case DFT_DIR:
2211                 result = osd_mkdir;
2212                 break;
2213         case DFT_REGULAR:
2214                 result = osd_mkreg;
2215                 break;
2216         case DFT_SYM:
2217                 result = osd_mksym;
2218                 break;
2219         case DFT_NODE:
2220                 result = osd_mknod;
2221                 break;
2222         case DFT_INDEX:
2223                 result = osd_mk_index;
2224                 break;
2225
2226         default:
2227                 LBUG();
2228                 break;
2229         }
2230         return result;
2231 }
2232
2233
2234 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2235                         struct dt_object *parent, struct dt_object *child,
2236                         umode_t child_mode)
2237 {
2238         LASSERT(ah);
2239
2240         ah->dah_parent = parent;
2241         ah->dah_mode = child_mode;
2242 }
2243
2244 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2245                           struct lu_attr *attr, struct dt_object_format *dof)
2246 {
2247         struct inode   *inode = obj->oo_inode;
2248         __u64           valid = attr->la_valid;
2249         int             result;
2250
2251         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2252
2253         if (dof->dof_type != DFT_NODE)
2254                 attr->la_valid &= ~LA_RDEV;
2255         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2256                 attr->la_valid &= ~LA_ATIME;
2257         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2258                 attr->la_valid &= ~LA_CTIME;
2259         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2260                 attr->la_valid &= ~LA_MTIME;
2261
2262         result = osd_quota_transfer(inode, attr);
2263         if (result)
2264                 return;
2265
2266         if (attr->la_valid != 0) {
2267                 result = osd_inode_setattr(info->oti_env, inode, attr);
2268                 /*
2269                  * The osd_inode_setattr() should always succeed here.  The
2270                  * only error that could be returned is EDQUOT when we are
2271                  * trying to change the UID or GID of the inode. However, this
2272                  * should not happen since quota enforcement is no longer
2273                  * enabled on ldiskfs (lquota takes care of it).
2274                  */
2275                 LASSERTF(result == 0, "%d\n", result);
2276                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2277         }
2278
2279         attr->la_valid = valid;
2280 }
2281
2282 /**
2283  * Helper function for osd_object_create()
2284  *
2285  * \retval 0, on success
2286  */
2287 static int __osd_object_create(struct osd_thread_info *info,
2288                                struct osd_object *obj, struct lu_attr *attr,
2289                                struct dt_allocation_hint *hint,
2290                                struct dt_object_format *dof,
2291                                struct thandle *th)
2292 {
2293         int     result;
2294         __u32   umask;
2295
2296         /* we drop umask so that permissions we pass are not affected */
2297         umask = current->fs->umask;
2298         current->fs->umask = 0;
2299
2300         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2301                                                   th);
2302         if (result == 0) {
2303                 osd_attr_init(info, obj, attr, dof);
2304                 osd_object_init0(obj);
2305                 /* bz 24037 */
2306                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2307                         unlock_new_inode(obj->oo_inode);
2308         }
2309
2310         /* restore previous umask value */
2311         current->fs->umask = umask;
2312
2313         return result;
2314 }
2315
2316 /**
2317  * Helper function for osd_object_create()
2318  *
2319  * \retval 0, on success
2320  */
2321 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2322                            const struct lu_fid *fid, struct thandle *th)
2323 {
2324         struct osd_thread_info *info = osd_oti_get(env);
2325         struct osd_inode_id    *id   = &info->oti_id;
2326         struct osd_device      *osd  = osd_obj2dev(obj);
2327         struct osd_thandle     *oh;
2328
2329         LASSERT(obj->oo_inode != NULL);
2330
2331         oh = container_of0(th, struct osd_thandle, ot_super);
2332         LASSERT(oh->ot_handle);
2333
2334         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2335         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2336 }
2337
2338 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2339                    u64 seq, struct lu_seq_range *range)
2340 {
2341         struct seq_server_site  *ss = osd_seq_site(osd);
2342
2343         if (fid_seq_is_idif(seq)) {
2344                 fld_range_set_ost(range);
2345                 range->lsr_index = idif_ost_idx(seq);
2346                 return 0;
2347         }
2348
2349         if (!fid_seq_in_fldb(seq)) {
2350                 fld_range_set_mdt(range);
2351                 if (ss != NULL)
2352                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2353                          * at all */
2354                         range->lsr_index = ss->ss_node_id;
2355                 return 0;
2356         }
2357
2358         LASSERT(ss != NULL);
2359         fld_range_set_any(range);
2360         /* OSD will only do local fld lookup */
2361         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2362 }
2363
2364 /*
2365  * Concurrency: no external locking is necessary.
2366  */
2367 static int osd_declare_object_create(const struct lu_env *env,
2368                                      struct dt_object *dt,
2369                                      struct lu_attr *attr,
2370                                      struct dt_allocation_hint *hint,
2371                                      struct dt_object_format *dof,
2372                                      struct thandle *handle)
2373 {
2374         struct osd_thandle      *oh;
2375         int                      rc;
2376         ENTRY;
2377
2378         LASSERT(handle != NULL);
2379
2380         oh = container_of0(handle, struct osd_thandle, ot_super);
2381         LASSERT(oh->ot_handle == NULL);
2382
2383         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2384                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2385         /* Reuse idle OI block may cause additional one OI block
2386          * to be changed. */
2387         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2388                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2389
2390         /* If this is directory, then we expect . and .. to be inserted as
2391          * well. The one directory block always needs to be created for the
2392          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2393          * block), there is no danger of needing a tree for the first block.
2394          */
2395         if (attr && S_ISDIR(attr->la_mode)) {
2396                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2397                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2398                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2399         }
2400
2401         if (!attr)
2402                 RETURN(0);
2403
2404         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2405                                    osd_dt_obj(dt), false, NULL, false);
2406         if (rc != 0)
2407                 RETURN(rc);
2408
2409         RETURN(rc);
2410 }
2411
2412 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2413                              struct lu_attr *attr,
2414                              struct dt_allocation_hint *hint,
2415                              struct dt_object_format *dof, struct thandle *th)
2416 {
2417         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2418         struct osd_object       *obj    = osd_dt_obj(dt);
2419         struct osd_thread_info  *info   = osd_oti_get(env);
2420         int result;
2421         ENTRY;
2422
2423         if (dt_object_exists(dt))
2424                 return -EEXIST;
2425
2426         LINVRNT(osd_invariant(obj));
2427         LASSERT(!dt_object_remote(dt));
2428         LASSERT(osd_write_locked(env, obj));
2429         LASSERT(th != NULL);
2430
2431         if (unlikely(fid_is_acct(fid)))
2432                 /* Quota files can't be created from the kernel any more,
2433                  * 'tune2fs -O quota' will take care of creating them */
2434                 RETURN(-EPERM);
2435
2436         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2437         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2438
2439         result = __osd_object_create(info, obj, attr, hint, dof, th);
2440         if (result == 0)
2441                 result = __osd_oi_insert(env, obj, fid, th);
2442
2443         LASSERT(ergo(result == 0,
2444                      dt_object_exists(dt) && !dt_object_remote(dt)));
2445
2446         LASSERT(osd_invariant(obj));
2447         RETURN(result);
2448 }
2449
2450 /**
2451  * Called to destroy on-disk representation of the object
2452  *
2453  * Concurrency: must be locked
2454  */
2455 static int osd_declare_object_destroy(const struct lu_env *env,
2456                                       struct dt_object *dt,
2457                                       struct thandle *th)
2458 {
2459         struct osd_object  *obj = osd_dt_obj(dt);
2460         struct inode       *inode = obj->oo_inode;
2461         struct osd_thandle *oh;
2462         int                 rc;
2463         ENTRY;
2464
2465         oh = container_of0(th, struct osd_thandle, ot_super);
2466         LASSERT(oh->ot_handle == NULL);
2467         LASSERT(inode);
2468
2469         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2470                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2471         /* Recycle idle OI leaf may cause additional three OI blocks
2472          * to be changed. */
2473         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2474                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2475         /* one less inode */
2476         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2477                                    -1, oh, obj, false, NULL, false);
2478         if (rc)
2479                 RETURN(rc);
2480         /* data to be truncated */
2481         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2482                                    0, oh, obj, true, NULL, false);
2483         RETURN(rc);
2484 }
2485
2486 static int osd_object_destroy(const struct lu_env *env,
2487                               struct dt_object *dt,
2488                               struct thandle *th)
2489 {
2490         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2491         struct osd_object      *obj = osd_dt_obj(dt);
2492         struct inode           *inode = obj->oo_inode;
2493         struct osd_device      *osd = osd_obj2dev(obj);
2494         struct osd_thandle     *oh;
2495         int                     result;
2496         ENTRY;
2497
2498         oh = container_of0(th, struct osd_thandle, ot_super);
2499         LASSERT(oh->ot_handle);
2500         LASSERT(inode);
2501         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2502
2503         if (unlikely(fid_is_acct(fid)))
2504                 RETURN(-EPERM);
2505
2506         if (S_ISDIR(inode->i_mode)) {
2507                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2508                         inode->i_nlink == 2);
2509                 /* it will check/delete the inode from remote parent,
2510                  * how to optimize it? unlink performance impaction XXX */
2511                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2512                 if (result != 0 && result != -ENOENT) {
2513                         CERROR("%s: delete inode "DFID": rc = %d\n",
2514                                osd_name(osd), PFID(fid), result);
2515                 }
2516                 spin_lock(&obj->oo_guard);
2517                 clear_nlink(inode);
2518                 spin_unlock(&obj->oo_guard);
2519                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2520         }
2521
2522         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2523
2524         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2525                                OI_CHECK_FLD);
2526
2527         /* XXX: add to ext3 orphan list */
2528         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2529
2530         /* not needed in the cache anymore */
2531         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2532
2533         RETURN(0);
2534 }
2535
2536 /**
2537  * Put the fid into lustre_mdt_attrs, and then place the structure
2538  * inode's ea. This fid should not be altered during the life time
2539  * of the inode.
2540  *
2541  * \retval +ve, on success
2542  * \retval -ve, on error
2543  *
2544  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2545  */
2546 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2547                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2548 {
2549         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2550         int                      rc;
2551         ENTRY;
2552
2553         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2554                 RETURN(0);
2555
2556         lustre_lma_init(lma, fid, compat, incompat);
2557         lustre_lma_swab(lma);
2558
2559         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2560                              XATTR_CREATE);
2561         /* LMA may already exist, but we need to check that all the
2562          * desired compat/incompat flags have been added. */
2563         if (unlikely(rc == -EEXIST)) {
2564                 if (compat == 0 && incompat == 0)
2565                         RETURN(0);
2566
2567                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2568                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2569                                      LMA_OLD_SIZE);
2570                 if (rc <= 0)
2571                         RETURN(-EINVAL);
2572
2573                 lustre_lma_swab(lma);
2574                 if (!(~lma->lma_compat & compat) &&
2575                     !(~lma->lma_incompat & incompat))
2576                         RETURN(0);
2577
2578                 lma->lma_compat |= compat;
2579                 lma->lma_incompat |= incompat;
2580                 lustre_lma_swab(lma);
2581                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2582                                      sizeof(*lma), XATTR_REPLACE);
2583         }
2584
2585         RETURN(rc);
2586 }
2587
2588 /**
2589  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2590  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2591  * To have compatilibility with 1.8 ldiskfs driver we need to have
2592  * magic number at start of fid data.
2593  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2594  * its inmemory API.
2595  */
2596 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2597                                          const struct lu_fid *fid)
2598 {
2599         if (!fid_is_namespace_visible(fid) ||
2600             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2601                 param->edp_magic = 0;
2602                 return;
2603         }
2604
2605         param->edp_magic = LDISKFS_LUFID_MAGIC;
2606         param->edp_len =  sizeof(struct lu_fid) + 1;
2607         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2608 }
2609
2610 /**
2611  * Try to read the fid from inode ea into dt_rec.
2612  *
2613  * \param fid object fid.
2614  *
2615  * \retval 0 on success
2616  */
2617 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2618                           __u32 ino, struct lu_fid *fid,
2619                           struct osd_inode_id *id)
2620 {
2621         struct osd_thread_info *info  = osd_oti_get(env);
2622         struct inode           *inode;
2623         ENTRY;
2624
2625         osd_id_gen(id, ino, OSD_OII_NOGEN);
2626         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2627         if (IS_ERR(inode))
2628                 RETURN(PTR_ERR(inode));
2629
2630         iput(inode);
2631         RETURN(0);
2632 }
2633
2634 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2635                                         struct inode *dir,
2636                                         struct inode *parent_dir,
2637                                         const struct lu_fid *dot_fid,
2638                                         const struct lu_fid *dot_dot_fid,
2639                                         struct osd_thandle *oth)
2640 {
2641         struct ldiskfs_dentry_param *dot_ldp;
2642         struct ldiskfs_dentry_param *dot_dot_ldp;
2643
2644         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2645         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2646
2647         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2648         dot_ldp->edp_magic = 0;
2649         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2650                                         dir, dot_ldp, dot_dot_ldp);
2651 }
2652
2653 /**
2654  * Create an local agent inode for remote entry
2655  */
2656 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2657                                                   struct osd_device *osd,
2658                                                   struct osd_object *pobj,
2659                                                   const struct lu_fid *fid,
2660                                                   __u32 type,
2661                                                   struct thandle *th)
2662 {
2663         struct osd_thread_info  *info = osd_oti_get(env);
2664         struct inode            *local;
2665         struct osd_thandle      *oh;
2666         int                     rc;
2667         ENTRY;
2668
2669         LASSERT(th);
2670         oh = container_of(th, struct osd_thandle, ot_super);
2671         LASSERT(oh->ot_handle->h_transaction != NULL);
2672
2673         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2674         if (IS_ERR(local)) {
2675                 CERROR("%s: create local error %d\n", osd_name(osd),
2676                        (int)PTR_ERR(local));
2677                 RETURN(local);
2678         }
2679
2680         /* Set special LMA flag for local agent inode */
2681         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2682         if (rc != 0) {
2683                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2684                        osd_name(osd), PFID(fid), rc);
2685                 RETURN(ERR_PTR(rc));
2686         }
2687
2688         if (!S_ISDIR(type))
2689                 RETURN(local);
2690
2691         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2692                                          lu_object_fid(&pobj->oo_dt.do_lu),
2693                                          fid, oh);
2694         if (rc != 0) {
2695                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2696                         osd_name(osd), PFID(fid), rc);
2697                 RETURN(ERR_PTR(rc));
2698         }
2699
2700         RETURN(local);
2701 }
2702
2703 /**
2704  * Delete local agent inode for remote entry
2705  */
2706 static int osd_delete_local_agent_inode(const struct lu_env *env,
2707                                         struct osd_device *osd,
2708                                         const struct lu_fid *fid,
2709                                         __u32 ino, struct osd_thandle *oh)
2710 {
2711         struct osd_thread_info  *oti = osd_oti_get(env);
2712         struct osd_inode_id     *id = &oti->oti_id;
2713         struct inode            *inode;
2714         ENTRY;
2715
2716         id->oii_ino = le32_to_cpu(ino);
2717         id->oii_gen = OSD_OII_NOGEN;
2718         inode = osd_iget(oti, osd, id);
2719         if (IS_ERR(inode)) {
2720                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2721                        PFID(fid), id->oii_ino, id->oii_gen);
2722                 RETURN(PTR_ERR(inode));
2723         }
2724
2725         clear_nlink(inode);
2726         mark_inode_dirty(inode);
2727         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2728                 osd_name(osd), PFID(fid), inode->i_ino);
2729         iput(inode);
2730         RETURN(0);
2731 }
2732
2733 /**
2734  * OSD layer object create function for interoperability mode (b11826).
2735  * This is mostly similar to osd_object_create(). Only difference being, fid is
2736  * inserted into inode ea here.
2737  *
2738  * \retval   0, on success
2739  * \retval -ve, on error
2740  */
2741 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2742                                 struct lu_attr *attr,
2743                                 struct dt_allocation_hint *hint,
2744                                 struct dt_object_format *dof,
2745                                 struct thandle *th)
2746 {
2747         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2748         struct osd_object       *obj    = osd_dt_obj(dt);
2749         struct osd_thread_info  *info   = osd_oti_get(env);
2750         int                      result;
2751
2752         ENTRY;
2753
2754         if (dt_object_exists(dt))
2755                 RETURN(-EEXIST);
2756
2757         LASSERT(osd_invariant(obj));
2758         LASSERT(!dt_object_remote(dt));
2759         LASSERT(osd_write_locked(env, obj));
2760         LASSERT(th != NULL);
2761
2762         if (unlikely(fid_is_acct(fid)))
2763                 /* Quota files can't be created from the kernel any more,
2764                  * 'tune2fs -O quota' will take care of creating them */
2765                 RETURN(-EPERM);
2766
2767         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2768         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2769
2770         result = __osd_object_create(info, obj, attr, hint, dof, th);
2771         if (result == 0)
2772                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2773                                 fid_is_on_ost(info, osd_obj2dev(obj),
2774                                               fid, OI_CHECK_FLD) ?
2775                                 LMAC_FID_ON_OST : 0, 0);
2776
2777         if (result == 0)
2778                 result = __osd_oi_insert(env, obj, fid, th);
2779
2780         LASSERT(ergo(result == 0,
2781                      dt_object_exists(dt) && !dt_object_remote(dt)));
2782         LINVRNT(osd_invariant(obj));
2783         RETURN(result);
2784 }
2785
2786 static int osd_declare_object_ref_add(const struct lu_env *env,
2787                                       struct dt_object *dt,
2788                                       struct thandle *handle)
2789 {
2790         struct osd_thandle       *oh;
2791
2792         /* it's possible that object doesn't exist yet */
2793         LASSERT(handle != NULL);
2794
2795         oh = container_of0(handle, struct osd_thandle, ot_super);
2796         LASSERT(oh->ot_handle == NULL);
2797
2798         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2799                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2800
2801         return 0;
2802 }
2803
2804 /*
2805  * Concurrency: @dt is write locked.
2806  */
2807 static int osd_object_ref_add(const struct lu_env *env,
2808                               struct dt_object *dt, struct thandle *th)
2809 {
2810         struct osd_object  *obj = osd_dt_obj(dt);
2811         struct inode       *inode = obj->oo_inode;
2812         struct osd_thandle *oh;
2813         int                 rc = 0;
2814
2815         if (!dt_object_exists(dt))
2816                 return -ENOENT;
2817
2818         LINVRNT(osd_invariant(obj));
2819         LASSERT(!dt_object_remote(dt));
2820         LASSERT(osd_write_locked(env, obj));
2821         LASSERT(th != NULL);
2822
2823         oh = container_of0(th, struct osd_thandle, ot_super);
2824         LASSERT(oh->ot_handle != NULL);
2825
2826         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2827
2828         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2829                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2830         /*
2831          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2832          * (65000) subdirectories by storing "1" in i_nlink if the link count
2833          * would otherwise overflow. Directory tranversal tools understand
2834          * that (st_nlink == 1) indicates that the filesystem dose not track
2835          * hard links count on the directory, and will not abort subdirectory
2836          * scanning early once (st_nlink - 2) subdirs have been found.
2837          *
2838          * This also has to properly handle the case of inodes with nlink == 0
2839          * in case they are being linked into the PENDING directory
2840          */
2841         spin_lock(&obj->oo_guard);
2842         if (unlikely(inode->i_nlink == 0))
2843                 /* inc_nlink from 0 may cause WARN_ON */
2844                 set_nlink(inode, 1);
2845         else {
2846                 ldiskfs_inc_count(oh->ot_handle, inode);
2847                 if (!S_ISDIR(inode->i_mode))
2848                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2849         }
2850         spin_unlock(&obj->oo_guard);
2851
2852         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2853         LINVRNT(osd_invariant(obj));
2854
2855         return rc;
2856 }
2857
2858 static int osd_declare_object_ref_del(const struct lu_env *env,
2859                                       struct dt_object *dt,
2860                                       struct thandle *handle)
2861 {
2862         struct osd_thandle *oh;
2863
2864         LASSERT(!dt_object_remote(dt));
2865         LASSERT(handle != NULL);
2866
2867         oh = container_of0(handle, struct osd_thandle, ot_super);
2868         LASSERT(oh->ot_handle == NULL);
2869
2870         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2871                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2872
2873         return 0;
2874 }
2875
2876 /*
2877  * Concurrency: @dt is write locked.
2878  */
2879 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2880                               struct thandle *th)
2881 {
2882         struct osd_object       *obj = osd_dt_obj(dt);
2883         struct inode            *inode = obj->oo_inode;
2884         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2885         struct osd_thandle      *oh;
2886
2887         if (!dt_object_exists(dt))
2888                 return -ENOENT;
2889
2890         LINVRNT(osd_invariant(obj));
2891         LASSERT(!dt_object_remote(dt));
2892         LASSERT(osd_write_locked(env, obj));
2893         LASSERT(th != NULL);
2894
2895         oh = container_of0(th, struct osd_thandle, ot_super);
2896         LASSERT(oh->ot_handle != NULL);
2897
2898         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2899
2900         spin_lock(&obj->oo_guard);
2901         /* That can be result of upgrade from old Lustre version and
2902          * applied only to local files.  Just skip this ref_del call.
2903          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2904         if (inode->i_nlink == 0) {
2905                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2906                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2907                              ", maybe an upgraded file? (LU-3915)\n",
2908                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2909                 spin_unlock(&obj->oo_guard);
2910                 return 0;
2911         }
2912
2913         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2914                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2915
2916         ldiskfs_dec_count(oh->ot_handle, inode);
2917         spin_unlock(&obj->oo_guard);
2918
2919         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2920         LINVRNT(osd_invariant(obj));
2921
2922         return 0;
2923 }
2924
2925 /*
2926  * Get the 64-bit version for an inode.
2927  */
2928 static int osd_object_version_get(const struct lu_env *env,
2929                                   struct dt_object *dt, dt_obj_version_t *ver)
2930 {
2931         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2932
2933         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2934                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2935         *ver = LDISKFS_I(inode)->i_fs_version;
2936         return 0;
2937 }
2938
2939 /*
2940  * Concurrency: @dt is read locked.
2941  */
2942 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2943                          struct lu_buf *buf, const char *name,
2944                          struct lustre_capa *capa)
2945 {
2946         struct osd_object      *obj    = osd_dt_obj(dt);
2947         struct inode           *inode  = obj->oo_inode;
2948         struct osd_thread_info *info   = osd_oti_get(env);
2949         struct dentry          *dentry = &info->oti_obj_dentry;
2950
2951         /* version get is not real XATTR but uses xattr API */
2952         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2953                 /* for version we are just using xattr API but change inode
2954                  * field instead */
2955                 if (buf->lb_len == 0)
2956                         return sizeof(dt_obj_version_t);
2957
2958                 if (buf->lb_len < sizeof(dt_obj_version_t))
2959                         return -ERANGE;
2960
2961                 osd_object_version_get(env, dt, buf->lb_buf);
2962
2963                 return sizeof(dt_obj_version_t);
2964         }
2965
2966         if (!dt_object_exists(dt))
2967                 return -ENOENT;
2968
2969         LASSERT(!dt_object_remote(dt));
2970         LASSERT(inode->i_op != NULL);
2971         LASSERT(inode->i_op->getxattr != NULL);
2972
2973         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2974                 return -EACCES;
2975
2976         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2977 }
2978
2979
2980 static int osd_declare_xattr_set(const struct lu_env *env,
2981                                  struct dt_object *dt,
2982                                  const struct lu_buf *buf, const char *name,
2983                                  int fl, struct thandle *handle)
2984 {
2985         struct osd_thandle *oh;
2986         int credits;
2987         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2988
2989         LASSERT(handle != NULL);
2990
2991         oh = container_of0(handle, struct osd_thandle, ot_super);
2992         LASSERT(oh->ot_handle == NULL);
2993
2994         /* optimistic optimization: LMA is set first and usually fit inode */
2995         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2996                 if (dt_object_exists(dt))
2997                         credits = 0;
2998                 else
2999                         credits = 1;
3000         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3001                 credits = 1;
3002         } else {
3003                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3004                 if (buf && buf->lb_len > sb->s_blocksize) {
3005                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
3006                                         sb->s_blocksize_bits;
3007                 }
3008                 /*
3009                  * xattr set may involve inode quota change, reserve credits for
3010                  * dquot_initialize()
3011                  */
3012                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3013         }
3014
3015         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3016
3017         return 0;
3018 }
3019
3020 /*
3021  * Set the 64-bit version for object
3022  */
3023 static void osd_object_version_set(const struct lu_env *env,
3024                                    struct dt_object *dt,
3025                                    dt_obj_version_t *new_version)
3026 {
3027         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3028
3029         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3030                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3031
3032         LDISKFS_I(inode)->i_fs_version = *new_version;
3033         /** Version is set after all inode operations are finished,
3034          *  so we should mark it dirty here */
3035         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3036 }
3037
3038 /*
3039  * Concurrency: @dt is write locked.
3040  */
3041 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3042                          const struct lu_buf *buf, const char *name, int fl,
3043                          struct thandle *handle, struct lustre_capa *capa)
3044 {
3045         struct osd_object      *obj      = osd_dt_obj(dt);
3046         struct inode           *inode    = obj->oo_inode;
3047         struct osd_thread_info *info     = osd_oti_get(env);
3048         int                     fs_flags = 0;
3049         ENTRY;
3050
3051         LASSERT(handle != NULL);
3052
3053         /* version set is not real XATTR */
3054         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3055                 /* for version we are just using xattr API but change inode
3056                  * field instead */
3057                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3058                 osd_object_version_set(env, dt, buf->lb_buf);
3059                 return sizeof(dt_obj_version_t);
3060         }
3061
3062         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3063                 return -EACCES;
3064
3065         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3066                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3067
3068         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3069         if (fl & LU_XATTR_REPLACE)
3070                 fs_flags |= XATTR_REPLACE;
3071
3072         if (fl & LU_XATTR_CREATE)
3073                 fs_flags |= XATTR_CREATE;
3074
3075         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3076                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3077                 int                      rc;
3078
3079                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3080                 if (rc != 0)
3081                         RETURN(rc);
3082
3083                 lma->lma_incompat |= LMAI_STRIPED;
3084                 lustre_lma_swab(lma);
3085                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3086                                      sizeof(*lma), XATTR_REPLACE);
3087                 if (rc != 0)
3088                         RETURN(rc);
3089         }
3090
3091         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3092             strcmp(name, XATTR_NAME_LINK) == 0)
3093                 return -ENOSPC;
3094
3095         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3096                                fs_flags);
3097 }
3098
3099 /*
3100  * Concurrency: @dt is read locked.
3101  */
3102 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3103                           const struct lu_buf *buf, struct lustre_capa *capa)
3104 {
3105         struct osd_object      *obj    = osd_dt_obj(dt);
3106         struct inode           *inode  = obj->oo_inode;
3107         struct osd_thread_info *info   = osd_oti_get(env);
3108         struct dentry          *dentry = &info->oti_obj_dentry;
3109
3110         if (!dt_object_exists(dt))
3111                 return -ENOENT;
3112
3113         LASSERT(!dt_object_remote(dt));
3114         LASSERT(inode->i_op != NULL);
3115         LASSERT(inode->i_op->listxattr != NULL);
3116
3117         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3118                 return -EACCES;
3119
3120         dentry->d_inode = inode;
3121         dentry->d_sb = inode->i_sb;
3122         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3123 }
3124
3125 static int osd_declare_xattr_del(const struct lu_env *env,
3126                                  struct dt_object *dt, const char *name,
3127                                  struct thandle *handle)
3128 {
3129         struct osd_thandle *oh;
3130         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3131
3132         LASSERT(!dt_object_remote(dt));
3133         LASSERT(handle != NULL);
3134
3135         oh = container_of0(handle, struct osd_thandle, ot_super);
3136         LASSERT(oh->ot_handle == NULL);
3137
3138         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3139                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3140         /*
3141          * xattr del may involve inode quota change, reserve credits for
3142          * dquot_initialize()
3143          */
3144         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3145
3146         return 0;
3147 }
3148
3149 /*
3150  * Concurrency: @dt is write locked.
3151  */
3152 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3153                          const char *name, struct thandle *handle,
3154                          struct lustre_capa *capa)
3155 {
3156         struct osd_object      *obj    = osd_dt_obj(dt);
3157         struct inode           *inode  = obj->oo_inode;
3158         struct osd_thread_info *info   = osd_oti_get(env);
3159         struct dentry          *dentry = &info->oti_obj_dentry;
3160         int                     rc;
3161
3162         if (!dt_object_exists(dt))
3163                 return -ENOENT;
3164
3165         LASSERT(!dt_object_remote(dt));
3166         LASSERT(inode->i_op != NULL);
3167         LASSERT(inode->i_op->removexattr != NULL);
3168         LASSERT(handle != NULL);
3169
3170         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3171                 return -EACCES;
3172
3173         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3174
3175         ll_vfs_dq_init(inode);
3176         dentry->d_inode = inode;
3177         dentry->d_sb = inode->i_sb;
3178         rc = inode->i_op->removexattr(dentry, name);
3179         return rc;
3180 }
3181
3182 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3183                                      struct dt_object *dt,
3184                                      struct lustre_capa *old, __u64 opc)
3185 {
3186         struct osd_thread_info *info = osd_oti_get(env);
3187         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3188         struct osd_object *obj = osd_dt_obj(dt);
3189         struct osd_device *osd = osd_obj2dev(obj);
3190         struct lustre_capa_key *key = &info->oti_capa_key;
3191         struct lustre_capa *capa = &info->oti_capa;
3192         struct obd_capa *oc;
3193         struct lu_capainfo *lci;
3194         int rc;
3195         ENTRY;
3196
3197         if (!osd->od_fl_capa)
3198                 RETURN(ERR_PTR(-ENOENT));
3199
3200         if (!dt_object_exists(dt))
3201                 RETURN(ERR_PTR(-ENOENT));
3202
3203         LASSERT(!dt_object_remote(dt));
3204         LINVRNT(osd_invariant(obj));
3205
3206         /* renewal sanity check */
3207         if (old && osd_object_auth(env, dt, old, opc))
3208                 RETURN(ERR_PTR(-EACCES));
3209
3210         lci = lu_capainfo_get(env);
3211         if (unlikely(lci == NULL))
3212                 RETURN(ERR_PTR(-ENOENT));
3213
3214         switch (lci->lci_auth) {
3215         case LC_ID_NONE:
3216                 RETURN(NULL);
3217         case LC_ID_PLAIN:
3218                 capa->lc_uid = i_uid_read(obj->oo_inode);
3219                 capa->lc_gid = i_gid_read(obj->oo_inode);
3220                 capa->lc_flags = LC_ID_PLAIN;
3221                 break;
3222         case LC_ID_CONVERT: {
3223                 __u32 d[4], s[4];
3224
3225                 s[0] = i_uid_read(obj->oo_inode);
3226                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3227                 s[2] = i_uid_read(obj->oo_inode);
3228                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3229                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3230                 if (unlikely(rc))
3231                         RETURN(ERR_PTR(rc));
3232
3233                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3234                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3235                 capa->lc_flags = LC_ID_CONVERT;
3236                 break;
3237         }
3238         default:
3239                 RETURN(ERR_PTR(-EINVAL));
3240         }
3241
3242         capa->lc_fid = *fid;
3243         capa->lc_opc = opc;
3244         capa->lc_flags |= osd->od_capa_alg << 24;
3245         capa->lc_timeout = osd->od_capa_timeout;
3246         capa->lc_expiry = 0;
3247
3248         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3249         if (oc) {
3250                 LASSERT(!capa_is_expired(oc));
3251                 RETURN(oc);
3252         }
3253
3254         spin_lock(&capa_lock);
3255         *key = osd->od_capa_keys[1];
3256         spin_unlock(&capa_lock);
3257
3258         capa->lc_keyid = key->lk_keyid;
3259         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3260
3261         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3262         if (rc) {
3263                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3264                 RETURN(ERR_PTR(rc));
3265         }
3266
3267         oc = capa_add(osd->od_capa_hash, capa);
3268         RETURN(oc);
3269 }
3270
3271 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3272                            __u64 start, __u64 end)
3273 {
3274         struct osd_object       *obj    = osd_dt_obj(dt);
3275         struct inode            *inode  = obj->oo_inode;
3276         struct osd_thread_info  *info   = osd_oti_get(env);
3277         struct dentry           *dentry = &info->oti_obj_dentry;
3278         struct file             *file   = &info->oti_file;
3279         int                     rc;
3280
3281         ENTRY;
3282
3283         dentry->d_inode = inode;
3284         dentry->d_sb = inode->i_sb;
3285         file->f_dentry = dentry;
3286         file->f_mapping = inode->i_mapping;
3287         file->f_op = inode->i_fop;
3288         set_file_inode(file, inode);
3289
3290         rc = ll_vfs_fsync_range(file, start, end, 0);
3291
3292         RETURN(rc);
3293 }
3294
3295 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3296                         void **data)
3297 {
3298         struct osd_object *obj = osd_dt_obj(dt);
3299         ENTRY;
3300
3301         *data = (void *)obj->oo_inode;
3302         RETURN(0);
3303 }
3304
3305 /*
3306  * Index operations.
3307  */
3308
3309 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3310                            const struct dt_index_features *feat)
3311 {
3312         struct iam_descr *descr;
3313
3314         if (osd_object_is_root(o))
3315                 return feat == &dt_directory_features;
3316
3317         LASSERT(o->oo_dir != NULL);
3318
3319         descr = o->oo_dir->od_container.ic_descr;
3320         if (feat == &dt_directory_features) {
3321                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3322                         return 1;
3323                 else
3324                         return 0;
3325         } else {
3326                 return
3327                         feat->dif_keysize_min <= descr->id_key_size &&
3328                         descr->id_key_size <= feat->dif_keysize_max &&
3329                         feat->dif_recsize_min <= descr->id_rec_size &&
3330                         descr->id_rec_size <= feat->dif_recsize_max &&
3331                         !(feat->dif_flags & (DT_IND_VARKEY |
3332                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3333                         ergo(feat->dif_flags & DT_IND_UPDATE,
3334                              1 /* XXX check that object (and file system) is
3335                                 * writable */);
3336         }
3337 }
3338
3339 static int osd_iam_container_init(const struct lu_env *env,
3340                                   struct osd_object *obj,
3341                                   struct osd_directory *dir)
3342 {
3343         struct iam_container *bag = &dir->od_container;
3344         int result;
3345
3346         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3347         if (result != 0)
3348                 return result;
3349
3350         result = iam_container_setup(bag);
3351         if (result == 0)
3352                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3353         else
3354                 iam_container_fini(bag);
3355
3356         return result;
3357 }
3358
3359
3360 /*
3361  * Concurrency: no external locking is necessary.
3362  */
3363 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3364                          const struct dt_index_features *feat)
3365 {
3366         int                      result;
3367         int                      skip_iam = 0;
3368         struct osd_object       *obj = osd_dt_obj(dt);
3369
3370         LINVRNT(osd_invariant(obj));
3371
3372         if (osd_object_is_root(obj)) {
3373                 dt->do_index_ops = &osd_index_ea_ops;
3374                 result = 0;
3375         } else if (feat == &dt_directory_features) {
3376                 dt->do_index_ops = &osd_index_ea_ops;
3377                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3378                         result = 0;
3379                 else
3380                         result = -ENOTDIR;
3381                 skip_iam = 1;
3382         } else if (unlikely(feat == &dt_otable_features)) {
3383                 dt->do_index_ops = &osd_otable_ops;
3384                 return 0;
3385         } else if (unlikely(feat == &dt_acct_features)) {
3386                 dt->do_index_ops = &osd_acct_index_ops;
3387                 result = 0;
3388                 skip_iam = 1;
3389         } else if (!osd_has_index(obj)) {
3390                 struct osd_directory *dir;
3391
3392                 OBD_ALLOC_PTR(dir);
3393                 if (dir != NULL) {
3394
3395                         spin_lock(&obj->oo_guard);
3396                         if (obj->oo_dir == NULL)
3397                                 obj->oo_dir = dir;
3398                         else
3399                                 /*
3400                                  * Concurrent thread allocated container data.
3401                                  */
3402                                 OBD_FREE_PTR(dir);
3403                         spin_unlock(&obj->oo_guard);
3404                         /*
3405                          * Now, that we have container data, serialize its
3406                          * initialization.
3407                          */
3408                         down_write(&obj->oo_ext_idx_sem);
3409                         /*
3410                          * recheck under lock.
3411                          */
3412                         if (!osd_has_index(obj))
3413                                 result = osd_iam_container_init(env, obj,
3414                                                                 obj->oo_dir);
3415                         else
3416                                 result = 0;
3417                         up_write(&obj->oo_ext_idx_sem);
3418                 } else {
3419                         result = -ENOMEM;
3420                 }
3421         } else {
3422                 result = 0;
3423         }
3424
3425         if (result == 0 && skip_iam == 0) {
3426                 if (!osd_iam_index_probe(env, obj, feat))
3427                         result = -ENOTDIR;
3428         }
3429         LINVRNT(osd_invariant(obj));
3430
3431         if (result == 0 && is_quota_glb_feat(feat) &&
3432             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3433                 result = osd_quota_migration(env, dt, feat);
3434
3435         return result;
3436 }
3437
3438 static int osd_otable_it_attr_get(const struct lu_env *env,
3439                                  struct dt_object *dt,
3440                                  struct lu_attr *attr,
3441                                  struct lustre_capa *capa)
3442 {
3443         attr->la_valid = 0;
3444         return 0;
3445 }
3446
3447 static const struct dt_object_operations osd_obj_ops = {
3448         .do_read_lock         = osd_object_read_lock,
3449         .do_write_lock        = osd_object_write_lock,
3450         .do_read_unlock       = osd_object_read_unlock,
3451         .do_write_unlock      = osd_object_write_unlock,
3452         .do_write_locked      = osd_object_write_locked,
3453         .do_attr_get          = osd_attr_get,
3454         .do_declare_attr_set  = osd_declare_attr_set,
3455         .do_attr_set          = osd_attr_set,
3456         .do_ah_init           = osd_ah_init,
3457         .do_declare_create    = osd_declare_object_create,
3458         .do_create            = osd_object_create,
3459         .do_declare_destroy   = osd_declare_object_destroy,
3460         .do_destroy           = osd_object_destroy,
3461         .do_index_try         = osd_index_try,
3462         .do_declare_ref_add   = osd_declare_object_ref_add,
3463         .do_ref_add           = osd_object_ref_add,
3464         .do_declare_ref_del   = osd_declare_object_ref_del,
3465         .do_ref_del           = osd_object_ref_del,
3466         .do_xattr_get         = osd_xattr_get,
3467         .do_declare_xattr_set = osd_declare_xattr_set,
3468         .do_xattr_set         = osd_xattr_set,
3469         .do_declare_xattr_del = osd_declare_xattr_del,
3470         .do_xattr_del         = osd_xattr_del,
3471         .do_xattr_list        = osd_xattr_list,
3472         .do_capa_get          = osd_capa_get,
3473         .do_object_sync       = osd_object_sync,
3474         .do_data_get          = osd_data_get,
3475 };
3476
3477 /**
3478  * dt_object_operations for interoperability mode
3479  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3480  */
3481 static const struct dt_object_operations osd_obj_ea_ops = {
3482         .do_read_lock         = osd_object_read_lock,
3483         .do_write_lock        = osd_object_write_lock,
3484         .do_read_unlock       = osd_object_read_unlock,
3485         .do_write_unlock      = osd_object_write_unlock,
3486         .do_write_locked      = osd_object_write_locked,
3487         .do_attr_get          = osd_attr_get,
3488         .do_declare_attr_set  = osd_declare_attr_set,
3489         .do_attr_set          = osd_attr_set,
3490         .do_ah_init           = osd_ah_init,
3491         .do_declare_create    = osd_declare_object_create,
3492         .do_create            = osd_object_ea_create,
3493         .do_declare_destroy   = osd_declare_object_destroy,
3494         .do_destroy           = osd_object_destroy,
3495         .do_index_try         = osd_index_try,
3496         .do_declare_ref_add   = osd_declare_object_ref_add,
3497         .do_ref_add           = osd_object_ref_add,
3498         .do_declare_ref_del   = osd_declare_object_ref_del,
3499         .do_ref_del           = osd_object_ref_del,
3500         .do_xattr_get         = osd_xattr_get,
3501         .do_declare_xattr_set = osd_declare_xattr_set,
3502         .do_xattr_set         = osd_xattr_set,
3503         .do_declare_xattr_del = osd_declare_xattr_del,
3504         .do_xattr_del         = osd_xattr_del,
3505         .do_xattr_list        = osd_xattr_list,
3506         .do_capa_get          = osd_capa_get,
3507         .do_object_sync       = osd_object_sync,
3508         .do_data_get          = osd_data_get,
3509 };
3510
3511 static const struct dt_object_operations osd_obj_otable_it_ops = {
3512         .do_attr_get    = osd_otable_it_attr_get,
3513         .do_index_try   = osd_index_try,
3514 };
3515
3516 static int osd_index_declare_iam_delete(const struct lu_env *env,
3517                                         struct dt_object *dt,
3518                                         const struct dt_key *key,
3519                                         struct thandle *handle)
3520 {
3521         struct osd_thandle    *oh;
3522
3523         oh = container_of0(handle, struct osd_thandle, ot_super);
3524         LASSERT(oh->ot_handle == NULL);
3525
3526         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3527                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3528
3529         return 0;
3530 }
3531
3532 /**
3533  *      delete a (key, value) pair from index \a dt specified by \a key
3534  *
3535  *      \param  dt      osd index object
3536  *      \param  key     key for index
3537  *      \param  rec     record reference
3538  *      \param  handle  transaction handler
3539  *
3540  *      \retval  0  success
3541  *      \retval -ve   failure
3542  */
3543 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3544                                 const struct dt_key *key,
3545                                 struct thandle *handle,
3546                                 struct lustre_capa *capa)
3547 {
3548         struct osd_thread_info *oti = osd_oti_get(env);
3549         struct osd_object      *obj = osd_dt_obj(dt);
3550         struct osd_thandle     *oh;
3551         struct iam_path_descr  *ipd;
3552         struct iam_container   *bag = &obj->oo_dir->od_container;
3553         int                     rc;
3554         ENTRY;
3555
3556         if (!dt_object_exists(dt))
3557                 RETURN(-ENOENT);
3558
3559         LINVRNT(osd_invariant(obj));
3560         LASSERT(!dt_object_remote(dt));
3561         LASSERT(bag->ic_object == obj->oo_inode);
3562         LASSERT(handle != NULL);
3563
3564         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3565                 RETURN(-EACCES);
3566
3567         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3568
3569         ipd = osd_idx_ipd_get(env, bag);
3570         if (unlikely(ipd == NULL))
3571                 RETURN(-ENOMEM);
3572
3573         oh = container_of0(handle, struct osd_thandle, ot_super);
3574         LASSERT(oh->ot_handle != NULL);
3575         LASSERT(oh->ot_handle->h_transaction != NULL);
3576
3577         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3578                 /* swab quota uid/gid provided by caller */
3579                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3580                 key = (const struct dt_key *)&oti->oti_quota_id;
3581         }
3582
3583         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3584         osd_ipd_put(env, bag, ipd);
3585         LINVRNT(osd_invariant(obj));
3586         RETURN(rc);
3587 }
3588
3589 static int osd_index_declare_ea_delete(const struct lu_env *env,
3590                                        struct dt_object *dt,
3591                                        const struct dt_key *key,
3592                                        struct thandle *handle)
3593 {
3594         struct osd_thandle *oh;
3595         struct inode       *inode;
3596         int                 rc;
3597         ENTRY;
3598
3599         LASSERT(!dt_object_remote(dt));
3600         LASSERT(handle != NULL);
3601
3602         oh = container_of0(handle, struct osd_thandle, ot_super);
3603         LASSERT(oh->ot_handle == NULL);
3604
3605         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3606                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3607
3608         inode = osd_dt_obj(dt)->oo_inode;
3609         LASSERT(inode);
3610
3611         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3612                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3613         RETURN(rc);
3614 }
3615
3616 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3617                                           struct dt_rec *fid)
3618 {
3619         struct osd_fid_pack *rec;
3620         int                  rc = -ENODATA;
3621
3622         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3623                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3624                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3625         }
3626         return rc;
3627 }
3628
3629 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3630                           const struct lu_fid *fid)
3631 {
3632         struct seq_server_site  *ss = osd_seq_site(osd);
3633         ENTRY;
3634
3635         /* FID seqs not in FLDB, must be local seq */
3636         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3637                 RETURN(0);
3638
3639         /* If FLD is not being initialized yet, it only happens during the
3640          * initialization, likely during mgs initialization, and we assume
3641          * this is local FID. */
3642         if (ss == NULL || ss->ss_server_fld == NULL)
3643                 RETURN(0);
3644
3645         /* Only check the local FLDB here */
3646         if (osd_seq_exists(env, osd, fid_seq(fid)))
3647                 RETURN(0);
3648
3649         RETURN(1);
3650 }
3651
3652 /**
3653  * Index delete function for interoperability mode (b11826).
3654  * It will remove the directory entry added by osd_index_ea_insert().
3655  * This entry is needed to maintain name->fid mapping.
3656  *
3657  * \param key,  key i.e. file entry to be deleted
3658  *
3659  * \retval   0, on success
3660  * \retval -ve, on error
3661  */
3662 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3663                                const struct dt_key *key, struct thandle *handle,
3664                                struct lustre_capa *capa)
3665 {
3666         struct osd_object          *obj = osd_dt_obj(dt);
3667         struct inode               *dir = obj->oo_inode;
3668         struct dentry              *dentry;
3669         struct osd_thandle         *oh;
3670         struct ldiskfs_dir_entry_2 *de = NULL;
3671         struct buffer_head         *bh;
3672         struct htree_lock          *hlock = NULL;
3673         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3674         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3675         int                        rc;
3676         ENTRY;
3677
3678         if (!dt_object_exists(dt))
3679                 RETURN(-ENOENT);
3680
3681         LINVRNT(osd_invariant(obj));
3682         LASSERT(!dt_object_remote(dt));
3683         LASSERT(handle != NULL);
3684
3685         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3686
3687         oh = container_of(handle, struct osd_thandle, ot_super);
3688         LASSERT(oh->ot_handle != NULL);
3689         LASSERT(oh->ot_handle->h_transaction != NULL);
3690
3691         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3692                 RETURN(-EACCES);
3693
3694         ll_vfs_dq_init(dir);
3695         dentry = osd_child_dentry_get(env, obj,
3696                                       (char *)key, strlen((char *)key));
3697
3698         if (obj->oo_hl_head != NULL) {
3699                 hlock = osd_oti_get(env)->oti_hlock;
3700                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3701                                    dir, LDISKFS_HLOCK_DEL);
3702         } else {
3703                 down_write(&obj->oo_ext_idx_sem);
3704         }
3705
3706         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3707         if (bh) {
3708                 __u32 ino = 0;
3709
3710                 /* If this is not the ".." entry, it might be a remote DNE
3711                  * entry and  we need to check if the FID is for a remote
3712                  * MDT.  If the FID is  not in the directory entry (e.g.
3713                  * upgraded 1.8 filesystem without dirdata enabled) then
3714                  * we need to get the FID from the LMA. For a remote directory
3715                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3716                  * case.
3717                  *
3718                  * Delete the entry before the agent inode in order to
3719                  * simplify error handling.  At worst an error after deleting
3720                  * the entry first might leak the agent inode afterward. The
3721                  * reverse would need filesystem abort in case of error deleting
3722                  * the entry after the agent had been removed, or leave a
3723                  * dangling entry pointing at a random inode. */
3724                 if (strcmp((char *)key, dotdot) != 0) {
3725                         LASSERT(de != NULL);
3726                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3727                         /* If Fid is not in dentry, try to get it from LMA */
3728                         if (rc == -ENODATA) {
3729                                 struct osd_inode_id *id;
3730                                 struct inode *inode;
3731
3732                                 /* Before trying to get fid from the inode,
3733                                  * check whether the inode is valid.
3734                                  *
3735                                  * If the inode has been deleted, do not go
3736                                  * ahead to do osd_ea_fid_get, which will set
3737                                  * the inode to bad inode, which might cause
3738                                  * the inode to be deleted uncorrectly */
3739                                 inode = ldiskfs_iget(osd_sb(osd),
3740                                                      le32_to_cpu(de->inode));
3741                                 if (IS_ERR(inode)) {
3742                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3743                                                "error.\n", osd_name(osd),
3744                                                PFID(fid));
3745                                         rc = PTR_ERR(inode);
3746                                 } else {
3747                                         if (likely(inode->i_nlink != 0)) {
3748                                                 id = &osd_oti_get(env)->oti_id;
3749                                                 rc = osd_ea_fid_get(env, obj,
3750                                                         le32_to_cpu(de->inode),
3751                                                                     fid, id);
3752                                         } else {
3753                                                 CDEBUG(D_INFO, "%s: %u "DFID
3754                                                        "deleted.\n",
3755                                                        osd_name(osd),
3756                                                        le32_to_cpu(de->inode),
3757                                                        PFID(fid));
3758                                                 rc = -ESTALE;
3759                                         }
3760                                         iput(inode);
3761                                 }
3762                         }
3763                         if (rc == 0 &&
3764                             unlikely(osd_remote_fid(env, osd, fid)))
3765                                 /* Need to delete agent inode */
3766                                 ino = le32_to_cpu(de->inode);
3767                 }
3768                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3769                 brelse(bh);
3770                 if (rc == 0 && unlikely(ino != 0)) {
3771                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3772                                                           oh);
3773                         if (rc != 0)
3774                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3775                                        osd_name(osd), PFID(fid), rc);
3776                 }
3777         } else {
3778                 rc = -ENOENT;
3779         }
3780         if (hlock != NULL)
3781                 ldiskfs_htree_unlock(hlock);
3782         else
3783                 up_write(&obj->oo_ext_idx_sem);
3784
3785         if (rc != 0)
3786                 GOTO(out, rc);
3787
3788         /* For inode on the remote MDT, .. will point to
3789          * /Agent directory, Check whether it needs to delete
3790          * from agent directory */
3791         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3792                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3793                                                    oh);
3794                 if (rc != 0 && rc != -ENOENT) {
3795                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3796                                osd_name(osd), PFID(fid), rc);
3797                 }
3798
3799                 if (rc == -ENOENT)
3800                         rc = 0;
3801
3802                 GOTO(out, rc);
3803         }
3804 out:
3805
3806         LASSERT(osd_invariant(obj));
3807         RETURN(rc);
3808 }
3809
3810 /**
3811  *      Lookup index for \a key and copy record to \a rec.
3812  *
3813  *      \param  dt      osd index object
3814  *      \param  key     key for index
3815  *      \param  rec     record reference
3816  *
3817  *      \retval  +ve  success : exact mach
3818  *      \retval  0    return record with key not greater than \a key
3819  *      \retval -ve   failure
3820  */
3821 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3822                                 struct dt_rec *rec, const struct dt_key *key,
3823                                 struct lustre_capa *capa)
3824 {
3825         struct osd_object      *obj = osd_dt_obj(dt);
3826         struct iam_path_descr  *ipd;
3827         struct iam_container   *bag = &obj->oo_dir->od_container;
3828         struct osd_thread_info *oti = osd_oti_get(env);
3829         struct iam_iterator    *it = &oti->oti_idx_it;
3830         struct iam_rec         *iam_rec;
3831         int                     rc;
3832         ENTRY;
3833
3834         if (!dt_object_exists(dt))
3835                 RETURN(-ENOENT);
3836
3837         LASSERT(osd_invariant(obj));
3838         LASSERT(!dt_object_remote(dt));
3839         LASSERT(bag->ic_object == obj->oo_inode);
3840
3841         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3842                 RETURN(-EACCES);
3843
3844         ipd = osd_idx_ipd_get(env, bag);
3845         if (IS_ERR(ipd))
3846                 RETURN(-ENOMEM);
3847
3848         /* got ipd now we can start iterator. */
3849         iam_it_init(it, bag, 0, ipd);
3850
3851         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3852                 /* swab quota uid/gid provided by caller */
3853                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3854                 key = (const struct dt_key *)&oti->oti_quota_id;
3855         }
3856
3857         rc = iam_it_get(it, (struct iam_key *)key);
3858         if (rc >= 0) {
3859                 if (S_ISDIR(obj->oo_inode->i_mode))
3860                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3861                 else
3862                         iam_rec = (struct iam_rec *) rec;
3863
3864                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3865
3866                 if (S_ISDIR(obj->oo_inode->i_mode))
3867                         osd_fid_unpack((struct lu_fid *) rec,
3868                                        (struct osd_fid_pack *)iam_rec);
3869                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3870                         osd_quota_unpack(obj, rec);
3871         }
3872
3873         iam_it_put(it);
3874         iam_it_fini(it);
3875         osd_ipd_put(env, bag, ipd);
3876
3877         LINVRNT(osd_invariant(obj));
3878
3879         RETURN(rc);
3880 }
3881
3882 static int osd_index_declare_iam_insert(const struct lu_env *env,
3883                                         struct dt_object *dt,
3884                                         const struct dt_rec *rec,
3885                                         const struct dt_key *key,
3886                                         struct thandle *handle)
3887 {
3888         struct osd_thandle *oh;
3889
3890         LASSERT(handle != NULL);
3891
3892         oh = container_of0(handle, struct osd_thandle, ot_super);
3893         LASSERT(oh->ot_handle == NULL);
3894
3895         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3896                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3897
3898         return 0;
3899 }
3900
3901 /**
3902  *      Inserts (key, value) pair in \a dt index object.
3903  *
3904  *      \param  dt      osd index object
3905  *      \param  key     key for index
3906  *      \param  rec     record reference
3907  *      \param  th      transaction handler
3908  *
3909  *      \retval  0  success
3910  *      \retval -ve failure
3911  */
3912 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3913                                 const struct dt_rec *rec,
3914                                 const struct dt_key *key, struct thandle *th,
3915                                 struct lustre_capa *capa, int ignore_quota)
3916 {
3917         struct osd_object     *obj = osd_dt_obj(dt);
3918         struct iam_path_descr *ipd;
3919         struct osd_thandle    *oh;
3920         struct iam_container  *bag;
3921         struct osd_thread_info *oti = osd_oti_get(env);
3922         struct iam_rec         *iam_rec;
3923         int                     rc;
3924         ENTRY;
3925
3926         if (!dt_object_exists(dt))
3927                 RETURN(-ENOENT);
3928
3929         LINVRNT(osd_invariant(obj));
3930         LASSERT(!dt_object_remote(dt));
3931
3932         bag = &obj->oo_dir->od_container;
3933         LASSERT(bag->ic_object == obj->oo_inode);
3934         LASSERT(th != NULL);
3935
3936         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3937                 RETURN(-EACCES);
3938
3939         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3940
3941         ipd = osd_idx_ipd_get(env, bag);
3942         if (unlikely(ipd == NULL))
3943                 RETURN(-ENOMEM);
3944
3945         oh = container_of0(th, struct osd_thandle, ot_super);
3946         LASSERT(oh->ot_handle != NULL);
3947         LASSERT(oh->ot_handle->h_transaction != NULL);
3948         if (S_ISDIR(obj->oo_inode->i_mode)) {
3949                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3950                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3951         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3952                 /* pack quota uid/gid */
3953                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3954                 key = (const struct dt_key *)&oti->oti_quota_id;
3955                 /* pack quota record */
3956                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3957                 iam_rec = (struct iam_rec *)rec;
3958         } else {
3959                 iam_rec = (struct iam_rec *)rec;
3960         }
3961
3962         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3963                         iam_rec, ipd);
3964         osd_ipd_put(env, bag, ipd);
3965         LINVRNT(osd_invariant(obj));
3966         RETURN(rc);
3967 }
3968
3969 /**
3970  * Calls ldiskfs_add_entry() to add directory entry
3971  * into the directory. This is required for
3972  * interoperability mode (b11826)
3973  *
3974  * \retval   0, on success
3975  * \retval -ve, on error
3976  */
3977 static int __osd_ea_add_rec(struct osd_thread_info *info,
3978                             struct osd_object *pobj, struct inode  *cinode,
3979                             const char *name, const struct lu_fid *fid,
3980                             struct htree_lock *hlock, struct thandle *th)
3981 {
3982         struct ldiskfs_dentry_param *ldp;
3983         struct dentry               *child;
3984         struct osd_thandle          *oth;
3985         int                          rc;
3986
3987         oth = container_of(th, struct osd_thandle, ot_super);
3988         LASSERT(oth->ot_handle != NULL);
3989         LASSERT(oth->ot_handle->h_transaction != NULL);
3990         LASSERT(pobj->oo_inode);
3991
3992         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3993         if (unlikely(pobj->oo_inode ==
3994                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3995                 ldp->edp_magic = 0;
3996         else
3997                 osd_get_ldiskfs_dirent_param(ldp, fid);
3998         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3999         child->d_fsdata = (void *)ldp;
4000         ll_vfs_dq_init(pobj->oo_inode);
4001         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
4002         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4003                 struct ldiskfs_dir_entry_2      *de;
4004                 struct buffer_head              *bh;
4005                 int                              rc1;
4006
4007                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4008                                             NULL, hlock);
4009                 if (bh != NULL) {
4010                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4011                                                                bh);
4012                         if (rc1 == 0) {
4013                                 if (S_ISDIR(cinode->i_mode))
4014                                         de->file_type = LDISKFS_DIRENT_LUFID |
4015                                                         LDISKFS_FT_REG_FILE;
4016                                 else
4017                                         de->file_type = LDISKFS_DIRENT_LUFID |
4018                                                         LDISKFS_FT_DIR;
4019                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4020                                                               NULL, bh);
4021                                 brelse(bh);
4022                         }
4023                 }
4024         }
4025
4026         RETURN(rc);
4027 }
4028
4029 /**
4030  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4031  * into the directory.Also sets flags into osd object to
4032  * indicate dot and dotdot are created. This is required for
4033  * interoperability mode (b11826)
4034  *
4035  * \param dir   directory for dot and dotdot fixup.
4036  * \param obj   child object for linking
4037  *
4038  * \retval   0, on success
4039  * \retval -ve, on error
4040  */
4041 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4042                               struct osd_object *dir,
4043                               struct inode *parent_dir, const char *name,
4044                               const struct lu_fid *dot_fid,
4045                               const struct lu_fid *dot_dot_fid,
4046                               struct thandle *th)
4047 {
4048         struct inode                *inode = dir->oo_inode;
4049         struct osd_thandle          *oth;
4050         int result = 0;
4051
4052         oth = container_of(th, struct osd_thandle, ot_super);
4053         LASSERT(oth->ot_handle->h_transaction != NULL);
4054         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4055
4056         if (strcmp(name, dot) == 0) {
4057                 if (dir->oo_compat_dot_created) {
4058                         result = -EEXIST;
4059                 } else {
4060                         LASSERT(inode == parent_dir);
4061                         dir->oo_compat_dot_created = 1;
4062                         result = 0;
4063                 }
4064         } else if (strcmp(name, dotdot) == 0) {
4065                 if (!dir->oo_compat_dot_created)
4066                         return -EINVAL;
4067                 /* in case of rename, dotdot is already created */
4068                 if (dir->oo_compat_dotdot_created) {
4069                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4070                                                 dot_dot_fid, NULL, th);
4071                 }
4072
4073                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4074                         struct lu_fid tfid = *dot_dot_fid;
4075
4076                         tfid.f_oid--;
4077                         result = osd_add_dot_dotdot_internal(info,
4078                                         dir->oo_inode, parent_dir, dot_fid,
4079                                         &tfid, oth);
4080                 } else {
4081                         result = osd_add_dot_dotdot_internal(info,
4082                                         dir->oo_inode, parent_dir, dot_fid,
4083                                         dot_dot_fid, oth);
4084                 }
4085
4086                 if (result == 0)
4087                         dir->oo_compat_dotdot_created = 1;
4088         }
4089
4090         return result;
4091 }
4092
4093
4094 /**
4095  * It will call the appropriate osd_add* function and return the
4096  * value, return by respective functions.
4097  */
4098 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4099                           struct inode *cinode, const char *name,
4100                           const struct lu_fid *fid, struct thandle *th)
4101 {
4102         struct osd_thread_info *info   = osd_oti_get(env);
4103         struct htree_lock      *hlock;
4104         int                     rc;
4105
4106         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4107
4108         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4109                                                    name[2] =='\0'))) {
4110                 if (hlock != NULL) {
4111                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4112                                            pobj->oo_inode, 0);
4113                 } else {
4114                         down_write(&pobj->oo_ext_idx_sem);
4115                 }
4116
4117                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4118                                         lu_object_fid(&pobj->oo_dt.do_lu),
4119                                         fid, th);
4120         } else {
4121                 if (hlock != NULL) {
4122                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4123                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4124                 } else {
4125                         down_write(&pobj->oo_ext_idx_sem);
4126                 }
4127
4128                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4129                         struct lu_fid *tfid = &info->oti_fid;
4130
4131                         *tfid = *fid;
4132                         tfid->f_ver = ~0;
4133                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4134                                               tfid, hlock, th);
4135                 } else {
4136                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4137                                               hlock, th);
4138                 }
4139         }
4140         if (hlock != NULL)
4141                 ldiskfs_htree_unlock(hlock);
4142         else
4143                 up_write(&pobj->oo_ext_idx_sem);
4144
4145         return rc;
4146 }
4147
4148 static void
4149 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4150                       struct osd_idmap_cache *oic)
4151 {
4152         struct osd_scrub    *scrub = &dev->od_scrub;
4153         struct lu_fid       *fid   = &oic->oic_fid;
4154         struct osd_inode_id *id    = &oti->oti_id;
4155         int                  once  = 0;
4156         int                  rc;
4157         ENTRY;
4158
4159         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4160                 RETURN_EXIT;
4161
4162         if (scrub->os_pos_current > id->oii_ino)
4163                 RETURN_EXIT;
4164
4165 again:
4166         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4167         if (rc != 0 && rc != -ENOENT)
4168                 RETURN_EXIT;
4169
4170         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4171                 RETURN_EXIT;
4172
4173         if (thread_is_running(&scrub->os_thread)) {
4174                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4175                 /* There is race condition between osd_oi_lookup and OI scrub.
4176                  * The OI scrub finished just after osd_oi_lookup() failure.
4177                  * Under such case, it is unnecessary to trigger OI scrub again,
4178                  * but try to call osd_oi_lookup() again. */
4179                 if (unlikely(rc == -EAGAIN))
4180                         goto again;
4181
4182                 RETURN_EXIT;
4183         }
4184
4185         if (!dev->od_noscrub && ++once == 1) {
4186                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4187                                      SS_CLEAR_FAILOUT);
4188                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4189                               ", rc = %d [2]\n",
4190                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4191                               PFID(fid), rc);
4192                 if (rc == 0 || rc == -EALREADY)
4193                         goto again;
4194         }
4195
4196         EXIT;
4197 }
4198
4199 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4200                                struct osd_device *dev,
4201                                struct osd_idmap_cache *oic,
4202                                struct lu_fid *fid, __u32 ino)
4203 {
4204         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4205         struct inode            *inode;
4206         int                      rc;
4207
4208         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4209         inode = osd_iget(oti, dev, &oic->oic_lid);
4210         if (IS_ERR(inode)) {
4211                 fid_zero(&oic->oic_fid);
4212                 return PTR_ERR(inode);
4213         }
4214
4215         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4216         iput(inode);
4217         if (rc != 0)
4218                 fid_zero(&oic->oic_fid);
4219         else
4220                 *fid = oic->oic_fid = lma->lma_self_fid;
4221         return rc;
4222 }
4223
4224 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4225                       struct osd_inode_id *id, const struct lu_fid *fid)
4226 {
4227         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4228                id->oii_ino, id->oii_gen, info);
4229         info->oti_cache.oic_lid = *id;
4230         info->oti_cache.oic_fid = *fid;
4231         info->oti_cache.oic_dev = osd;
4232 }
4233
4234 /**
4235  * Get parent FID from the linkEA.
4236  *
4237  * For a directory which parent resides on remote MDT, to satisfy the
4238  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4239  * the other hand, to make the lookup(..) on the directory can return
4240  * the real parent FID, we append the real parent FID after its ".."
4241  * name entry in the /REMOTE_PARENT_DIR.
4242  *
4243  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4244  * backup. So after the restore, we cannot get the right parent FID from
4245  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4246  * we have stored the real parent FID in the directory object's linkEA,
4247  * we can parse the linkEA for the real parent FID.
4248  *
4249  * \param[in] env       pointer to the thread context
4250  * \param[in] obj       pointer to the object to be handled
4251  * \param[out]fid       pointer to the buffer to hold the parent FID
4252  *
4253  * \retval              0 for getting the real parent FID successfully
4254  * \retval              negative error number on failure
4255  */
4256 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4257                                     struct osd_object *obj,
4258                                     struct lu_fid *fid)
4259 {
4260         struct osd_thread_info  *oti    = osd_oti_get(env);
4261         struct lu_buf           *buf    = &oti->oti_big_buf;
4262         struct dentry           *dentry = &oti->oti_obj_dentry;
4263         struct inode            *inode  = obj->oo_inode;
4264         struct linkea_data       ldata  = { NULL };
4265         int                      rc;
4266         ENTRY;
4267
4268         fid_zero(fid);
4269         if (!S_ISDIR(inode->i_mode))
4270                 RETURN(-EIO);
4271
4272 again:
4273         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4274                              buf->lb_buf, buf->lb_len);
4275         if (rc == -ERANGE) {
4276                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4277                                      NULL, 0);
4278                 if (rc > 0) {
4279                         lu_buf_realloc(buf, rc);
4280                         if (buf->lb_buf == NULL)
4281                                 RETURN(-ENOMEM);
4282
4283                         goto again;
4284                 }
4285         }
4286
4287         if (unlikely(rc == 0))
4288                 RETURN(-ENODATA);
4289
4290         if (rc < 0)
4291                 RETURN(rc);
4292
4293         if (unlikely(buf->lb_buf == NULL)) {
4294                 lu_buf_realloc(buf, rc);
4295                 if (buf->lb_buf == NULL)
4296                         RETURN(-ENOMEM);
4297
4298                 goto again;
4299         }
4300
4301         ldata.ld_buf = buf;
4302         rc = linkea_init(&ldata);
4303         if (rc == 0) {
4304                 linkea_first_entry(&ldata);
4305                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4306         }
4307
4308         RETURN(rc);
4309 }
4310
4311 /**
4312  * Calls ->lookup() to find dentry. From dentry get inode and
4313  * read inode's ea to get fid. This is required for  interoperability
4314  * mode (b11826)
4315  *
4316  * \retval   0, on success
4317  * \retval -ve, on error
4318  */
4319 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4320                              struct dt_rec *rec, const struct dt_key *key)
4321 {
4322         struct inode                    *dir    = obj->oo_inode;
4323         struct dentry                   *dentry;
4324         struct ldiskfs_dir_entry_2      *de;
4325         struct buffer_head              *bh;
4326         struct lu_fid                   *fid = (struct lu_fid *) rec;
4327         struct htree_lock               *hlock = NULL;
4328         int                             ino;
4329         int                             rc;
4330         ENTRY;
4331
4332         LASSERT(dir->i_op != NULL);
4333         LASSERT(dir->i_op->lookup != NULL);
4334
4335         dentry = osd_child_dentry_get(env, obj,
4336                                       (char *)key, strlen((char *)key));
4337
4338         if (obj->oo_hl_head != NULL) {
4339                 hlock = osd_oti_get(env)->oti_hlock;
4340                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4341                                    dir, LDISKFS_HLOCK_LOOKUP);
4342         } else {
4343                 down_read(&obj->oo_ext_idx_sem);
4344         }
4345
4346         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4347         if (bh) {
4348                 struct osd_thread_info *oti = osd_oti_get(env);
4349                 struct osd_inode_id *id = &oti->oti_id;
4350                 struct osd_idmap_cache *oic = &oti->oti_cache;
4351                 struct osd_device *dev = osd_obj2dev(obj);
4352
4353                 ino = le32_to_cpu(de->inode);
4354                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4355                         brelse(bh);
4356                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4357                         GOTO(out, rc);
4358                 }
4359
4360                 rc = osd_get_fid_from_dentry(de, rec);
4361
4362                 /* done with de, release bh */
4363                 brelse(bh);
4364                 if (rc != 0) {
4365                         if (unlikely(ino == osd_remote_parent_ino(dev)))
4366                                 /* If the parent is on remote MDT, and there
4367                                  * is no FID-in-dirent, then we have to get
4368                                  * the parent FID from the linkEA.  */
4369                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
4370                         else
4371                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4372                 } else {
4373                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4374                 }
4375
4376                 if (rc != 0) {
4377                         fid_zero(&oic->oic_fid);
4378                         GOTO(out, rc);
4379                 }
4380
4381                 if (osd_remote_fid(env, dev, fid))
4382                         GOTO(out, rc = 0);
4383
4384                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4385                 osd_consistency_check(oti, dev, oic);
4386         } else {
4387                 rc = -ENOENT;
4388         }
4389
4390         GOTO(out, rc);
4391
4392 out:
4393         if (hlock != NULL)
4394                 ldiskfs_htree_unlock(hlock);
4395         else
4396                 up_read(&obj->oo_ext_idx_sem);
4397         return rc;
4398 }
4399
4400 /**
4401  * Find the osd object for given fid.
4402  *
4403  * \param fid need to find the osd object having this fid
4404  *
4405  * \retval osd_object on success
4406  * \retval        -ve on error
4407  */
4408 static struct osd_object *osd_object_find(const struct lu_env *env,
4409                                           struct dt_object *dt,
4410                                           const struct lu_fid *fid)
4411 {
4412         struct lu_device  *ludev = dt->do_lu.lo_dev;
4413         struct osd_object *child = NULL;
4414         struct lu_object  *luch;
4415         struct lu_object  *lo;
4416
4417         /*
4418          * at this point topdev might not exist yet
4419          * (i.e. MGS is preparing profiles). so we can
4420          * not rely on topdev and instead lookup with
4421          * our device passed as topdev. this can't work
4422          * if the object isn't cached yet (as osd doesn't
4423          * allocate lu_header). IOW, the object must be
4424          * in the cache, otherwise lu_object_alloc() crashes
4425          * -bzzz
4426          */
4427         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4428                                  ludev : ludev->ld_site->ls_top_dev,
4429                                  fid, NULL);
4430         if (!IS_ERR(luch)) {
4431                 if (lu_object_exists(luch)) {
4432                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4433                         if (lo != NULL)
4434                                 child = osd_obj(lo);
4435                         else
4436                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4437                                                 "lu_object can't be located"
4438                                                 DFID"\n", PFID(fid));
4439
4440                         if (child == NULL) {
4441                                 lu_object_put(env, luch);
4442                                 CERROR("Unable to get osd_object\n");
4443                                 child = ERR_PTR(-ENOENT);
4444                         }
4445                 } else {
4446                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4447                                         "lu_object does not exists "DFID"\n",
4448                                         PFID(fid));
4449                         lu_object_put(env, luch);
4450                         child = ERR_PTR(-ENOENT);
4451                 }
4452         } else {
4453                 child = ERR_CAST(luch);
4454         }
4455
4456         return child;
4457 }
4458
4459 /**
4460  * Put the osd object once done with it.
4461  *
4462  * \param obj osd object that needs to be put
4463  */
4464 static inline void osd_object_put(const struct lu_env *env,
4465                                   struct osd_object *obj)
4466 {
4467         lu_object_put(env, &obj->oo_dt.do_lu);
4468 }
4469
4470 static int osd_index_declare_ea_insert(const struct lu_env *env,
4471                                        struct dt_object *dt,
4472                                        const struct dt_rec *rec,
4473                                        const struct dt_key *key,
4474                                        struct thandle *handle)
4475 {
4476         struct osd_thandle      *oh;
4477         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4478         struct lu_fid           *fid = (struct lu_fid *)rec;
4479         int                     rc;
4480         ENTRY;
4481
4482         LASSERT(!dt_object_remote(dt));
4483         LASSERT(handle != NULL);
4484
4485         oh = container_of0(handle, struct osd_thandle, ot_super);
4486         LASSERT(oh->ot_handle == NULL);
4487
4488         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4489                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4490
4491         if (osd_dt_obj(dt)->oo_inode != NULL) {
4492                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4493
4494                 /* We ignore block quota on meta pool (MDTs), so needn't
4495                  * calculate how many blocks will be consumed by this index
4496                  * insert */
4497                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4498                                            i_gid_read(inode), 0, oh,
4499                                            osd_dt_obj(dt), true, NULL, false);
4500         }
4501
4502         if (fid == NULL)
4503                 RETURN(0);
4504
4505         rc = osd_remote_fid(env, osd, fid);
4506         if (rc <= 0)
4507                 RETURN(rc);
4508
4509         rc = 0;
4510
4511         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4512                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4513         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4514                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4515         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4516                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4517
4518         RETURN(rc);
4519 }
4520
4521 /**
4522  * Index add function for interoperability mode (b11826).
4523  * It will add the directory entry.This entry is needed to
4524  * maintain name->fid mapping.
4525  *
4526  * \param key it is key i.e. file entry to be inserted
4527  * \param rec it is value of given key i.e. fid
4528  *
4529  * \retval   0, on success
4530  * \retval -ve, on error
4531  */
4532 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4533                                const struct dt_rec *rec,
4534                                const struct dt_key *key, struct thandle *th,
4535                                struct lustre_capa *capa, int ignore_quota)
4536 {
4537         struct osd_object       *obj = osd_dt_obj(dt);
4538         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4539         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4540         const struct lu_fid     *fid    = rec1->rec_fid;
4541         const char              *name = (const char *)key;
4542         struct osd_thread_info  *oti   = osd_oti_get(env);
4543         struct osd_inode_id     *id    = &oti->oti_id;
4544         struct inode            *child_inode = NULL;
4545         struct osd_object       *child = NULL;
4546         int                     rc;
4547         ENTRY;
4548
4549         if (!dt_object_exists(dt))
4550                 RETURN(-ENOENT);
4551
4552         LASSERT(osd_invariant(obj));
4553         LASSERT(!dt_object_remote(dt));
4554         LASSERT(th != NULL);
4555
4556         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4557
4558         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4559                 RETURN(-EACCES);
4560
4561         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4562
4563         rc = osd_remote_fid(env, osd, fid);
4564         if (rc < 0) {
4565                 CERROR("%s: Can not find object "DFID" rc %d\n",
4566                        osd_name(osd), PFID(fid), rc);
4567                 RETURN(rc);
4568         }
4569
4570         if (rc == 1) {
4571                 /* Insert remote entry */
4572                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4573                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4574                         struct osd_thandle      *oh;
4575
4576                         /* If parent on remote MDT, we need put this object
4577                          * under AGENT */
4578                         oh = container_of(th, typeof(*oh), ot_super);
4579                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4580                         if (rc != 0) {
4581                                 CERROR("%s: add "DFID" error: rc = %d\n",
4582                                        osd_name(osd),
4583                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4584                                 RETURN(rc);
4585                         }
4586
4587                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4588                 } else {
4589                         child_inode = osd_create_local_agent_inode(env, osd,
4590                                         obj, fid, rec1->rec_type & S_IFMT, th);
4591                         if (IS_ERR(child_inode))
4592                                 RETURN(PTR_ERR(child_inode));
4593                 }
4594         } else {
4595                 /* Insert local entry */
4596                 child = osd_object_find(env, dt, fid);
4597                 if (IS_ERR(child)) {
4598                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4599                                osd_name(osd), PFID(fid),
4600                                id->oii_ino, id->oii_gen,
4601                                (int)PTR_ERR(child));
4602                         RETURN(PTR_ERR(child));
4603                 }
4604                 child_inode = igrab(child->oo_inode);
4605         }
4606
4607         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4608
4609         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4610                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4611
4612         iput(child_inode);
4613         if (child != NULL)
4614                 osd_object_put(env, child);
4615         LASSERT(osd_invariant(obj));
4616         RETURN(rc);
4617 }
4618
4619 /**
4620  *  Initialize osd Iterator for given osd index object.
4621  *
4622  *  \param  dt      osd index object
4623  */
4624
4625 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4626                                      struct dt_object *dt,
4627                                      __u32 unused,
4628                                      struct lustre_capa *capa)
4629 {
4630         struct osd_it_iam      *it;
4631         struct osd_thread_info *oti = osd_oti_get(env);
4632         struct osd_object      *obj = osd_dt_obj(dt);
4633         struct lu_object       *lo  = &dt->do_lu;
4634         struct iam_path_descr  *ipd;
4635         struct iam_container   *bag = &obj->oo_dir->od_container;
4636
4637         if (!dt_object_exists(dt))
4638                 return ERR_PTR(-ENOENT);
4639
4640         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4641                 return ERR_PTR(-EACCES);
4642
4643         if (oti->oti_it_inline) {
4644                 OBD_ALLOC_PTR(it);
4645                 if (it == NULL)
4646                         return ERR_PTR(-ENOMEM);
4647         } else {
4648                 it = &oti->oti_it;
4649                 oti->oti_it_inline = 1;
4650         }
4651
4652         ipd = osd_it_ipd_get(env, bag);
4653         if (likely(ipd != NULL)) {
4654                 it->oi_obj = obj;
4655                 it->oi_ipd = ipd;
4656                 lu_object_get(lo);
4657                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4658                 return (struct dt_it *)it;
4659         } else {
4660                 if (it != &oti->oti_it)
4661                         OBD_FREE_PTR(it);
4662                 else
4663                         oti->oti_it_inline = 0;
4664
4665                 return ERR_PTR(-ENOMEM);
4666         }
4667 }
4668
4669 /**
4670  * free given Iterator.
4671  */
4672
4673 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4674 {
4675         struct osd_thread_info  *oti = osd_oti_get(env);
4676         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4677         struct osd_object       *obj = it->oi_obj;
4678
4679         iam_it_fini(&it->oi_it);
4680         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4681         lu_object_put(env, &obj->oo_dt.do_lu);
4682         if (it != &oti->oti_it)
4683                 OBD_FREE_PTR(it);
4684         else
4685                 oti->oti_it_inline = 0;
4686 }
4687
4688 /**
4689  *  Move Iterator to record specified by \a key
4690  *
4691  *  \param  di      osd iterator
4692  *  \param  key     key for index
4693  *
4694  *  \retval +ve  di points to record with least key not larger than key
4695  *  \retval  0   di points to exact matched key
4696  *  \retval -ve  failure
4697  */
4698
4699 static int osd_it_iam_get(const struct lu_env *env,
4700                           struct dt_it *di, const struct dt_key *key)
4701 {
4702         struct osd_thread_info  *oti = osd_oti_get(env);
4703         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4704
4705         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4706                 /* swab quota uid/gid */
4707                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4708                 key = (struct dt_key *)&oti->oti_quota_id;
4709         }
4710
4711         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4712 }
4713
4714 /**
4715  *  Release Iterator
4716  *
4717  *  \param  di      osd iterator
4718  */
4719 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4720 {
4721         struct osd_it_iam *it = (struct osd_it_iam *)di;
4722
4723         iam_it_put(&it->oi_it);
4724 }
4725
4726 /**
4727  *  Move iterator by one record
4728  *
4729  *  \param  di      osd iterator
4730  *
4731  *  \retval +1   end of container reached
4732  *  \retval  0   success
4733  *  \retval -ve  failure
4734  */
4735
4736 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4737 {
4738         struct osd_it_iam *it = (struct osd_it_iam *)di;
4739
4740         return iam_it_next(&it->oi_it);
4741 }
4742
4743 /**
4744  * Return pointer to the key under iterator.
4745  */
4746
4747 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4748                                  const struct dt_it *di)
4749 {
4750         struct osd_thread_info *oti = osd_oti_get(env);
4751         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4752         struct osd_object      *obj = it->oi_obj;
4753         struct dt_key          *key;
4754
4755         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4756
4757         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4758                 /* swab quota uid/gid */
4759                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4760                 key = (struct dt_key *)&oti->oti_quota_id;
4761         }
4762
4763         return key;
4764 }
4765
4766 /**
4767  * Return size of key under iterator (in bytes)
4768  */
4769
4770 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4771 {
4772         struct osd_it_iam *it = (struct osd_it_iam *)di;
4773
4774         return iam_it_key_size(&it->oi_it);
4775 }
4776
4777 static inline void
4778 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4779 {
4780         /* check if file type is required */
4781         if (ent->lde_attrs & LUDA_TYPE) {
4782                 struct luda_type *lt;
4783                 int align = sizeof(*lt) - 1;
4784
4785                 len = (len + align) & ~align;
4786                 lt = (struct luda_type *)(ent->lde_name + len);
4787                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4788         }
4789
4790         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4791 }
4792
4793 /**
4794  * build lu direct from backend fs dirent.
4795  */
4796
4797 static inline void
4798 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4799                    char *name, __u16 namelen, __u16 type, __u32 attr)
4800 {
4801         ent->lde_attrs = attr | LUDA_FID;
4802         fid_cpu_to_le(&ent->lde_fid, fid);
4803
4804         ent->lde_hash = cpu_to_le64(offset);
4805         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4806
4807         strncpy(ent->lde_name, name, namelen);
4808         ent->lde_name[namelen] = '\0';
4809         ent->lde_namelen = cpu_to_le16(namelen);
4810
4811         /* append lustre attributes */
4812         osd_it_append_attrs(ent, namelen, type);
4813 }
4814
4815 /**
4816  * Return pointer to the record under iterator.
4817  */
4818 static int osd_it_iam_rec(const struct lu_env *env,
4819                           const struct dt_it *di,
4820                           struct dt_rec *dtrec, __u32 attr)
4821 {
4822         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4823         struct osd_thread_info *info = osd_oti_get(env);
4824         ENTRY;
4825
4826         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4827                 const struct osd_fid_pack *rec;
4828                 struct lu_fid             *fid = &info->oti_fid;
4829                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4830                 char                      *name;
4831                 int                        namelen;
4832                 __u64                      hash;
4833                 int                        rc;
4834
4835                 name = (char *)iam_it_key_get(&it->oi_it);
4836                 if (IS_ERR(name))
4837                         RETURN(PTR_ERR(name));
4838
4839                 namelen = iam_it_key_size(&it->oi_it);
4840
4841                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4842                 if (IS_ERR(rec))
4843                         RETURN(PTR_ERR(rec));
4844
4845                 rc = osd_fid_unpack(fid, rec);
4846                 if (rc)
4847                         RETURN(rc);
4848
4849                 hash = iam_it_store(&it->oi_it);
4850
4851                 /* IAM does not store object type in IAM index (dir) */
4852                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4853                                    0, LUDA_FID);
4854         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4855                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4856                            (struct iam_rec *)dtrec);
4857                 osd_quota_unpack(it->oi_obj, dtrec);
4858         } else {
4859                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4860                            (struct iam_rec *)dtrec);
4861         }
4862
4863         RETURN(0);
4864 }
4865
4866 /**
4867  * Returns cookie for current Iterator position.
4868  */
4869 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4870 {
4871         struct osd_it_iam *it = (struct osd_it_iam *)di;
4872
4873         return iam_it_store(&it->oi_it);
4874 }
4875
4876 /**
4877  * Restore iterator from cookie.
4878  *
4879  * \param  di      osd iterator
4880  * \param  hash    Iterator location cookie
4881  *
4882  * \retval +ve  di points to record with least key not larger than key.
4883  * \retval  0   di points to exact matched key
4884  * \retval -ve  failure
4885  */
4886
4887 static int osd_it_iam_load(const struct lu_env *env,
4888                            const struct dt_it *di, __u64 hash)
4889 {
4890         struct osd_it_iam *it = (struct osd_it_iam *)di;
4891
4892         return iam_it_load(&it->oi_it, hash);
4893 }
4894
4895 static const struct dt_index_operations osd_index_iam_ops = {
4896         .dio_lookup         = osd_index_iam_lookup,
4897         .dio_declare_insert = osd_index_declare_iam_insert,
4898         .dio_insert         = osd_index_iam_insert,
4899         .dio_declare_delete = osd_index_declare_iam_delete,
4900         .dio_delete         = osd_index_iam_delete,
4901         .dio_it     = {
4902                 .init     = osd_it_iam_init,
4903                 .fini     = osd_it_iam_fini,
4904                 .get      = osd_it_iam_get,
4905                 .put      = osd_it_iam_put,
4906                 .next     = osd_it_iam_next,
4907                 .key      = osd_it_iam_key,
4908                 .key_size = osd_it_iam_key_size,
4909                 .rec      = osd_it_iam_rec,
4910                 .store    = osd_it_iam_store,
4911                 .load     = osd_it_iam_load
4912         }
4913 };
4914
4915
4916 /**
4917  * Creates or initializes iterator context.
4918  *
4919  * \retval struct osd_it_ea, iterator structure on success
4920  *
4921  */
4922 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4923                                     struct dt_object *dt,
4924                                     __u32 attr,
4925                                     struct lustre_capa *capa)
4926 {
4927         struct osd_object       *obj  = osd_dt_obj(dt);
4928         struct osd_thread_info  *info = osd_oti_get(env);
4929         struct osd_it_ea        *it;
4930         struct file             *file;
4931         struct lu_object        *lo   = &dt->do_lu;
4932         struct dentry           *obj_dentry = &info->oti_it_dentry;
4933         ENTRY;
4934
4935         if (!dt_object_exists(dt))
4936                 RETURN(ERR_PTR(-ENOENT));
4937
4938         if (info->oti_it_inline) {
4939                 OBD_ALLOC_PTR(it);
4940                 if (it == NULL)
4941                         RETURN(ERR_PTR(-ENOMEM));
4942         } else {
4943                 it = &info->oti_it_ea;
4944                 info->oti_it_inline = 1;
4945         }
4946
4947         obj_dentry->d_inode = obj->oo_inode;
4948         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4949         obj_dentry->d_name.hash = 0;
4950
4951         it->oie_rd_dirent       = 0;
4952         it->oie_it_dirent       = 0;
4953         it->oie_dirent          = NULL;
4954         it->oie_buf             = info->oti_it_ea_buf;
4955         it->oie_obj             = obj;
4956
4957         file = &it->oie_file;
4958         /* Reset the "file" totally to avoid to reuse any old value from
4959          * former readdir handling, the "file->f_pos" should be zero. */
4960         memset(file, 0, sizeof(*file));
4961         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4962         if (attr & LUDA_64BITHASH)
4963                 file->f_mode    = FMODE_64BITHASH;
4964         else
4965                 file->f_mode    = FMODE_32BITHASH;
4966         file->f_dentry          = obj_dentry;
4967         file->f_mapping         = obj->oo_inode->i_mapping;
4968         file->f_op              = obj->oo_inode->i_fop;
4969         set_file_inode(file, obj->oo_inode);
4970
4971         lu_object_get(lo);
4972         RETURN((struct dt_it *) it);
4973 }
4974
4975 /**
4976  * Destroy or finishes iterator context.
4977  *
4978  * \param di iterator structure to be destroyed
4979  */
4980 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4981 {
4982         struct osd_thread_info  *info   = osd_oti_get(env);
4983         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4984         struct osd_object       *obj    = it->oie_obj;
4985         struct inode            *inode  = obj->oo_inode;
4986
4987         ENTRY;
4988         it->oie_file.f_op->release(inode, &it->oie_file);
4989         lu_object_put(env, &obj->oo_dt.do_lu);
4990         if (it != &info->oti_it_ea)
4991                 OBD_FREE_PTR(it);
4992         else
4993                 info->oti_it_inline = 0;
4994         EXIT;
4995 }
4996
4997 /**
4998  * It position the iterator at given key, so that next lookup continues from
4999  * that key Or it is similar to dio_it->load() but based on a key,
5000  * rather than file position.
5001  *
5002  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5003  * to the beginning.
5004  *
5005  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5006  */
5007 static int osd_it_ea_get(const struct lu_env *env,
5008                          struct dt_it *di, const struct dt_key *key)
5009 {
5010         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
5011
5012         ENTRY;
5013         LASSERT(((const char *)key)[0] == '\0');
5014         it->oie_file.f_pos      = 0;
5015         it->oie_rd_dirent       = 0;
5016         it->oie_it_dirent       = 0;
5017         it->oie_dirent          = NULL;
5018
5019         RETURN(+1);
5020 }
5021
5022 /**
5023  * Does nothing
5024  */
5025 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5026 {
5027 }
5028
5029 struct osd_filldir_cbs {
5030 #ifdef HAVE_DIR_CONTEXT
5031         struct dir_context ctx;
5032 #endif
5033         struct osd_it_ea  *it;
5034 };
5035 /**
5036  * It is called internally by ->readdir(). It fills the
5037  * iterator's in-memory data structure with required
5038  * information i.e. name, namelen, rec_size etc.
5039  *
5040  * \param buf in which information to be filled in.
5041  * \param name name of the file in given dir
5042  *
5043  * \retval 0 on success
5044  * \retval 1 on buffer full
5045  */
5046 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5047                                loff_t offset, __u64 ino,
5048                                unsigned d_type)
5049 {
5050         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5051         struct osd_object       *obj  = it->oie_obj;
5052         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5053         struct lu_fid           *fid  = &ent->oied_fid;
5054         struct osd_fid_pack     *rec;
5055         ENTRY;
5056
5057         /* this should never happen */
5058         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5059                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5060                 RETURN(-EIO);
5061         }
5062
5063         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5064             OSD_IT_EA_BUFSIZE)
5065                 RETURN(1);
5066
5067         /* "." is just the object itself. */
5068         if (namelen == 1 && name[0] == '.') {
5069                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5070         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5071                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5072                 if (osd_fid_unpack(fid, rec) != 0)
5073                         fid_zero(fid);
5074         } else {
5075                 fid_zero(fid);
5076         }
5077         d_type &= ~LDISKFS_DIRENT_LUFID;
5078
5079         /* NOT export local root. */
5080         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5081                 ino = obj->oo_inode->i_ino;
5082                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5083         }
5084
5085         ent->oied_ino     = ino;
5086         ent->oied_off     = offset;
5087         ent->oied_namelen = namelen;
5088         ent->oied_type    = d_type;
5089
5090         memcpy(ent->oied_name, name, namelen);
5091
5092         it->oie_rd_dirent++;
5093         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5094         RETURN(0);
5095 }
5096
5097 /**
5098  * Calls ->readdir() to load a directory entry at a time
5099  * and stored it in iterator's in-memory data structure.
5100  *
5101  * \param di iterator's in memory structure
5102  *
5103  * \retval   0 on success
5104  * \retval -ve on error
5105  * \retval +1 reach the end of entry
5106  */
5107 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5108                                const struct dt_it *di)
5109 {
5110         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5111         struct osd_object  *obj   = it->oie_obj;
5112         struct inode       *inode = obj->oo_inode;
5113         struct htree_lock  *hlock = NULL;
5114         struct file        *filp  = &it->oie_file;
5115         int                 rc = 0;
5116         struct osd_filldir_cbs buf = {
5117 #ifdef HAVE_DIR_CONTEXT
5118                 .ctx.actor = osd_ldiskfs_filldir,
5119 #endif
5120                 .it = it
5121         };
5122
5123         ENTRY;
5124         it->oie_dirent = it->oie_buf;
5125         it->oie_rd_dirent = 0;
5126
5127         if (obj->oo_hl_head != NULL) {
5128                 hlock = osd_oti_get(env)->oti_hlock;
5129                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5130                                    inode, LDISKFS_HLOCK_READDIR);
5131         } else {
5132                 down_read(&obj->oo_ext_idx_sem);
5133         }
5134
5135 #ifdef HAVE_DIR_CONTEXT
5136         buf.ctx.pos = filp->f_pos;
5137         rc = inode->i_fop->iterate(filp, &buf.ctx);
5138         filp->f_pos = buf.ctx.pos;
5139 #else
5140         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5141 #endif
5142
5143         if (hlock != NULL)
5144                 ldiskfs_htree_unlock(hlock);
5145         else
5146                 up_read(&obj->oo_ext_idx_sem);
5147
5148         if (it->oie_rd_dirent == 0) {
5149                 /*If it does not get any dirent, it means it has been reached
5150                  *to the end of the dir */
5151                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5152                 if (rc == 0)
5153                         rc = 1;
5154         } else {
5155                 it->oie_dirent = it->oie_buf;
5156                 it->oie_it_dirent = 1;
5157         }
5158
5159         RETURN(rc);
5160 }
5161
5162 /**
5163  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5164  * to load a directory entry at a time and stored it in
5165  * iterator's in-memory data structure.
5166  *
5167  * \param di iterator's in memory structure
5168  *
5169  * \retval +ve iterator reached to end
5170  * \retval   0 iterator not reached to end
5171  * \retval -ve on error
5172  */
5173 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5174 {
5175         struct osd_it_ea *it = (struct osd_it_ea *)di;
5176         int rc;
5177
5178         ENTRY;
5179
5180         if (it->oie_it_dirent < it->oie_rd_dirent) {
5181                 it->oie_dirent =
5182                         (void *) it->oie_dirent +
5183                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5184                                        it->oie_dirent->oied_namelen);
5185                 it->oie_it_dirent++;
5186                 RETURN(0);
5187         } else {
5188                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5189                         rc = +1;
5190                 else
5191                         rc = osd_ldiskfs_it_fill(env, di);
5192         }
5193
5194         RETURN(rc);
5195 }
5196
5197 /**
5198  * Returns the key at current position from iterator's in memory structure.
5199  *
5200  * \param di iterator's in memory structure
5201  *
5202  * \retval key i.e. struct dt_key on success
5203  */
5204 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5205                                     const struct dt_it *di)
5206 {
5207         struct osd_it_ea *it = (struct osd_it_ea *)di;
5208
5209         return (struct dt_key *)it->oie_dirent->oied_name;
5210 }
5211
5212 /**
5213  * Returns the key's size at current position from iterator's in memory structure.
5214  *
5215  * \param di iterator's in memory structure
5216  *
5217  * \retval key_size i.e. struct dt_key on success
5218  */
5219 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5220 {
5221         struct osd_it_ea *it = (struct osd_it_ea *)di;
5222
5223         return it->oie_dirent->oied_namelen;
5224 }
5225
5226 static int
5227 osd_dirent_update(handle_t *jh, struct super_block *sb,
5228                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5229                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5230 {
5231         struct osd_fid_pack *rec;
5232         int                  rc;
5233         ENTRY;
5234
5235         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5236         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5237
5238         rc = ldiskfs_journal_get_write_access(jh, bh);
5239         if (rc != 0)
5240                 RETURN(rc);
5241
5242         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5243         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5244         rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5245
5246         RETURN(rc);
5247 }
5248
5249 static inline int
5250 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5251 {
5252         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5253             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5254                 return 1;
5255         else
5256                 return 0;
5257 }
5258
5259 static inline int
5260 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5261 {
5262         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5263                  "dot_dotdot = %d\n", dot_dotdot);
5264
5265         if (LDISKFS_DIR_REC_LEN(de) >=
5266             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5267                 return 1;
5268         else
5269                 return 0;
5270 }
5271
5272 static int
5273 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5274                     struct inode *dir, struct inode *inode,
5275                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5276                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5277                     struct htree_lock *hlock)
5278 {
5279         struct dentry               *dentry;
5280         struct osd_fid_pack         *rec;
5281         struct ldiskfs_dentry_param *ldp;
5282         int                          rc;
5283         ENTRY;
5284
5285         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5286                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5287                 RETURN(0);
5288
5289         /* There is enough space to hold the FID-in-dirent. */
5290         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5291                                  dir->i_sb->s_blocksize)) {
5292                 rc = ldiskfs_journal_get_write_access(jh, bh);
5293                 if (rc != 0)
5294                         RETURN(rc);
5295
5296                 de->name[de->name_len] = 0;
5297                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5298                 rec->fp_len = sizeof(struct lu_fid) + 1;
5299                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5300                 de->file_type |= LDISKFS_DIRENT_LUFID;
5301
5302                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5303
5304                 RETURN(rc);
5305         }
5306
5307         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5308         if (rc != 0)
5309                 RETURN(rc);
5310
5311         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5312                                            ent->oied_namelen);
5313         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5314         osd_get_ldiskfs_dirent_param(ldp, fid);
5315         dentry->d_fsdata = (void *)ldp;
5316         ll_vfs_dq_init(dir);
5317         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5318         /* It is too bad, we cannot reinsert the name entry back.
5319          * That means we lose it! */
5320         if (rc != 0)
5321                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5322                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5323                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5324                        dir->i_ino, dir->i_generation,
5325                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5326
5327         RETURN(rc);
5328 }
5329
5330 static int
5331 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5332                         struct osd_it_ea *it, struct lu_fid *fid,
5333                         struct osd_inode_id *id, __u32 *attr)
5334 {
5335         struct osd_thread_info     *info        = osd_oti_get(env);
5336         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5337         struct osd_device          *dev         = osd_obj2dev(obj);
5338         struct super_block         *sb          = osd_sb(dev);
5339         const char                 *devname     =
5340                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5341         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5342         struct inode               *dir         = obj->oo_inode;
5343         struct htree_lock          *hlock       = NULL;
5344         struct buffer_head         *bh          = NULL;
5345         handle_t                   *jh          = NULL;
5346         struct ldiskfs_dir_entry_2 *de;
5347         struct dentry              *dentry;
5348         struct inode               *inode;
5349         int                         credits;
5350         int                         rc;
5351         int                         dot_dotdot  = 0;
5352         bool                        dirty       = false;
5353         ENTRY;
5354
5355         if (ent->oied_name[0] == '.') {
5356                 if (ent->oied_namelen == 1)
5357                         dot_dotdot = 1;
5358                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5359                         dot_dotdot = 2;
5360         }
5361
5362         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5363                                       ent->oied_namelen);
5364
5365         /* We need to ensure that the name entry is still valid.
5366          * Because it may be removed or renamed by other already.
5367          *
5368          * The unlink or rename operation will start journal before PDO lock,
5369          * so to avoid deadlock, here we need to start journal handle before
5370          * related PDO lock also. But because we do not know whether there
5371          * will be something to be repaired before PDO lock, we just start
5372          * journal without conditions.
5373          *
5374          * We may need to remove the name entry firstly, then insert back.
5375          * One credit is for user quota file update.
5376          * One credit is for group quota file update.
5377          * Two credits are for dirty inode. */
5378         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5379                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5380
5381 again:
5382         if (dev->od_dirent_journal) {
5383                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5384                 if (IS_ERR(jh)) {
5385                         rc = PTR_ERR(jh);
5386                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5387                                "check_repair, dir = %lu/%u, credits = %d, "
5388                                "name = %.*s: rc = %d\n",
5389                                devname, dir->i_ino, dir->i_generation, credits,
5390                                ent->oied_namelen, ent->oied_name, rc);
5391                         RETURN(rc);
5392                 }
5393
5394                 if (obj->oo_hl_head != NULL) {
5395                         hlock = osd_oti_get(env)->oti_hlock;
5396                         /* "0" means exclusive lock for the whole directory.
5397                          * We need to prevent others access such name entry
5398                          * during the delete + insert. Neither HLOCK_ADD nor
5399                          * HLOCK_DEL cannot guarantee the atomicity. */
5400                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5401                 } else {
5402                         down_write(&obj->oo_ext_idx_sem);
5403                 }
5404         } else {
5405                 if (obj->oo_hl_head != NULL) {
5406                         hlock = osd_oti_get(env)->oti_hlock;
5407                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5408                                            LDISKFS_HLOCK_LOOKUP);
5409                 } else {
5410                         down_read(&obj->oo_ext_idx_sem);
5411                 }
5412         }
5413
5414         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5415         /* For dot/dotdot entry, if there is not enough space to hold the
5416          * FID-in-dirent, just keep them there. It only happens when the
5417          * device upgraded from 1.8 or restored from MDT file-level backup.
5418          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5419          * and needs to get FID from LMA when readdir, it will not affect the
5420          * performance much. */
5421         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5422             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5423                 *attr |= LUDA_IGNORE;
5424                 GOTO(out_journal, rc = 0);
5425         }
5426
5427         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5428         inode = osd_iget(info, dev, id);
5429         if (IS_ERR(inode)) {
5430                 rc = PTR_ERR(inode);
5431                 if (rc == -ENOENT || rc == -ESTALE)
5432                         rc = 1;
5433                 else
5434                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5435                                "check_repair, dir = %lu/%u, name = %.*s: "
5436                                "rc = %d\n",
5437                                devname, dir->i_ino, dir->i_generation,
5438                                ent->oied_namelen, ent->oied_name, rc);
5439
5440                 GOTO(out_journal, rc);
5441         }
5442
5443         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5444         if (rc == 0) {
5445                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5446
5447                 if (fid_is_sane(fid)) {
5448                         /* FID-in-dirent is valid. */
5449                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5450                                 GOTO(out_inode, rc = 0);
5451
5452                         /* Do not repair under dryrun mode. */
5453                         if (*attr & LUDA_VERIFY_DRYRUN) {
5454                                 *attr |= LUDA_REPAIR;
5455                                 GOTO(out_inode, rc = 0);
5456                         }
5457
5458                         if (!dev->od_dirent_journal) {
5459                                 iput(inode);
5460                                 brelse(bh);
5461                                 if (hlock != NULL)
5462                                         ldiskfs_htree_unlock(hlock);
5463                                 else
5464                                         up_read(&obj->oo_ext_idx_sem);
5465                                 dev->od_dirent_journal = 1;
5466                                 goto again;
5467                         }
5468
5469                         *fid = lma->lma_self_fid;
5470                         dirty = true;
5471                         /* Update the FID-in-dirent. */
5472                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5473                         if (rc == 0)
5474                                 *attr |= LUDA_REPAIR;
5475                         else
5476                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5477                                        "in the dirent, dir = %lu/%u, "
5478                                        "name = %.*s, "DFID": rc = %d\n",
5479                                        devname, dir->i_ino, dir->i_generation,
5480                                        ent->oied_namelen, ent->oied_name,
5481                                        PFID(fid), rc);
5482                 } else {
5483                         /* Do not repair under dryrun mode. */
5484                         if (*attr & LUDA_VERIFY_DRYRUN) {
5485                                 *fid = lma->lma_self_fid;
5486                                 *attr |= LUDA_REPAIR;
5487                                 GOTO(out_inode, rc = 0);
5488                         }
5489
5490                         if (!dev->od_dirent_journal) {
5491                                 iput(inode);
5492                                 brelse(bh);
5493                                 if (hlock != NULL)
5494                                         ldiskfs_htree_unlock(hlock);
5495                                 else
5496                                         up_read(&obj->oo_ext_idx_sem);
5497                                 dev->od_dirent_journal = 1;
5498                                 goto again;
5499                         }
5500
5501                         *fid = lma->lma_self_fid;
5502                         dirty = true;
5503                         /* Append the FID-in-dirent. */
5504                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5505                                                  fid, bh, de, hlock);
5506                         if (rc == 0)
5507                                 *attr |= LUDA_REPAIR;
5508                         else
5509                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5510                                        "after the dirent, dir = %lu/%u, "
5511                                        "name = %.*s, "DFID": rc = %d\n",
5512                                        devname, dir->i_ino, dir->i_generation,
5513                                        ent->oied_namelen, ent->oied_name,
5514                                        PFID(fid), rc);
5515                 }
5516         } else if (rc == -ENODATA) {
5517                 /* Do not repair under dryrun mode. */
5518                 if (*attr & LUDA_VERIFY_DRYRUN) {
5519                         if (fid_is_sane(fid)) {
5520                                 *attr |= LUDA_REPAIR;
5521                         } else {
5522                                 lu_igif_build(fid, inode->i_ino,
5523                                               inode->i_generation);
5524                                 *attr |= LUDA_UPGRADE;
5525                         }
5526                         GOTO(out_inode, rc = 0);
5527                 }
5528
5529                 if (!dev->od_dirent_journal) {
5530                         iput(inode);
5531                         brelse(bh);
5532                         if (hlock != NULL)
5533                                 ldiskfs_htree_unlock(hlock);
5534                         else
5535                                 up_read(&obj->oo_ext_idx_sem);
5536                         dev->od_dirent_journal = 1;
5537                         goto again;
5538                 }
5539
5540                 dirty = true;
5541                 if (unlikely(fid_is_sane(fid))) {
5542                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5543                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5544                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5545                         if (rc == 0)
5546                                 *attr |= LUDA_REPAIR;
5547                         else
5548                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5549                                        "update dirent, dir = %lu/%u, "
5550                                        "name = %.*s, "DFID": rc = %d\n",
5551                                        devname, dir->i_ino, dir->i_generation,
5552                                        ent->oied_namelen, ent->oied_name,
5553                                        PFID(fid), rc);
5554                 } else {
5555                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5556                         /* It is probably IGIF object. Only aappend the
5557                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5558                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5559                                                  fid, bh, de, hlock);
5560                         if (rc == 0)
5561                                 *attr |= LUDA_UPGRADE;
5562                         else
5563                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5564                                        "after the dirent, dir = %lu/%u, "
5565                                        "name = %.*s, "DFID": rc = %d\n",
5566                                        devname, dir->i_ino, dir->i_generation,
5567                                        ent->oied_namelen, ent->oied_name,
5568                                        PFID(fid), rc);
5569                 }
5570         }
5571
5572         GOTO(out_inode, rc);
5573
5574 out_inode:
5575         iput(inode);
5576
5577 out_journal:
5578         brelse(bh);
5579         if (hlock != NULL) {
5580                 ldiskfs_htree_unlock(hlock);
5581         } else {
5582                 if (dev->od_dirent_journal)
5583                         up_write(&obj->oo_ext_idx_sem);
5584                 else
5585                         up_read(&obj->oo_ext_idx_sem);
5586         }
5587         if (jh != NULL)
5588                 ldiskfs_journal_stop(jh);
5589         if (rc >= 0 && !dirty)
5590                 dev->od_dirent_journal = 0;
5591         return rc;
5592 }
5593
5594 /**
5595  * Returns the value at current position from iterator's in memory structure.
5596  *
5597  * \param di struct osd_it_ea, iterator's in memory structure
5598  * \param attr attr requested for dirent.
5599  * \param lde lustre dirent
5600  *
5601  * \retval   0 no error and \param lde has correct lustre dirent.
5602  * \retval -ve on error
5603  */
5604 static inline int osd_it_ea_rec(const struct lu_env *env,
5605                                 const struct dt_it *di,
5606                                 struct dt_rec *dtrec, __u32 attr)
5607 {
5608         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5609         struct osd_object      *obj   = it->oie_obj;
5610         struct osd_device      *dev   = osd_obj2dev(obj);
5611         struct osd_thread_info *oti   = osd_oti_get(env);
5612         struct osd_inode_id    *id    = &oti->oti_id;
5613         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5614         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5615         __u32                   ino   = it->oie_dirent->oied_ino;
5616         int                     rc    = 0;
5617         ENTRY;
5618
5619         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5620
5621         if (attr & LUDA_VERIFY) {
5622                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5623                         attr |= LUDA_IGNORE;
5624                         /* If the parent is on remote MDT, and there
5625                          * is no FID-in-dirent, then we have to get
5626                          * the parent FID from the linkEA.  */
5627                         osd_get_pfid_from_linkea(env, obj, fid);
5628                 } else {
5629                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5630                                                      &attr);
5631                 }
5632         } else {
5633                 attr &= ~LU_DIRENT_ATTRS_MASK;
5634                 if (!fid_is_sane(fid)) {
5635                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5636                             likely(it->oie_dirent->oied_namelen != 2 ||
5637                                    it->oie_dirent->oied_name[0] != '.' ||
5638                                    it->oie_dirent->oied_name[1] != '.'))
5639                                 RETURN(-ENOENT);
5640
5641                         if (unlikely(ino == osd_remote_parent_ino(dev)))
5642                                 /* If the parent is on remote MDT, and there
5643                                  * is no FID-in-dirent, then we have to get
5644                                  * the parent FID from the linkEA.  */
5645                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5646                         else
5647                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5648                 } else {
5649                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5650                 }
5651         }
5652
5653         /* Pack the entry anyway, at least the offset is right. */
5654         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5655                            it->oie_dirent->oied_name,
5656                            it->oie_dirent->oied_namelen,
5657                            it->oie_dirent->oied_type, attr);
5658
5659         if (rc < 0)
5660                 RETURN(rc);
5661
5662         if (osd_remote_fid(env, dev, fid))
5663                 RETURN(0);
5664
5665         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5666                 osd_add_oi_cache(oti, dev, id, fid);
5667
5668         RETURN(rc > 0 ? 0 : rc);
5669 }
5670
5671 /**
5672  * Returns the record size size at current position.
5673  *
5674  * This function will return record(lu_dirent) size in bytes.
5675  *
5676  * \param[in] env       execution environment
5677  * \param[in] di        iterator's in memory structure
5678  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5679  *                      calculate the lu_dirent size.
5680  *
5681  * \retval      record size(in bytes & in memory) of the current lu_dirent
5682  *              entry.
5683  */
5684 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5685                               __u32 attr)
5686 {
5687         struct osd_it_ea *it = (struct osd_it_ea *)di;
5688
5689         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5690 }
5691
5692 /**
5693  * Returns a cookie for current position of the iterator head, so that
5694  * user can use this cookie to load/start the iterator next time.
5695  *
5696  * \param di iterator's in memory structure
5697  *
5698  * \retval cookie for current position, on success
5699  */
5700 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5701 {
5702         struct osd_it_ea *it = (struct osd_it_ea *)di;
5703
5704         return it->oie_dirent->oied_off;
5705 }
5706
5707 /**
5708  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5709  * to load a directory entry at a time and stored it i inn,
5710  * in iterator's in-memory data structure.
5711  *
5712  * \param di struct osd_it_ea, iterator's in memory structure
5713  *
5714  * \retval +ve on success
5715  * \retval -ve on error
5716  */
5717 static int osd_it_ea_load(const struct lu_env *env,
5718                           const struct dt_it *di, __u64 hash)
5719 {
5720         struct osd_it_ea *it = (struct osd_it_ea *)di;
5721         int rc;
5722
5723         ENTRY;
5724         it->oie_file.f_pos = hash;
5725
5726         rc =  osd_ldiskfs_it_fill(env, di);
5727         if (rc > 0)
5728                 rc = -ENODATA;
5729
5730         if (rc == 0)
5731                 rc = +1;
5732
5733         RETURN(rc);
5734 }
5735
5736 /**
5737  * Index lookup function for interoperability mode (b11826).
5738  *
5739  * \param key,  key i.e. file name to be searched
5740  *
5741  * \retval +ve, on success
5742  * \retval -ve, on error
5743  */
5744 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5745                                struct dt_rec *rec, const struct dt_key *key,
5746                                struct lustre_capa *capa)
5747 {
5748         struct osd_object *obj = osd_dt_obj(dt);
5749         int rc = 0;
5750
5751         ENTRY;
5752
5753         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5754         LINVRNT(osd_invariant(obj));
5755
5756         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5757                 return -EACCES;
5758
5759         rc = osd_ea_lookup_rec(env, obj, rec, key);
5760         if (rc == 0)
5761                 rc = +1;
5762         RETURN(rc);
5763 }
5764
5765 /**
5766  * Index and Iterator operations for interoperability
5767  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5768  */
5769 static const struct dt_index_operations osd_index_ea_ops = {
5770         .dio_lookup         = osd_index_ea_lookup,
5771         .dio_declare_insert = osd_index_declare_ea_insert,
5772         .dio_insert         = osd_index_ea_insert,
5773         .dio_declare_delete = osd_index_declare_ea_delete,
5774         .dio_delete         = osd_index_ea_delete,
5775         .dio_it     = {
5776                 .init     = osd_it_ea_init,
5777                 .fini     = osd_it_ea_fini,
5778                 .get      = osd_it_ea_get,
5779                 .put      = osd_it_ea_put,
5780                 .next     = osd_it_ea_next,
5781                 .key      = osd_it_ea_key,
5782                 .key_size = osd_it_ea_key_size,
5783                 .rec      = osd_it_ea_rec,
5784                 .rec_size = osd_it_ea_rec_size,
5785                 .store    = osd_it_ea_store,
5786                 .load     = osd_it_ea_load
5787         }
5788 };
5789
5790 static void *osd_key_init(const struct lu_context *ctx,
5791                           struct lu_context_key *key)
5792 {
5793         struct osd_thread_info *info;
5794
5795         OBD_ALLOC_PTR(info);
5796         if (info == NULL)
5797                 return ERR_PTR(-ENOMEM);
5798
5799         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5800         if (info->oti_it_ea_buf == NULL)
5801                 goto out_free_info;
5802
5803         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5804
5805         info->oti_hlock = ldiskfs_htree_lock_alloc();
5806         if (info->oti_hlock == NULL)
5807                 goto out_free_ea;
5808
5809         return info;
5810
5811  out_free_ea:
5812         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5813  out_free_info:
5814         OBD_FREE_PTR(info);
5815         return ERR_PTR(-ENOMEM);
5816 }
5817
5818 static void osd_key_fini(const struct lu_context *ctx,
5819                          struct lu_context_key *key, void* data)
5820 {
5821         struct osd_thread_info *info = data;
5822
5823         if (info->oti_hlock != NULL)
5824                 ldiskfs_htree_lock_free(info->oti_hlock);
5825         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5826         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5827         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5828         lu_buf_free(&info->oti_big_buf);
5829         OBD_FREE_PTR(info);
5830 }
5831
5832 static void osd_key_exit(const struct lu_context *ctx,
5833                          struct lu_context_key *key, void *data)
5834 {
5835         struct osd_thread_info *info = data;
5836
5837         LASSERT(info->oti_r_locks == 0);
5838         LASSERT(info->oti_w_locks == 0);
5839         LASSERT(info->oti_txns    == 0);
5840 }
5841
5842 /* type constructor/destructor: osd_type_init, osd_type_fini */
5843 LU_TYPE_INIT_FINI(osd, &osd_key);
5844
5845 struct lu_context_key osd_key = {
5846         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5847         .lct_init = osd_key_init,
5848         .lct_fini = osd_key_fini,
5849         .lct_exit = osd_key_exit
5850 };
5851
5852
5853 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5854                            const char *name, struct lu_device *next)
5855 {
5856         struct osd_device *osd = osd_dev(d);
5857
5858         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5859             >= sizeof(osd->od_svname))
5860                 return -E2BIG;
5861         return osd_procfs_init(osd, name);
5862 }
5863
5864 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5865 {
5866         struct seq_server_site  *ss = osd_seq_site(osd);
5867         int                     rc;
5868         ENTRY;
5869
5870         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5871                 RETURN(0);
5872
5873         if (unlikely(ss == NULL))
5874                 RETURN(-ENODEV);
5875
5876         OBD_ALLOC_PTR(osd->od_cl_seq);
5877         if (osd->od_cl_seq == NULL)
5878                 RETURN(-ENOMEM);
5879
5880         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5881                              osd->od_svname, ss->ss_server_seq);
5882
5883         if (rc != 0) {
5884                 OBD_FREE_PTR(osd->od_cl_seq);
5885                 osd->od_cl_seq = NULL;
5886         }
5887
5888         RETURN(rc);
5889 }
5890
5891 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5892 {
5893         if (osd->od_cl_seq == NULL)
5894                 return;
5895
5896         seq_client_fini(osd->od_cl_seq);
5897         OBD_FREE_PTR(osd->od_cl_seq);
5898         osd->od_cl_seq = NULL;
5899 }
5900
5901 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5902 {
5903         ENTRY;
5904
5905         /* shutdown quota slave instance associated with the device */
5906         if (o->od_quota_slave != NULL) {
5907                 qsd_fini(env, o->od_quota_slave);
5908                 o->od_quota_slave = NULL;
5909         }
5910
5911         osd_fid_fini(env, o);
5912
5913         RETURN(0);
5914 }
5915
5916 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5917 {
5918         ENTRY;
5919
5920         if (o->od_mnt != NULL) {
5921                 shrink_dcache_sb(osd_sb(o));
5922                 osd_sync(env, &o->od_dt_dev);
5923
5924                 mntput(o->od_mnt);
5925                 o->od_mnt = NULL;
5926         }
5927
5928         EXIT;
5929 }
5930
5931 static int osd_mount(const struct lu_env *env,
5932                      struct osd_device *o, struct lustre_cfg *cfg)
5933 {
5934         const char              *name  = lustre_cfg_string(cfg, 0);
5935         const char              *dev  = lustre_cfg_string(cfg, 1);
5936         const char              *opts;
5937         unsigned long            page, s_flags, lmd_flags = 0;
5938         struct page             *__page;
5939         struct file_system_type *type;
5940         char                    *options = NULL;
5941         char                    *str;
5942         struct osd_thread_info  *info = osd_oti_get(env);
5943         struct lu_fid           *fid = &info->oti_fid;
5944         struct inode            *inode;
5945         int                      rc = 0;
5946         ENTRY;
5947
5948         if (o->od_mnt != NULL)
5949                 RETURN(0);
5950
5951         if (strlen(dev) >= sizeof(o->od_mntdev))
5952                 RETURN(-E2BIG);
5953         strcpy(o->od_mntdev, dev);
5954
5955         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5956         if (__page == NULL)
5957                 GOTO(out, rc = -ENOMEM);
5958
5959         str = lustre_cfg_string(cfg, 2);
5960         s_flags = simple_strtoul(str, NULL, 0);
5961         str = strstr(str, ":");
5962         if (str)
5963                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5964         opts = lustre_cfg_string(cfg, 3);
5965         page = (unsigned long)page_address(__page);
5966         options = (char *)page;
5967         *options = '\0';
5968         if (opts == NULL)
5969                 strcat(options, "user_xattr,acl");
5970         else
5971                 strcat(options, opts);
5972
5973         /* Glom up mount options */
5974         if (*options != '\0')
5975                 strcat(options, ",");
5976         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5977
5978         type = get_fs_type("ldiskfs");
5979         if (!type) {
5980                 CERROR("%s: cannot find ldiskfs module\n", name);
5981                 GOTO(out, rc = -ENODEV);
5982         }
5983
5984         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5985         module_put(type->owner);
5986
5987         if (IS_ERR(o->od_mnt)) {
5988                 rc = PTR_ERR(o->od_mnt);
5989                 o->od_mnt = NULL;
5990                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5991                 GOTO(out, rc);
5992         }
5993
5994 #ifdef HAVE_DEV_SET_RDONLY
5995         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5996                 CERROR("%s: underlying device %s is marked as read-only. "
5997                        "Setup failed\n", name, dev);
5998                 GOTO(out_mnt, rc = -EROFS);
5999         }
6000 #endif
6001
6002         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6003                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6004                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6005                 GOTO(out_mnt, rc = -EINVAL);
6006         }
6007
6008 #ifdef LDISKFS_MOUNT_DIRDATA
6009         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6010                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6011                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6012         else if (!o->od_is_ost)
6013                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6014                       "enabling the dirdata feature. If you do not want to "
6015                       "downgrade to Lustre-1.x again, you can enable it via "
6016                       "'tune2fs -O dirdata device'\n", name, dev);
6017 #endif
6018         inode = osd_sb(o)->s_root->d_inode;
6019         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6020         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6021         if (rc != 0) {
6022                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6023                 GOTO(out_mnt, rc);
6024         }
6025
6026         if (lmd_flags & LMD_FLG_NOSCRUB)
6027                 o->od_noscrub = 1;
6028
6029         GOTO(out, rc = 0);
6030
6031 out_mnt:
6032         mntput(o->od_mnt);
6033         o->od_mnt = NULL;
6034
6035 out:
6036         if (__page)
6037                 OBD_PAGE_FREE(__page);
6038
6039         return rc;
6040 }
6041
6042 static struct lu_device *osd_device_fini(const struct lu_env *env,
6043                                          struct lu_device *d)
6044 {
6045         struct osd_device *o = osd_dev(d);
6046         ENTRY;
6047
6048         osd_shutdown(env, o);
6049         osd_procfs_fini(o);
6050         osd_scrub_cleanup(env, o);
6051         osd_obj_map_fini(o);
6052         osd_umount(env, o);
6053
6054         RETURN(NULL);
6055 }
6056
6057 static int osd_device_init0(const struct lu_env *env,
6058                             struct osd_device *o,
6059                             struct lustre_cfg *cfg)
6060 {
6061         struct lu_device        *l = osd2lu_dev(o);
6062         struct osd_thread_info *info;
6063         int                     rc;
6064         int                     cplen = 0;
6065
6066         /* if the module was re-loaded, env can loose its keys */
6067         rc = lu_env_refill((struct lu_env *) env);
6068         if (rc)
6069                 GOTO(out, rc);
6070         info = osd_oti_get(env);
6071         LASSERT(info);
6072
6073         l->ld_ops = &osd_lu_ops;
6074         o->od_dt_dev.dd_ops = &osd_dt_ops;
6075
6076         spin_lock_init(&o->od_osfs_lock);
6077         mutex_init(&o->od_otable_mutex);
6078
6079         o->od_capa_hash = init_capa_hash();
6080         if (o->od_capa_hash == NULL)
6081                 GOTO(out, rc = -ENOMEM);
6082
6083         o->od_read_cache = 1;
6084         o->od_writethrough_cache = 1;
6085         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6086
6087         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6088                         sizeof(o->od_svname));
6089         if (cplen >= sizeof(o->od_svname)) {
6090                 rc = -E2BIG;
6091                 GOTO(out_capa, rc);
6092         }
6093
6094         if (server_name_is_ost(o->od_svname))
6095                 o->od_is_ost = 1;
6096
6097         o->od_full_scrub_ratio = OFSR_DEFAULT;
6098         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6099         rc = osd_mount(env, o, cfg);
6100         if (rc != 0)
6101                 GOTO(out_capa, rc);
6102
6103         rc = osd_obj_map_init(env, o);
6104         if (rc != 0)
6105                 GOTO(out_mnt, rc);
6106
6107         rc = lu_site_init(&o->od_site, l);
6108         if (rc != 0)
6109                 GOTO(out_compat, rc);
6110         o->od_site.ls_bottom_dev = l;
6111
6112         rc = lu_site_init_finish(&o->od_site);
6113         if (rc != 0)
6114                 GOTO(out_site, rc);
6115
6116         /* self-repair LMA by default */
6117         o->od_lma_self_repair = 1;
6118
6119         INIT_LIST_HEAD(&o->od_ios_list);
6120         /* setup scrub, including OI files initialization */
6121         rc = osd_scrub_setup(env, o);
6122         if (rc < 0)
6123                 GOTO(out_site, rc);
6124
6125         rc = osd_procfs_init(o, o->od_svname);
6126         if (rc != 0) {
6127                 CERROR("%s: can't initialize procfs: rc = %d\n",
6128                        o->od_svname, rc);
6129                 GOTO(out_scrub, rc);
6130         }
6131
6132         LASSERT(l->ld_site->ls_linkage.next != NULL);
6133         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6134
6135         /* initialize quota slave instance */
6136         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6137                                      o->od_proc_entry);
6138         if (IS_ERR(o->od_quota_slave)) {
6139                 rc = PTR_ERR(o->od_quota_slave);
6140                 o->od_quota_slave = NULL;
6141                 GOTO(out_procfs, rc);
6142         }
6143
6144         RETURN(0);
6145
6146 out_procfs:
6147         osd_procfs_fini(o);
6148 out_scrub:
6149         osd_scrub_cleanup(env, o);
6150 out_site:
6151         lu_site_fini(&o->od_site);
6152 out_compat:
6153         osd_obj_map_fini(o);
6154 out_mnt:
6155         osd_umount(env, o);
6156 out_capa:
6157         cleanup_capa_hash(o->od_capa_hash);
6158 out:
6159         return rc;
6160 }
6161
6162 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6163                                           struct lu_device_type *t,
6164                                           struct lustre_cfg *cfg)
6165 {
6166         struct osd_device *o;
6167         int                rc;
6168
6169         OBD_ALLOC_PTR(o);
6170         if (o == NULL)
6171                 return ERR_PTR(-ENOMEM);
6172
6173         rc = dt_device_init(&o->od_dt_dev, t);
6174         if (rc == 0) {
6175                 /* Because the ctx might be revived in dt_device_init,
6176                  * refill the env here */
6177                 lu_env_refill((struct lu_env *)env);
6178                 rc = osd_device_init0(env, o, cfg);
6179                 if (rc)
6180                         dt_device_fini(&o->od_dt_dev);
6181         }
6182
6183         if (unlikely(rc != 0))
6184                 OBD_FREE_PTR(o);
6185
6186         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6187 }
6188
6189 static struct lu_device *osd_device_free(const struct lu_env *env,
6190                                          struct lu_device *d)
6191 {
6192         struct osd_device *o = osd_dev(d);
6193         ENTRY;
6194
6195         cleanup_capa_hash(o->od_capa_hash);
6196         /* XXX: make osd top device in order to release reference */
6197         d->ld_site->ls_top_dev = d;
6198         lu_site_purge(env, d->ld_site, -1);
6199         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6200                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6201                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6202         }
6203         lu_site_fini(&o->od_site);
6204         dt_device_fini(&o->od_dt_dev);
6205         OBD_FREE_PTR(o);
6206         RETURN(NULL);
6207 }
6208
6209 static int osd_process_config(const struct lu_env *env,
6210                               struct lu_device *d, struct lustre_cfg *cfg)
6211 {
6212         struct osd_device               *o = osd_dev(d);
6213         int                             rc;
6214         ENTRY;
6215
6216         switch (cfg->lcfg_command) {
6217         case LCFG_SETUP:
6218                 rc = osd_mount(env, o, cfg);
6219                 break;
6220         case LCFG_CLEANUP:
6221                 lu_dev_del_linkage(d->ld_site, d);
6222                 rc = osd_shutdown(env, o);
6223                 break;
6224         case LCFG_PARAM:
6225                 LASSERT(&o->od_dt_dev);
6226                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6227                                               cfg, &o->od_dt_dev);
6228                 if (rc > 0 || rc == -ENOSYS)
6229                         rc = class_process_proc_param(PARAM_OST,
6230                                                       lprocfs_osd_obd_vars,
6231                                                       cfg, &o->od_dt_dev);
6232                 break;
6233         default:
6234                 rc = -ENOSYS;
6235         }
6236
6237         RETURN(rc);
6238 }
6239
6240 static int osd_recovery_complete(const struct lu_env *env,
6241                                  struct lu_device *d)
6242 {
6243         struct osd_device       *osd = osd_dev(d);
6244         int                      rc = 0;
6245         ENTRY;
6246
6247         if (osd->od_quota_slave == NULL)
6248                 RETURN(0);
6249
6250         /* start qsd instance on recovery completion, this notifies the quota
6251          * slave code that we are about to process new requests now */
6252         rc = qsd_start(env, osd->od_quota_slave);
6253         RETURN(rc);
6254 }
6255
6256 /*
6257  * we use exports to track all osd users
6258  */
6259 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6260                            struct obd_device *obd, struct obd_uuid *cluuid,
6261                            struct obd_connect_data *data, void *localdata)
6262 {
6263         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6264         struct lustre_handle  conn;
6265         int                   rc;
6266         ENTRY;
6267
6268         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6269
6270         rc = class_connect(&conn, obd, cluuid);
6271         if (rc)
6272                 RETURN(rc);
6273
6274         *exp = class_conn2export(&conn);
6275
6276         spin_lock(&osd->od_osfs_lock);
6277         osd->od_connects++;
6278         spin_unlock(&osd->od_osfs_lock);
6279
6280         RETURN(0);
6281 }
6282
6283 /*
6284  * once last export (we don't count self-export) disappeared
6285  * osd can be released
6286  */
6287 static int osd_obd_disconnect(struct obd_export *exp)
6288 {
6289         struct obd_device *obd = exp->exp_obd;
6290         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6291         int                rc, release = 0;
6292         ENTRY;
6293
6294         /* Only disconnect the underlying layers on the final disconnect. */
6295         spin_lock(&osd->od_osfs_lock);
6296         osd->od_connects--;
6297         if (osd->od_connects == 0)
6298                 release = 1;
6299         spin_unlock(&osd->od_osfs_lock);
6300
6301         rc = class_disconnect(exp); /* bz 9811 */
6302
6303         if (rc == 0 && release)
6304                 class_manual_cleanup(obd);
6305         RETURN(rc);
6306 }
6307
6308 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6309                        struct lu_device *dev)
6310 {
6311         struct osd_device *osd = osd_dev(dev);
6312         int                result = 0;
6313         ENTRY;
6314
6315         if (osd->od_quota_slave != NULL) {
6316                 /* set up quota slave objects */
6317                 result = qsd_prepare(env, osd->od_quota_slave);
6318                 if (result != 0)
6319                         RETURN(result);
6320         }
6321
6322         result = osd_fid_init(env, osd);
6323
6324         RETURN(result);
6325 }
6326
6327 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6328                          struct lu_fid *fid, struct md_op_data *op_data)
6329 {
6330         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6331
6332         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6333 }
6334
6335 static const struct lu_object_operations osd_lu_obj_ops = {
6336         .loo_object_init      = osd_object_init,
6337         .loo_object_delete    = osd_object_delete,
6338         .loo_object_release   = osd_object_release,
6339         .loo_object_free      = osd_object_free,
6340         .loo_object_print     = osd_object_print,
6341         .loo_object_invariant = osd_object_invariant
6342 };
6343
6344 const struct lu_device_operations osd_lu_ops = {
6345         .ldo_object_alloc      = osd_object_alloc,
6346         .ldo_process_config    = osd_process_config,
6347         .ldo_recovery_complete = osd_recovery_complete,
6348         .ldo_prepare           = osd_prepare,
6349 };
6350
6351 static const struct lu_device_type_operations osd_device_type_ops = {
6352         .ldto_init = osd_type_init,
6353         .ldto_fini = osd_type_fini,
6354
6355         .ldto_start = osd_type_start,
6356         .ldto_stop  = osd_type_stop,
6357
6358         .ldto_device_alloc = osd_device_alloc,
6359         .ldto_device_free  = osd_device_free,
6360
6361         .ldto_device_init    = osd_device_init,
6362         .ldto_device_fini    = osd_device_fini
6363 };
6364
6365 static struct lu_device_type osd_device_type = {
6366         .ldt_tags     = LU_DEVICE_DT,
6367         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6368         .ldt_ops      = &osd_device_type_ops,
6369         .ldt_ctx_tags = LCT_LOCAL,
6370 };
6371
6372 /*
6373  * lprocfs legacy support.
6374  */
6375 static struct obd_ops osd_obd_device_ops = {
6376         .o_owner = THIS_MODULE,
6377         .o_connect      = osd_obd_connect,
6378         .o_disconnect   = osd_obd_disconnect,
6379         .o_fid_alloc    = osd_fid_alloc,
6380 };
6381
6382 static int __init osd_mod_init(void)
6383 {
6384         int rc;
6385
6386         osd_oi_mod_init();
6387
6388         rc = lu_kmem_init(ldiskfs_caches);
6389         if (rc)
6390                 return rc;
6391
6392         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6393                                  lprocfs_osd_module_vars,
6394                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6395         if (rc)
6396                 lu_kmem_fini(ldiskfs_caches);
6397         return rc;
6398 }
6399
6400 static void __exit osd_mod_exit(void)
6401 {
6402         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6403         lu_kmem_fini(ldiskfs_caches);
6404 }
6405
6406 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6407 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6408 MODULE_LICENSE("GPL");
6409
6410 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);