lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76 #include <lustre_linkea.h>
  77
  78 int ldiskfs_pdo = 1;
  79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  80                 "ldiskfs with parallel directory operations");
  81
  82 int ldiskfs_track_declares_assert;
  83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  84                 "LBUG during tracking of declares");
  85
  86 /* Slab to allocate dynlocks */
  87 struct kmem_cache *dynlock_cachep;
  88
  89 static struct lu_kmem_descr ldiskfs_caches[] = {
  90         {
  91                 .ckd_cache = &dynlock_cachep,
  92                 .ckd_name  = "dynlock_cache",
  93                 .ckd_size  = sizeof(struct dynlock_handle)
  94         },
  95         {
  96                 .ckd_cache = NULL
  97         }
  98 };
  99
 100 static const char dot[] = ".";
 101 static const char dotdot[] = "..";
 102 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 103
 104 static const struct lu_object_operations      osd_lu_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ops;
 106 static const struct dt_object_operations      osd_obj_ea_ops;
 107 static const struct dt_object_operations      osd_obj_otable_it_ops;
 108 static const struct dt_index_operations       osd_index_iam_ops;
 109 static const struct dt_index_operations       osd_index_ea_ops;
 110
 111 int osd_trans_declare_op2rb[] = {
 112         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 113         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 114         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 115         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 116         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 117         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 118         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 119         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 120         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 121         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 122         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 123         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 124 };
 125
 126 static int osd_has_index(const struct osd_object *obj)
 127 {
 128         return obj->oo_dt.do_index_ops != NULL;
 129 }
 130
 131 static int osd_object_invariant(const struct lu_object *l)
 132 {
 133         return osd_invariant(osd_obj(l));
 134 }
 135
 136 /*
 137  * Concurrency: doesn't matter
 138  */
 139
 140 /*
 141  * Concurrency: doesn't matter
 142  */
 143 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 144 {
 145         struct osd_thread_info *oti = osd_oti_get(env);
 146         return oti->oti_w_locks > 0 && o->oo_owner == env;
 147 }
 148
 149 /*
 150  * Concurrency: doesn't access mutable data
 151  */
 152 static int osd_root_get(const struct lu_env *env,
 153                         struct dt_device *dev, struct lu_fid *f)
 154 {
 155         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 156         return 0;
 157 }
 158
 159 /*
 160  * OSD object methods.
 161  */
 162
 163 /*
 164  * Concurrency: no concurrent access is possible that early in object
 165  * life-cycle.
 166  */
 167 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 168                                           const struct lu_object_header *hdr,
 169                                           struct lu_device *d)
 170 {
 171         struct osd_object *mo;
 172
 173         OBD_ALLOC_PTR(mo);
 174         if (mo != NULL) {
 175                 struct lu_object *l;
 176
 177                 l = &mo->oo_dt.do_lu;
 178                 dt_object_init(&mo->oo_dt, NULL, d);
 179                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 180                 l->lo_ops = &osd_lu_obj_ops;
 181                 init_rwsem(&mo->oo_sem);
 182                 init_rwsem(&mo->oo_ext_idx_sem);
 183                 spin_lock_init(&mo->oo_guard);
 184                 return l;
 185         } else {
 186                 return NULL;
 187         }
 188 }
 189
 190 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 191                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 192 {
 193         int rc;
 194
 195         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 196         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 197                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 198         if (rc > 0) {
 199                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 200                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 201                 rc = 0;
 202                 lustre_lma_swab(lma);
 203                 /* Check LMA compatibility */
 204                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 205                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 206                               "for fid = "DFID", ino = %lu\n",
 207                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 208                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 209                               PFID(&lma->lma_self_fid), inode->i_ino);
 210                         rc = -EOPNOTSUPP;
 211                 }
 212         } else if (rc == 0) {
 213                 rc = -ENODATA;
 214         }
 215
 216         return rc;
 217 }
 218
 219 /*
 220  * retrieve object from backend ext fs.
 221  **/
 222 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 223                        struct osd_inode_id *id)
 224 {
 225         struct inode *inode = NULL;
 226
 227         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 228         if (IS_ERR(inode)) {
 229                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 230                        id->oii_ino, PTR_ERR(inode));
 231         } else if (id->oii_gen != OSD_OII_NOGEN &&
 232                    inode->i_generation != id->oii_gen) {
 233                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 234                        "i_generation = %u\n",
 235                        id->oii_ino, id->oii_gen, inode->i_generation);
 236                 iput(inode);
 237                 inode = ERR_PTR(-ESTALE);
 238         } else if (inode->i_nlink == 0) {
 239                 /* due to parallel readdir and unlink,
 240                 * we can have dead inode here. */
 241                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 242                 iput(inode);
 243                 inode = ERR_PTR(-ESTALE);
 244         } else if (is_bad_inode(inode)) {
 245                 CWARN("%.16s: bad inode: ino = %u\n",
 246                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 247                 iput(inode);
 248                 inode = ERR_PTR(-ENOENT);
 249         } else {
 250                 if (id->oii_gen == OSD_OII_NOGEN)
 251                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 252
 253                 /* Do not update file c/mtime in ldiskfs.
 254                  * NB: we don't have any lock to protect this because we don't
 255                  * have reference on osd_object now, but contention with
 256                  * another lookup + attr_set can't happen in the tiny window
 257                  * between if (...) and set S_NOCMTIME. */
 258                 if (!(inode->i_flags & S_NOCMTIME))
 259                         inode->i_flags |= S_NOCMTIME;
 260         }
 261         return inode;
 262 }
 263
 264 static struct inode *
 265 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 266              struct osd_inode_id *id, struct lu_fid *fid)
 267 {
 268         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 269         struct inode            *inode;
 270         int                      rc;
 271
 272         inode = osd_iget(info, dev, id);
 273         if (IS_ERR(inode))
 274                 return inode;
 275
 276         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 277         if (rc == 0) {
 278                 *fid = lma->lma_self_fid;
 279         } else if (rc == -ENODATA) {
 280                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 281                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 282                 else
 283                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 284         } else {
 285                 iput(inode);
 286                 inode = ERR_PTR(rc);
 287         }
 288         return inode;
 289 }
 290
 291 static struct inode *osd_iget_check(struct osd_thread_info *info,
 292                                     struct osd_device *dev,
 293                                     const struct lu_fid *fid,
 294                                     struct osd_inode_id *id,
 295                                     bool in_oi)
 296 {
 297         struct inode    *inode;
 298         int              rc     = 0;
 299         ENTRY;
 300
 301         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 302         if (IS_ERR(inode)) {
 303                 rc = PTR_ERR(inode);
 304                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 305                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 306                                id->oii_ino, rc);
 307
 308                         GOTO(put, rc);
 309                 }
 310
 311                 goto check_oi;
 312         }
 313
 314         if (is_bad_inode(inode)) {
 315                 rc = -ENOENT;
 316                 if (!in_oi) {
 317                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 318
 319                         GOTO(put, rc);
 320                 }
 321
 322                 goto check_oi;
 323         }
 324
 325         if (id->oii_gen != OSD_OII_NOGEN &&
 326             inode->i_generation != id->oii_gen) {
 327                 rc = -ESTALE;
 328                 if (!in_oi) {
 329                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 330                                "oii_gen = %u, i_generation = %u\n",
 331                                id->oii_ino, id->oii_gen, inode->i_generation);
 332
 333                         GOTO(put, rc);
 334                 }
 335
 336                 goto check_oi;
 337         }
 338
 339         if (inode->i_nlink == 0) {
 340                 rc = -ENOENT;
 341                 if (!in_oi) {
 342                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 343
 344                         GOTO(put, rc);
 345                 }
 346
 347                 goto check_oi;
 348         }
 349
 350 check_oi:
 351         if (rc != 0) {
 352                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 353
 354                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 355                 /* XXX: There are some possible cases:
 356                  *      1. rc = 0.
 357                  *         Backup/restore caused the OI invalid.
 358                  *      2. rc = 0.
 359                  *         Someone unlinked the object but NOT removed
 360                  *         the OI mapping, such as mount target device
 361                  *         as ldiskfs, and modify something directly.
 362                  *      3. rc = -ENOENT.
 363                  *         Someone just removed the object between the
 364                  *         former oi_lookup and the iget. It is normal.
 365                  *      4. Other failure cases.
 366                  *
 367                  *      Generally, when the device is mounted, it will
 368                  *      auto check whether the system is restored from
 369                  *      file-level backup or not. We trust such detect
 370                  *      to distinguish the 1st case from the 2nd case. */
 371                 if (rc == 0) {
 372                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 373                             inode->i_generation == id->oii_gen)
 374                                 rc = -ENOENT;
 375                         else
 376                                 rc = -EREMCHG;
 377                 }
 378         } else {
 379                 if (id->oii_gen == OSD_OII_NOGEN)
 380                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 381
 382                 /* Do not update file c/mtime in ldiskfs.
 383                  * NB: we don't have any lock to protect this because we don't
 384                  * have reference on osd_object now, but contention with
 385                  * another lookup + attr_set can't happen in the tiny window
 386                  * between if (...) and set S_NOCMTIME. */
 387                 if (!(inode->i_flags & S_NOCMTIME))
 388                         inode->i_flags |= S_NOCMTIME;
 389         }
 390
 391         GOTO(put, rc);
 392
 393 put:
 394         if (rc != 0) {
 395                 if (!IS_ERR(inode))
 396                         iput(inode);
 397
 398                 inode = ERR_PTR(rc);
 399         }
 400
 401         return inode;
 402 }
 403
 404 /**
 405  * \retval +v: new filter_fid, does not contain self-fid
 406  * \retval 0:  filter_fid_old, contains self-fid
 407  * \retval -v: other failure cases
 408  */
 409 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 410                  struct dentry *dentry, struct lu_fid *fid)
 411 {
 412         struct filter_fid_old   *ff     = &info->oti_ff;
 413         struct ost_id           *ostid  = &info->oti_ostid;
 414         int                      rc;
 415
 416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 417         if (rc == sizeof(*ff)) {
 418                 rc = 0;
 419                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 420                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 421                 /* XXX: should use real OST index in the future. LU-3569 */
 422                 ostid_to_fid(fid, ostid, 0);
 423         } else if (rc == sizeof(struct filter_fid)) {
 424                 rc = 1;
 425         } else if (rc >= 0) {
 426                 rc = -EINVAL;
 427         }
 428
 429         return rc;
 430 }
 431
 432 static int osd_lma_self_repair(struct osd_thread_info *info,
 433                                struct osd_device *osd, struct inode *inode,
 434                                const struct lu_fid *fid, __u32 compat)
 435 {
 436         handle_t *jh;
 437         int       rc;
 438
 439         LASSERT(current->journal_info == NULL);
 440
 441         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 442                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 443         if (IS_ERR(jh)) {
 444                 rc = PTR_ERR(jh);
 445                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 446                       osd_name(osd), rc);
 447                 return rc;
 448         }
 449
 450         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 451         if (rc != 0)
 452                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 453                       osd_name(osd), rc);
 454         ldiskfs_journal_stop(jh);
 455         return rc;
 456 }
 457
 458 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 459 {
 460         struct osd_thread_info  *info   = osd_oti_get(env);
 461         struct osd_device       *osd    = osd_obj2dev(obj);
 462         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 463         struct inode            *inode  = obj->oo_inode;
 464         struct dentry           *dentry = &info->oti_obj_dentry;
 465         struct lu_fid           *fid    = NULL;
 466         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 467         int                      rc;
 468         ENTRY;
 469
 470         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 471         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 472                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 473         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 474                 fid = &lma->lma_self_fid;
 475                 rc = osd_get_idif(info, inode, dentry, fid);
 476                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 477                         /* For the given OST-object, if it has neither LMA nor
 478                          * FID in XATTR_NAME_FID, then the given FID (which is
 479                          * contained in the @obj, from client RPC for locating
 480                          * the OST-object) is trusted. We use it to generate
 481                          * the LMA. */
 482                         osd_lma_self_repair(info, osd, inode, rfid,
 483                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 484                                 LMAC_FID_ON_OST : 0);
 485                         RETURN(0);
 486                 }
 487         }
 488
 489         if (unlikely(rc == -ENODATA))
 490                 RETURN(0);
 491
 492         if (rc < 0)
 493                 RETURN(rc);
 494
 495         if (rc > 0) {
 496                 rc = 0;
 497                 lustre_lma_swab(lma);
 498                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 499                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 500                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 501                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 502                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 503                               PFID(rfid), inode->i_ino);
 504                         rc = -EOPNOTSUPP;
 505                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 506                         fid = &lma->lma_self_fid;
 507                 }
 508         }
 509
 510         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 511                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 512                         struct ost_id   *oi   = &info->oti_ostid;
 513                         struct lu_fid   *fid1 = &info->oti_fid3;
 514                         __u32            idx  = fid_idif_ost_idx(rfid);
 515
 516                         /* For old IDIF, the OST index is not part of the IDIF,
 517                          * Means that different OSTs may have the same IDIFs.
 518                          * Under such case, we need to make some compatible
 519                          * check to make sure to trigger OI scrub properly. */
 520                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 521                                 /* Given @rfid is new, LMA is old. */
 522                                 fid_to_ostid(fid, oi);
 523                                 ostid_to_fid(fid1, oi, idx);
 524                                 if (lu_fid_eq(fid1, rfid)) {
 525                                         if (osd->od_lma_self_repair)
 526                                                 osd_lma_self_repair(info, osd,
 527                                                         inode, rfid,
 528                                                         LMAC_FID_ON_OST);
 529                                         RETURN(0);
 530                                 }
 531                         }
 532                 }
 533
 534                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 535                        osd_name(osd), PFID(rfid), PFID(fid));
 536                 rc = -EREMCHG;
 537         }
 538
 539         RETURN(rc);
 540 }
 541
 542 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 543                           const struct lu_fid *fid,
 544                           const struct lu_object_conf *conf)
 545 {
 546         struct osd_thread_info *info;
 547         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 548         struct osd_device      *dev;
 549         struct osd_idmap_cache *oic;
 550         struct osd_inode_id    *id;
 551         struct inode           *inode;
 552         struct osd_scrub       *scrub;
 553         struct scrub_file      *sf;
 554         int                     result;
 555         int                     saved  = 0;
 556         bool                    in_oi  = false;
 557         bool                    triggered = false;
 558         ENTRY;
 559
 560         LINVRNT(osd_invariant(obj));
 561         LASSERT(obj->oo_inode == NULL);
 562         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 563
 564         dev = osd_dev(ldev);
 565         scrub = &dev->od_scrub;
 566         sf = &scrub->os_file;
 567         info = osd_oti_get(env);
 568         LASSERT(info);
 569         oic = &info->oti_cache;
 570
 571         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 572                 RETURN(-ENOENT);
 573
 574         /* For the object is created as locking anchor, or for the object to
 575          * be created on disk. No need to osd_oi_lookup() at here because FID
 576          * shouldn't never be re-used, if it's really a duplicate FID from
 577          * unexpected reason, we should be able to detect it later by calling
 578          * do_create->osd_oi_insert(). */
 579         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 580                 GOTO(out, result = 0);
 581
 582         /* Search order: 1. per-thread cache. */
 583         if (lu_fid_eq(fid, &oic->oic_fid) &&
 584             likely(oic->oic_dev == dev)) {
 585                 id = &oic->oic_lid;
 586                 goto iget;
 587         }
 588
 589         id = &info->oti_id;
 590         if (!list_empty(&scrub->os_inconsistent_items)) {
 591                 /* Search order: 2. OI scrub pending list. */
 592                 result = osd_oii_lookup(dev, fid, id);
 593                 if (result == 0)
 594                         goto iget;
 595         }
 596
 597         /* Search order: 3. OI files. */
 598         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 599         if (result == -ENOENT) {
 600                 if (!fid_is_norm(fid) ||
 601                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 602                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 603                                       sf->sf_oi_bitmap))
 604                         GOTO(out, result = 0);
 605
 606                 goto trigger;
 607         }
 608
 609         if (result != 0)
 610                 GOTO(out, result);
 611
 612         in_oi = true;
 613
 614 iget:
 615         inode = osd_iget_check(info, dev, fid, id, in_oi);
 616         if (IS_ERR(inode)) {
 617                 result = PTR_ERR(inode);
 618                 if (result == -ENOENT || result == -ESTALE) {
 619                         if (!in_oi)
 620                                 fid_zero(&oic->oic_fid);
 621
 622                         GOTO(out, result = -ENOENT);
 623                 } else if (result == -EREMCHG) {
 624
 625 trigger:
 626                         if (!in_oi)
 627                                 fid_zero(&oic->oic_fid);
 628
 629                         if (unlikely(triggered))
 630                                 GOTO(out, result = saved);
 631
 632                         triggered = true;
 633                         if (thread_is_running(&scrub->os_thread)) {
 634                                 result = -EINPROGRESS;
 635                         } else if (!dev->od_noscrub) {
 636                                 /* Since we do not know the right OI mapping,
 637                                  * we have to trigger OI scrub to scan the
 638                                  * whole device. */
 639                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
 640                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 641                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 642                                               "for "DFID", rc = %d [1]\n",
 643                                               osd_name(dev), PFID(fid),result);
 644                                 if (result == 0 || result == -EALREADY)
 645                                         result = -EINPROGRESS;
 646                                 else
 647                                         result = -EREMCHG;
 648                         }
 649
 650                         /* We still have chance to get the valid inode: for the
 651                          * object which is referenced by remote name entry, the
 652                          * object on the local MDT will be linked under the dir
 653                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 654                          *
 655                          * We do not know whether the object for the given FID
 656                          * is referenced by some remote name entry or not, and
 657                          * especially for DNE II, a multiple-linked object may
 658                          * have many name entries reside on many MDTs.
 659                          *
 660                          * To simplify the operation, OSD will not distinguish
 661                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 662                          * only happened for the RPC from other MDT during the
 663                          * OI scrub, or for the client side RPC with FID only,
 664                          * such as FID to path, or from old connected client. */
 665                         saved = result;
 666                         result = osd_lookup_in_remote_parent(info, dev,
 667                                                              fid, id);
 668                         if (result == 0) {
 669                                 in_oi = false;
 670                                 goto iget;
 671                         }
 672
 673                         result = saved;
 674                 }
 675
 676                 GOTO(out, result);
 677         }
 678
 679         obj->oo_inode = inode;
 680         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 681
 682         result = osd_check_lma(env, obj);
 683         if (result != 0) {
 684                 iput(inode);
 685                 obj->oo_inode = NULL;
 686                 if (result == -EREMCHG) {
 687                         if (!in_oi) {
 688                                 result = osd_oi_lookup(info, dev, fid, id,
 689                                                        OI_CHECK_FLD);
 690                                 if (result != 0) {
 691                                         fid_zero(&oic->oic_fid);
 692                                         GOTO(out, result);
 693                                 }
 694                         }
 695
 696                         goto trigger;
 697                 }
 698
 699                 GOTO(out, result);
 700         }
 701
 702         obj->oo_compat_dot_created = 1;
 703         obj->oo_compat_dotdot_created = 1;
 704
 705         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 706                 GOTO(out, result = 0);
 707
 708         LASSERT(obj->oo_hl_head == NULL);
 709         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 710         if (obj->oo_hl_head == NULL) {
 711                 obj->oo_inode = NULL;
 712                 iput(inode);
 713                 GOTO(out, result = -ENOMEM);
 714         }
 715         GOTO(out, result = 0);
 716
 717 out:
 718         LINVRNT(osd_invariant(obj));
 719         return result;
 720 }
 721
 722 /*
 723  * Concurrency: shouldn't matter.
 724  */
 725 static void osd_object_init0(struct osd_object *obj)
 726 {
 727         LASSERT(obj->oo_inode != NULL);
 728         obj->oo_dt.do_body_ops = &osd_body_ops;
 729         obj->oo_dt.do_lu.lo_header->loh_attr |=
 730                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 731 }
 732
 733 /*
 734  * Concurrency: no concurrent access is possible that early in object
 735  * life-cycle.
 736  */
 737 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 738                            const struct lu_object_conf *conf)
 739 {
 740         struct osd_object *obj = osd_obj(l);
 741         int result;
 742
 743         LINVRNT(osd_invariant(obj));
 744
 745         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 746                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 747                 l->lo_header->loh_attr |= LOHA_EXISTS;
 748                 return 0;
 749         }
 750
 751         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 752         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 753         if (result == 0 && obj->oo_inode != NULL)
 754                 osd_object_init0(obj);
 755
 756         LINVRNT(osd_invariant(obj));
 757         return result;
 758 }
 759
 760 /*
 761  * Concurrency: no concurrent access is possible that late in object
 762  * life-cycle.
 763  */
 764 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 765 {
 766         struct osd_object *obj = osd_obj(l);
 767
 768         LINVRNT(osd_invariant(obj));
 769
 770         dt_object_fini(&obj->oo_dt);
 771         if (obj->oo_hl_head != NULL)
 772                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 773         OBD_FREE_PTR(obj);
 774 }
 775
 776 /*
 777  * Concurrency: no concurrent access is possible that late in object
 778  * life-cycle.
 779  */
 780 static void osd_index_fini(struct osd_object *o)
 781 {
 782         struct iam_container *bag;
 783
 784         if (o->oo_dir != NULL) {
 785                 bag = &o->oo_dir->od_container;
 786                 if (o->oo_inode != NULL) {
 787                         if (bag->ic_object == o->oo_inode)
 788                                 iam_container_fini(bag);
 789                 }
 790                 OBD_FREE_PTR(o->oo_dir);
 791                 o->oo_dir = NULL;
 792         }
 793 }
 794
 795 /*
 796  * Concurrency: no concurrent access is possible that late in object
 797  * life-cycle (for all existing callers, that is. New callers have to provide
 798  * their own locking.)
 799  */
 800 static int osd_inode_unlinked(const struct inode *inode)
 801 {
 802         return inode->i_nlink == 0;
 803 }
 804
 805 enum {
 806         OSD_TXN_OI_DELETE_CREDITS    = 20,
 807         OSD_TXN_INODE_DELETE_CREDITS = 20
 808 };
 809
 810 /*
 811  * Journal
 812  */
 813
 814 #if OSD_THANDLE_STATS
 815 /**
 816  * Set time when the handle is allocated
 817  */
 818 static void osd_th_alloced(struct osd_thandle *oth)
 819 {
 820         oth->oth_alloced = cfs_time_current();
 821 }
 822
 823 /**
 824  * Set time when the handle started
 825  */
 826 static void osd_th_started(struct osd_thandle *oth)
 827 {
 828         oth->oth_started = cfs_time_current();
 829 }
 830
 831 /**
 832  * Helper function to convert time interval to microseconds packed in
 833  * long int.
 834  */
 835 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 836 {
 837         struct timeval val;
 838
 839         cfs_duration_usec(cfs_time_sub(end, start), &val);
 840         return val.tv_sec * 1000000 + val.tv_usec;
 841 }
 842
 843 /**
 844  * Check whether the we deal with this handle for too long.
 845  */
 846 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 847                                 cfs_time_t alloced, cfs_time_t started,
 848                                 cfs_time_t closed)
 849 {
 850         cfs_time_t now = cfs_time_current();
 851
 852         LASSERT(dev != NULL);
 853
 854         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 855                             interval_to_usec(alloced, started));
 856         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 857                             interval_to_usec(started, closed));
 858         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 859                             interval_to_usec(closed, now));
 860
 861         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 862                 CWARN("transaction handle %p was open for too long: "
 863                       "now "CFS_TIME_T" ,"
 864                       "alloced "CFS_TIME_T" ,"
 865                       "started "CFS_TIME_T" ,"
 866                       "closed "CFS_TIME_T"\n",
 867                       oth, now, alloced, started, closed);
 868                 libcfs_debug_dumpstack(NULL);
 869         }
 870 }
 871
 872 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 873 {                                                                       \
 874         cfs_time_t __closed = cfs_time_current();                       \
 875         cfs_time_t __alloced = oth->oth_alloced;                        \
 876         cfs_time_t __started = oth->oth_started;                        \
 877                                                                         \
 878         expr;                                                           \
 879         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 880 }
 881
 882 #else /* OSD_THANDLE_STATS */
 883
 884 #define osd_th_alloced(h)                  do {} while(0)
 885 #define osd_th_started(h)                  do {} while(0)
 886 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 887
 888 #endif /* OSD_THANDLE_STATS */
 889
 890 /*
 891  * Concurrency: doesn't access mutable data.
 892  */
 893 static int osd_param_is_not_sane(const struct osd_device *dev,
 894                                  const struct thandle *th)
 895 {
 896         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 897
 898         return oh->ot_credits > osd_transaction_size(dev);
 899 }
 900
 901 /*
 902  * Concurrency: shouldn't matter.
 903  */
 904 static void osd_trans_commit_cb(struct super_block *sb,
 905                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 906 {
 907         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 908         struct thandle     *th  = &oh->ot_super;
 909         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 910         struct dt_txn_commit_cb *dcb, *tmp;
 911
 912         LASSERT(oh->ot_handle == NULL);
 913
 914         if (error)
 915                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 916
 917         dt_txn_hook_commit(th);
 918
 919         /* call per-transaction callbacks if any */
 920         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 921                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 922                          "commit callback entry: magic=%x name='%s'\n",
 923                          dcb->dcb_magic, dcb->dcb_name);
 924                 list_del_init(&dcb->dcb_linkage);
 925                 dcb->dcb_func(NULL, th, dcb, error);
 926         }
 927
 928         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 929         lu_device_put(lud);
 930         th->th_dev = NULL;
 931
 932         lu_context_exit(&th->th_ctx);
 933         lu_context_fini(&th->th_ctx);
 934         thandle_put(th);
 935 }
 936
 937 static struct thandle *osd_trans_create(const struct lu_env *env,
 938                                         struct dt_device *d)
 939 {
 940         struct osd_thread_info  *oti = osd_oti_get(env);
 941         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 942         struct osd_thandle      *oh;
 943         struct thandle          *th;
 944         ENTRY;
 945
 946         /* on pending IO in this thread should left from prev. request */
 947         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 948
 949         th = ERR_PTR(-ENOMEM);
 950         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 951         if (oh != NULL) {
 952                 oh->ot_quota_trans = &oti->oti_quota_trans;
 953                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 954                 th = &oh->ot_super;
 955                 th->th_dev = d;
 956                 th->th_result = 0;
 957                 th->th_tags = LCT_TX_HANDLE;
 958                 oh->ot_credits = 0;
 959                 atomic_set(&th->th_refc, 1);
 960                 th->th_alloc_size = sizeof(*oh);
 961                 oti->oti_dev = osd_dt_dev(d);
 962                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 963                 osd_th_alloced(oh);
 964
 965                 memset(oti->oti_declare_ops, 0,
 966                                         sizeof(oti->oti_declare_ops));
 967                 memset(oti->oti_declare_ops_rb, 0,
 968                                         sizeof(oti->oti_declare_ops_rb));
 969                 memset(oti->oti_declare_ops_cred, 0,
 970                                         sizeof(oti->oti_declare_ops_cred));
 971                 oti->oti_rollback = false;
 972         }
 973         RETURN(th);
 974 }
 975
 976 /*
 977  * Concurrency: shouldn't matter.
 978  */
 979 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 980                            struct thandle *th)
 981 {
 982         struct osd_thread_info *oti = osd_oti_get(env);
 983         struct osd_device  *dev = osd_dt_dev(d);
 984         handle_t           *jh;
 985         struct osd_thandle *oh;
 986         int rc;
 987
 988         ENTRY;
 989
 990         LASSERT(current->journal_info == NULL);
 991
 992         oh = container_of0(th, struct osd_thandle, ot_super);
 993         LASSERT(oh != NULL);
 994         LASSERT(oh->ot_handle == NULL);
 995
 996         rc = dt_txn_hook_start(env, d, th);
 997         if (rc != 0)
 998                 GOTO(out, rc);
 999
1000         if (unlikely(osd_param_is_not_sane(dev, th))) {
1001                 static unsigned long last_printed;
1002                 static int last_credits;
1003
1004                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1005                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1006                       oh->ot_credits,
1007                       osd_journal(dev)->j_max_transaction_buffers);
1008                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_CREATE],
1010                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1011                       oti->oti_declare_ops[OSD_OT_DESTROY],
1012                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1013                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1015                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1016                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1017                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1018                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1019                       oti->oti_declare_ops[OSD_OT_WRITE],
1020                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1021                       oti->oti_declare_ops[OSD_OT_PUNCH],
1022                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1023                       oti->oti_declare_ops[OSD_OT_QUOTA],
1024                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1025                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_INSERT],
1027                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1028                       oti->oti_declare_ops[OSD_OT_DELETE],
1029                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1030                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1031                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1032                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1033                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1034                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1035
1036                 if (last_credits != oh->ot_credits &&
1037                     time_after(jiffies, last_printed +
1038                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1039                         libcfs_debug_dumpstack(NULL);
1040                         last_credits = oh->ot_credits;
1041                         last_printed = jiffies;
1042                 }
1043                 /* XXX Limit the credits to 'max_transaction_buffers', and
1044                  *     let the underlying filesystem to catch the error if
1045                  *     we really need so many credits.
1046                  *
1047                  *     This should be removed when we can calculate the
1048                  *     credits precisely. */
1049                 oh->ot_credits = osd_transaction_size(dev);
1050         }
1051
1052         /*
1053          * XXX temporary stuff. Some abstraction layer should
1054          * be used.
1055          */
1056         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1057         osd_th_started(oh);
1058         if (!IS_ERR(jh)) {
1059                 oh->ot_handle = jh;
1060                 LASSERT(oti->oti_txns == 0);
1061                 lu_context_init(&th->th_ctx, th->th_tags);
1062                 lu_context_enter(&th->th_ctx);
1063
1064                 lu_device_get(&d->dd_lu_dev);
1065                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1066                               "osd-tx", th);
1067                 oti->oti_txns++;
1068                 rc = 0;
1069         } else {
1070                 rc = PTR_ERR(jh);
1071         }
1072 out:
1073         RETURN(rc);
1074 }
1075
1076 static int osd_seq_exists(const struct lu_env *env,
1077                               struct osd_device *osd, obd_seq seq)
1078 {
1079         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1080         struct seq_server_site  *ss = osd_seq_site(osd);
1081         int                     rc;
1082         ENTRY;
1083
1084         LASSERT(ss != NULL);
1085         LASSERT(ss->ss_server_fld != NULL);
1086
1087         rc = osd_fld_lookup(env, osd, seq, range);
1088         if (rc != 0) {
1089                 if (rc != -ENOENT)
1090                         CERROR("%s: can't lookup FLD sequence "LPX64
1091                                ": rc = %d\n", osd_name(osd), seq, rc);
1092                 RETURN(0);
1093         }
1094
1095         RETURN(ss->ss_node_id == range->lsr_index);
1096 }
1097
1098 /*
1099  * Concurrency: shouldn't matter.
1100  */
1101 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1102                           struct thandle *th)
1103 {
1104         int                     rc = 0;
1105         struct osd_thandle     *oh;
1106         struct osd_thread_info *oti = osd_oti_get(env);
1107         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1108         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1109         struct lquota_trans    *qtrans;
1110         ENTRY;
1111
1112         oh = container_of0(th, struct osd_thandle, ot_super);
1113
1114         qtrans = oh->ot_quota_trans;
1115         oh->ot_quota_trans = NULL;
1116
1117         if (oh->ot_handle != NULL) {
1118                 handle_t *hdl = oh->ot_handle;
1119
1120                 /*
1121                  * add commit callback
1122                  * notice we don't do this in osd_trans_start()
1123                  * as underlying transaction can change during truncate
1124                  */
1125                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1126                                          &oh->ot_jcb);
1127
1128                 LASSERT(oti->oti_txns == 1);
1129                 oti->oti_txns--;
1130                 rc = dt_txn_hook_stop(env, th);
1131                 if (rc != 0)
1132                         CERROR("Failure in transaction hook: %d\n", rc);
1133
1134                 /* hook functions might modify th_sync */
1135                 hdl->h_sync = th->th_sync;
1136
1137                 oh->ot_handle = NULL;
1138                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1139                                   rc = ldiskfs_journal_stop(hdl));
1140                 if (rc != 0)
1141                         CERROR("Failure to stop transaction: %d\n", rc);
1142         } else {
1143                 thandle_put(&oh->ot_super);
1144         }
1145
1146         /* inform the quota slave device that the transaction is stopping */
1147         qsd_op_end(env, qsd, qtrans);
1148
1149         /* as we want IO to journal and data IO be concurrent, we don't block
1150          * awaiting data IO completion in osd_do_bio(), instead we wait here
1151          * once transaction is submitted to the journal. all reqular requests
1152          * don't do direct IO (except read/write), thus this wait_event becomes
1153          * no-op for them.
1154          *
1155          * IMPORTANT: we have to wait till any IO submited by the thread is
1156          * completed otherwise iobuf may be corrupted by different request
1157          */
1158         wait_event(iobuf->dr_wait,
1159                        atomic_read(&iobuf->dr_numreqs) == 0);
1160         osd_fini_iobuf(oti->oti_dev, iobuf);
1161         if (!rc)
1162                 rc = iobuf->dr_error;
1163
1164         RETURN(rc);
1165 }
1166
1167 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1168 {
1169         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1170                                                ot_super);
1171
1172         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1173         LASSERT(&dcb->dcb_func != NULL);
1174         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1175
1176         return 0;
1177 }
1178
1179 /*
1180  * Called just before object is freed. Releases all resources except for
1181  * object itself (that is released by osd_object_free()).
1182  *
1183  * Concurrency: no concurrent access is possible that late in object
1184  * life-cycle.
1185  */
1186 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1187 {
1188         struct osd_object *obj   = osd_obj(l);
1189         struct inode      *inode = obj->oo_inode;
1190
1191         LINVRNT(osd_invariant(obj));
1192
1193         /*
1194          * If object is unlinked remove fid->ino mapping from object index.
1195          */
1196
1197         osd_index_fini(obj);
1198         if (inode != NULL) {
1199                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1200                 qid_t                    uid = i_uid_read(inode);
1201                 qid_t                    gid = i_gid_read(inode);
1202
1203                 iput(inode);
1204                 obj->oo_inode = NULL;
1205
1206                 if (qsd != NULL) {
1207                         struct osd_thread_info  *info = osd_oti_get(env);
1208                         struct lquota_id_info   *qi = &info->oti_qi;
1209
1210                         /* Release granted quota to master if necessary */
1211                         qi->lqi_id.qid_uid = uid;
1212                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1213
1214                         qi->lqi_id.qid_uid = gid;
1215                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1216                 }
1217         }
1218 }
1219
1220 /*
1221  * Concurrency: ->loo_object_release() is called under site spin-lock.
1222  */
1223 static void osd_object_release(const struct lu_env *env,
1224                                struct lu_object *l)
1225 {
1226 }
1227
1228 /*
1229  * Concurrency: shouldn't matter.
1230  */
1231 static int osd_object_print(const struct lu_env *env, void *cookie,
1232                             lu_printer_t p, const struct lu_object *l)
1233 {
1234         struct osd_object *o = osd_obj(l);
1235         struct iam_descr  *d;
1236
1237         if (o->oo_dir != NULL)
1238                 d = o->oo_dir->od_container.ic_descr;
1239         else
1240                 d = NULL;
1241         return (*p)(env, cookie,
1242                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1243                     o, o->oo_inode,
1244                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1245                     o->oo_inode ? o->oo_inode->i_generation : 0,
1246                     d ? d->id_ops->id_name : "plain");
1247 }
1248
1249 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1250
1251 /*
1252  * Concurrency: shouldn't matter.
1253  */
1254 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1255                struct obd_statfs *sfs)
1256 {
1257         struct osd_device  *osd = osd_dt_dev(d);
1258         struct super_block *sb = osd_sb(osd);
1259         struct kstatfs     *ksfs;
1260         int result = 0;
1261
1262         if (unlikely(osd->od_mnt == NULL))
1263                 return -EINPROGRESS;
1264
1265         /* osd_lproc.c call this without env, allocate ksfs for that case */
1266         if (unlikely(env == NULL)) {
1267                 OBD_ALLOC_PTR(ksfs);
1268                 if (ksfs == NULL)
1269                         return -ENOMEM;
1270         } else {
1271                 ksfs = &osd_oti_get(env)->oti_ksfs;
1272         }
1273
1274         spin_lock(&osd->od_osfs_lock);
1275         result = sb->s_op->statfs(sb->s_root, ksfs);
1276         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1277                 statfs_pack(sfs, ksfs);
1278                 if (sb->s_flags & MS_RDONLY)
1279                         sfs->os_state = OS_STATE_READONLY;
1280         }
1281
1282         spin_unlock(&osd->od_osfs_lock);
1283
1284         if (unlikely(env == NULL))
1285                 OBD_FREE_PTR(ksfs);
1286
1287         /* Reserve a small amount of space for local objects like last_rcvd,
1288          * llog, quota files, ... */
1289         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1290                 sfs->os_bavail = 0;
1291         } else {
1292                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1293                 /** Take out metadata overhead for indirect blocks */
1294                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1295         }
1296
1297         return result;
1298 }
1299
1300 /**
1301  * Estimate space needed for file creations. We assume the largest filename
1302  * which is 2^64 - 1, hence a filename of 20 chars.
1303  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1304  */
1305 #ifdef __LDISKFS_DIR_REC_LEN
1306 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1307 #else
1308 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1309 #endif
1310
1311 /*
1312  * Concurrency: doesn't access mutable data.
1313  */
1314 static void osd_conf_get(const struct lu_env *env,
1315                          const struct dt_device *dev,
1316                          struct dt_device_param *param)
1317 {
1318         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1319         int                ea_overhead;
1320
1321         /*
1322          * XXX should be taken from not-yet-existing fs abstraction layer.
1323          */
1324         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1325         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1326         param->ddp_block_shift  = sb->s_blocksize_bits;
1327         param->ddp_mount_type     = LDD_MT_LDISKFS;
1328         param->ddp_maxbytes       = sb->s_maxbytes;
1329         /* Overhead estimate should be fairly accurate, so we really take a tiny
1330          * error margin which also avoids fragmenting the filesystem too much */
1331         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1332         /* inode are statically allocated, so per-inode space consumption
1333          * is the space consumed by the directory entry */
1334         param->ddp_inodespace     = PER_OBJ_USAGE;
1335         /* per-fragment overhead to be used by the client code */
1336         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1337         param->ddp_mntopts      = 0;
1338         if (test_opt(sb, XATTR_USER))
1339                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1340         if (test_opt(sb, POSIX_ACL))
1341                 param->ddp_mntopts |= MNTOPT_ACL;
1342
1343         /* LOD might calculate the max stripe count based on max_ea_size,
1344          * so we need take account in the overhead as well,
1345          * xattr_header + magic + xattr_entry_head */
1346         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1347                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1348
1349 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1350         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1351                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1352                                                                 ea_overhead;
1353         else
1354 #endif
1355                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1356 }
1357
1358 /*
1359  * Concurrency: shouldn't matter.
1360  */
1361 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1362 {
1363         int rc;
1364
1365         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1366
1367         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1368
1369         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1370                LUSTRE_OSD_LDISKFS_NAME, rc);
1371
1372         return rc;
1373 }
1374
1375 /**
1376  * Start commit for OSD device.
1377  *
1378  * An implementation of dt_commit_async method for OSD device.
1379  * Asychronously starts underlayng fs sync and thereby a transaction
1380  * commit.
1381  *
1382  * \param env environment
1383  * \param d dt device
1384  *
1385  * \see dt_device_operations
1386  */
1387 static int osd_commit_async(const struct lu_env *env,
1388                             struct dt_device *d)
1389 {
1390         struct super_block *s = osd_sb(osd_dt_dev(d));
1391         ENTRY;
1392
1393         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1394         RETURN(s->s_op->sync_fs(s, 0));
1395 }
1396
1397 /*
1398  * Concurrency: shouldn't matter.
1399  */
1400
1401 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1402 {
1403         struct super_block *sb = osd_sb(osd_dt_dev(d));
1404         struct block_device *dev = sb->s_bdev;
1405 #ifdef HAVE_DEV_SET_RDONLY
1406         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1407         int rc = 0;
1408 #else
1409         int rc = -EOPNOTSUPP;
1410 #endif
1411         ENTRY;
1412
1413 #ifdef HAVE_DEV_SET_RDONLY
1414         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1415
1416         if (jdev && (jdev != dev)) {
1417                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1418                        (long)jdev);
1419                 dev_set_rdonly(jdev);
1420         }
1421         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1422         dev_set_rdonly(dev);
1423 #else
1424         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1425                osd_dt_dev(d)->od_svname, (long)dev, rc);
1426 #endif
1427         RETURN(rc);
1428 }
1429
1430 /*
1431  * Concurrency: serialization provided by callers.
1432  */
1433 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1434                               int mode, unsigned long timeout, __u32 alg,
1435                               struct lustre_capa_key *keys)
1436 {
1437         struct osd_device *dev = osd_dt_dev(d);
1438         ENTRY;
1439
1440         dev->od_fl_capa = mode;
1441         dev->od_capa_timeout = timeout;
1442         dev->od_capa_alg = alg;
1443         dev->od_capa_keys = keys;
1444         RETURN(0);
1445 }
1446
1447 /**
1448  * Note: we do not count into QUOTA here.
1449  * If we mount with --data_journal we may need more.
1450  */
1451 const int osd_dto_credits_noquota[DTO_NR] = {
1452         /**
1453          * Insert.
1454          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1455          * SINGLEDATA_TRANS_BLOCKS(8)
1456          * XXX Note: maybe iam need more, since iam have more level than
1457          *           EXT3 htree.
1458          */
1459         [DTO_INDEX_INSERT]  = 16,
1460         /**
1461          * Delete
1462          * just modify a single entry, probably merge few within a block
1463          */
1464         [DTO_INDEX_DELETE]  = 1,
1465         /**
1466          * Used for OI scrub
1467          */
1468         [DTO_INDEX_UPDATE]  = 16,
1469         /**
1470          * 4(inode, inode bits, groups, GDT)
1471          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1472          */
1473         [DTO_OBJECT_CREATE] = 4,
1474         /**
1475          * 4(inode, inode bits, groups, GDT)
1476          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1477          */
1478         [DTO_OBJECT_DELETE] = 4,
1479         /**
1480          * Attr set credits (inode)
1481          */
1482         [DTO_ATTR_SET_BASE] = 1,
1483         /**
1484          * Xattr set. The same as xattr of EXT3.
1485          * DATA_TRANS_BLOCKS(14)
1486          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1487          * are also counted in. Do not know why?
1488          */
1489         [DTO_XATTR_SET]     = 14,
1490         /**
1491          * credits for inode change during write.
1492          */
1493         [DTO_WRITE_BASE]    = 3,
1494         /**
1495          * credits for single block write.
1496          */
1497         [DTO_WRITE_BLOCK]   = 14,
1498         /**
1499          * Attr set credits for chown.
1500          * This is extra credits for setattr, and it is null without quota
1501          */
1502         [DTO_ATTR_SET_CHOWN] = 0
1503 };
1504
1505 static const struct dt_device_operations osd_dt_ops = {
1506         .dt_root_get       = osd_root_get,
1507         .dt_statfs         = osd_statfs,
1508         .dt_trans_create   = osd_trans_create,
1509         .dt_trans_start    = osd_trans_start,
1510         .dt_trans_stop     = osd_trans_stop,
1511         .dt_trans_cb_add   = osd_trans_cb_add,
1512         .dt_conf_get       = osd_conf_get,
1513         .dt_sync           = osd_sync,
1514         .dt_ro             = osd_ro,
1515         .dt_commit_async   = osd_commit_async,
1516         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1517 };
1518
1519 static void osd_object_read_lock(const struct lu_env *env,
1520                                  struct dt_object *dt, unsigned role)
1521 {
1522         struct osd_object *obj = osd_dt_obj(dt);
1523         struct osd_thread_info *oti = osd_oti_get(env);
1524
1525         LINVRNT(osd_invariant(obj));
1526
1527         LASSERT(obj->oo_owner != env);
1528         down_read_nested(&obj->oo_sem, role);
1529
1530         LASSERT(obj->oo_owner == NULL);
1531         oti->oti_r_locks++;
1532 }
1533
1534 static void osd_object_write_lock(const struct lu_env *env,
1535                                   struct dt_object *dt, unsigned role)
1536 {
1537         struct osd_object *obj = osd_dt_obj(dt);
1538         struct osd_thread_info *oti = osd_oti_get(env);
1539
1540         LINVRNT(osd_invariant(obj));
1541
1542         LASSERT(obj->oo_owner != env);
1543         down_write_nested(&obj->oo_sem, role);
1544
1545         LASSERT(obj->oo_owner == NULL);
1546         obj->oo_owner = env;
1547         oti->oti_w_locks++;
1548 }
1549
1550 static void osd_object_read_unlock(const struct lu_env *env,
1551                                    struct dt_object *dt)
1552 {
1553         struct osd_object *obj = osd_dt_obj(dt);
1554         struct osd_thread_info *oti = osd_oti_get(env);
1555
1556         LINVRNT(osd_invariant(obj));
1557
1558         LASSERT(oti->oti_r_locks > 0);
1559         oti->oti_r_locks--;
1560         up_read(&obj->oo_sem);
1561 }
1562
1563 static void osd_object_write_unlock(const struct lu_env *env,
1564                                     struct dt_object *dt)
1565 {
1566         struct osd_object *obj = osd_dt_obj(dt);
1567         struct osd_thread_info *oti = osd_oti_get(env);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         LASSERT(obj->oo_owner == env);
1572         LASSERT(oti->oti_w_locks > 0);
1573         oti->oti_w_locks--;
1574         obj->oo_owner = NULL;
1575         up_write(&obj->oo_sem);
1576 }
1577
1578 static int osd_object_write_locked(const struct lu_env *env,
1579                                    struct dt_object *dt)
1580 {
1581         struct osd_object *obj = osd_dt_obj(dt);
1582
1583         LINVRNT(osd_invariant(obj));
1584
1585         return obj->oo_owner == env;
1586 }
1587
1588 static int capa_is_sane(const struct lu_env *env,
1589                         struct osd_device *dev,
1590                         struct lustre_capa *capa,
1591                         struct lustre_capa_key *keys)
1592 {
1593         struct osd_thread_info *oti = osd_oti_get(env);
1594         struct lustre_capa *tcapa = &oti->oti_capa;
1595         struct obd_capa *oc;
1596         int i, rc = 0;
1597         ENTRY;
1598
1599         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1600         if (oc) {
1601                 if (capa_is_expired(oc)) {
1602                         DEBUG_CAPA(D_ERROR, capa, "expired");
1603                         rc = -ESTALE;
1604                 }
1605                 capa_put(oc);
1606                 RETURN(rc);
1607         }
1608
1609         if (capa_is_expired_sec(capa)) {
1610                 DEBUG_CAPA(D_ERROR, capa, "expired");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         spin_lock(&capa_lock);
1615         for (i = 0; i < 2; i++) {
1616                 if (keys[i].lk_keyid == capa->lc_keyid) {
1617                         oti->oti_capa_key = keys[i];
1618                         break;
1619                 }
1620         }
1621         spin_unlock(&capa_lock);
1622
1623         if (i == 2) {
1624                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1625                 RETURN(-ESTALE);
1626         }
1627
1628         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1629         if (rc)
1630                 RETURN(rc);
1631
1632         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1633                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1634                 RETURN(-EACCES);
1635         }
1636
1637         oc = capa_add(dev->od_capa_hash, capa);
1638         capa_put(oc);
1639
1640         RETURN(0);
1641 }
1642
1643 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1644                     struct lustre_capa *capa, __u64 opc)
1645 {
1646         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1647         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1648         struct lu_capainfo *lci;
1649         int rc;
1650
1651         if (!osd->od_fl_capa)
1652                 return 0;
1653
1654         if (capa == BYPASS_CAPA)
1655                 return 0;
1656
1657         lci = lu_capainfo_get(env);
1658         if (unlikely(lci == NULL))
1659                 return 0;
1660
1661         if (lci->lci_auth == LC_ID_NONE)
1662                 return 0;
1663
1664         if (capa == NULL) {
1665                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1666                        osd_name(osd), PFID(fid), -EACCES);
1667                 return -EACCES;
1668         }
1669
1670         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1671                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1672                            PFID(fid));
1673                 return -EACCES;
1674         }
1675
1676         if (!capa_opc_supported(capa, opc)) {
1677                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1678                 return -EACCES;
1679         }
1680
1681         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1682         if (rc != 0) {
1683                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1684                 return -EACCES;
1685         }
1686
1687         return 0;
1688 }
1689
1690 static struct timespec *osd_inode_time(const struct lu_env *env,
1691                                        struct inode *inode, __u64 seconds)
1692 {
1693         struct osd_thread_info  *oti = osd_oti_get(env);
1694         struct timespec         *t   = &oti->oti_time;
1695
1696         t->tv_sec = seconds;
1697         t->tv_nsec = 0;
1698         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1699         return t;
1700 }
1701
1702
1703 static void osd_inode_getattr(const struct lu_env *env,
1704                               struct inode *inode, struct lu_attr *attr)
1705 {
1706         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1707                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1708                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1709                            LA_TYPE;
1710
1711         attr->la_atime   = LTIME_S(inode->i_atime);
1712         attr->la_mtime   = LTIME_S(inode->i_mtime);
1713         attr->la_ctime   = LTIME_S(inode->i_ctime);
1714         attr->la_mode    = inode->i_mode;
1715         attr->la_size    = i_size_read(inode);
1716         attr->la_blocks  = inode->i_blocks;
1717         attr->la_uid     = i_uid_read(inode);
1718         attr->la_gid     = i_gid_read(inode);
1719         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1720         attr->la_nlink   = inode->i_nlink;
1721         attr->la_rdev    = inode->i_rdev;
1722         attr->la_blksize = 1 << inode->i_blkbits;
1723         attr->la_blkbits = inode->i_blkbits;
1724 }
1725
1726 static int osd_attr_get(const struct lu_env *env,
1727                         struct dt_object *dt,
1728                         struct lu_attr *attr,
1729                         struct lustre_capa *capa)
1730 {
1731         struct osd_object *obj = osd_dt_obj(dt);
1732
1733         LASSERT(dt_object_exists(dt));
1734         LASSERT(!dt_object_remote(dt));
1735         LINVRNT(osd_invariant(obj));
1736
1737         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1738                 return -EACCES;
1739
1740         spin_lock(&obj->oo_guard);
1741         osd_inode_getattr(env, obj->oo_inode, attr);
1742         spin_unlock(&obj->oo_guard);
1743         return 0;
1744 }
1745
1746 static int osd_declare_attr_set(const struct lu_env *env,
1747                                 struct dt_object *dt,
1748                                 const struct lu_attr *attr,
1749                                 struct thandle *handle)
1750 {
1751         struct osd_thandle     *oh;
1752         struct osd_object      *obj;
1753         struct osd_thread_info *info = osd_oti_get(env);
1754         struct lquota_id_info  *qi = &info->oti_qi;
1755         qid_t                   uid;
1756         qid_t                   gid;
1757         long long               bspace;
1758         int                     rc = 0;
1759         bool                    enforce;
1760         ENTRY;
1761
1762         LASSERT(dt != NULL);
1763         LASSERT(handle != NULL);
1764
1765         obj = osd_dt_obj(dt);
1766         LASSERT(osd_invariant(obj));
1767
1768         oh = container_of0(handle, struct osd_thandle, ot_super);
1769         LASSERT(oh->ot_handle == NULL);
1770
1771         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1772                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1773
1774         if (attr == NULL || obj->oo_inode == NULL)
1775                 RETURN(rc);
1776
1777         bspace   = obj->oo_inode->i_blocks;
1778         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1779         bspace   = toqb(bspace);
1780
1781         /* Changing ownership is always preformed by super user, it should not
1782          * fail with EDQUOT.
1783          *
1784          * We still need to call the osd_declare_qid() to calculate the journal
1785          * credits for updating quota accounting files and to trigger quota
1786          * space adjustment once the operation is completed.*/
1787         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1788                 /* USERQUOTA */
1789                 uid = i_uid_read(obj->oo_inode);
1790                 qi->lqi_type = USRQUOTA;
1791                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1792                 /* inode accounting */
1793                 qi->lqi_is_blk = false;
1794
1795                 /* one more inode for the new uid ... */
1796                 qi->lqi_id.qid_uid = attr->la_uid;
1797                 qi->lqi_space      = 1;
1798                 /* Reserve credits for the new uid */
1799                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1800                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1801                         rc = 0;
1802                 if (rc)
1803                         RETURN(rc);
1804
1805                 /* and one less inode for the current uid */
1806                 qi->lqi_id.qid_uid = uid;
1807                 qi->lqi_space      = -1;
1808                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1809                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1810                         rc = 0;
1811                 if (rc)
1812                         RETURN(rc);
1813
1814                 /* block accounting */
1815                 qi->lqi_is_blk = true;
1816
1817                 /* more blocks for the new uid ... */
1818                 qi->lqi_id.qid_uid = attr->la_uid;
1819                 qi->lqi_space      = bspace;
1820                 /*
1821                  * Credits for the new uid has been reserved, re-use "obj"
1822                  * to save credit reservation.
1823                  */
1824                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1825                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1826                         rc = 0;
1827                 if (rc)
1828                         RETURN(rc);
1829
1830                 /* and finally less blocks for the current uid */
1831                 qi->lqi_id.qid_uid = uid;
1832                 qi->lqi_space      = -bspace;
1833                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1834                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1835                         rc = 0;
1836                 if (rc)
1837                         RETURN(rc);
1838
1839                 /* GROUP QUOTA */
1840                 gid = i_gid_read(obj->oo_inode);
1841                 qi->lqi_type = GRPQUOTA;
1842                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1843
1844                 /* inode accounting */
1845                 qi->lqi_is_blk = false;
1846
1847                 /* one more inode for the new gid ... */
1848                 qi->lqi_id.qid_gid = attr->la_gid;
1849                 qi->lqi_space      = 1;
1850                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1851                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1852                         rc = 0;
1853                 if (rc)
1854                         RETURN(rc);
1855
1856                 /* and one less inode for the current gid */
1857                 qi->lqi_id.qid_gid = gid;
1858                 qi->lqi_space      = -1;
1859                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1860                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1861                         rc = 0;
1862                 if (rc)
1863                         RETURN(rc);
1864
1865                 /* block accounting */
1866                 qi->lqi_is_blk = true;
1867
1868                 /* more blocks for the new gid ... */
1869                 qi->lqi_id.qid_gid = attr->la_gid;
1870                 qi->lqi_space      = bspace;
1871                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1872                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1873                         rc = 0;
1874                 if (rc)
1875                         RETURN(rc);
1876
1877                 /* and finally less blocks for the current gid */
1878                 qi->lqi_id.qid_gid = gid;
1879                 qi->lqi_space      = -bspace;
1880                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1881                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1882                         rc = 0;
1883                 if (rc)
1884                         RETURN(rc);
1885         }
1886
1887         RETURN(rc);
1888 }
1889
1890 static int osd_inode_setattr(const struct lu_env *env,
1891                              struct inode *inode, const struct lu_attr *attr)
1892 {
1893         __u64 bits = attr->la_valid;
1894
1895         /* Only allow set size for regular file */
1896         if (!S_ISREG(inode->i_mode))
1897                 bits &= ~(LA_SIZE | LA_BLOCKS);
1898
1899         if (bits == 0)
1900                 return 0;
1901
1902         if (bits & LA_ATIME)
1903                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1904         if (bits & LA_CTIME)
1905                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1906         if (bits & LA_MTIME)
1907                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1908         if (bits & LA_SIZE) {
1909                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1910                 i_size_write(inode, attr->la_size);
1911         }
1912
1913 #if 0
1914         /* OSD should not change "i_blocks" which is used by quota.
1915          * "i_blocks" should be changed by ldiskfs only. */
1916         if (bits & LA_BLOCKS)
1917                 inode->i_blocks = attr->la_blocks;
1918 #endif
1919         if (bits & LA_MODE)
1920                 inode->i_mode = (inode->i_mode & S_IFMT) |
1921                                 (attr->la_mode & ~S_IFMT);
1922         if (bits & LA_UID)
1923                 i_uid_write(inode, attr->la_uid);
1924         if (bits & LA_GID)
1925                 i_gid_write(inode, attr->la_gid);
1926         if (bits & LA_NLINK)
1927                 set_nlink(inode, attr->la_nlink);
1928         if (bits & LA_RDEV)
1929                 inode->i_rdev = attr->la_rdev;
1930
1931         if (bits & LA_FLAGS) {
1932                 /* always keep S_NOCMTIME */
1933                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1934                                  S_NOCMTIME;
1935         }
1936         return 0;
1937 }
1938
1939 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1940 {
1941         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1942             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1943                 struct iattr    iattr;
1944                 int             rc;
1945
1946                 ll_vfs_dq_init(inode);
1947                 iattr.ia_valid = 0;
1948                 if (attr->la_valid & LA_UID)
1949                         iattr.ia_valid |= ATTR_UID;
1950                 if (attr->la_valid & LA_GID)
1951                         iattr.ia_valid |= ATTR_GID;
1952                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1953                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1954
1955                 rc = ll_vfs_dq_transfer(inode, &iattr);
1956                 if (rc) {
1957                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1958                                "enforcement enabled on the ldiskfs "
1959                                "filesystem?\n", inode->i_sb->s_id, rc);
1960                         return rc;
1961                 }
1962         }
1963         return 0;
1964 }
1965
1966 static int osd_attr_set(const struct lu_env *env,
1967                         struct dt_object *dt,
1968                         const struct lu_attr *attr,
1969                         struct thandle *handle,
1970                         struct lustre_capa *capa)
1971 {
1972         struct osd_object *obj = osd_dt_obj(dt);
1973         struct inode      *inode;
1974         int rc;
1975
1976         LASSERT(handle != NULL);
1977         LASSERT(dt_object_exists(dt));
1978         LASSERT(!dt_object_remote(dt));
1979         LASSERT(osd_invariant(obj));
1980
1981         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1982                 return -EACCES;
1983
1984         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1985
1986         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1987                 struct osd_thread_info  *oti  = osd_oti_get(env);
1988                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1989                 struct lu_fid           *fid1 = &oti->oti_fid;
1990                 struct osd_inode_id     *id   = &oti->oti_id;
1991                 struct iam_path_descr   *ipd;
1992                 struct iam_container    *bag;
1993                 struct osd_thandle      *oh;
1994                 int                      rc;
1995
1996                 fid_cpu_to_be(fid1, fid0);
1997                 memset(id, 1, sizeof(*id));
1998                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1999                                   fid0)->oi_dir.od_container;
2000                 ipd = osd_idx_ipd_get(env, bag);
2001                 if (unlikely(ipd == NULL))
2002                         RETURN(-ENOMEM);
2003
2004                 oh = container_of0(handle, struct osd_thandle, ot_super);
2005                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2006                                 (const struct iam_rec *)id, ipd);
2007                 osd_ipd_put(env, bag, ipd);
2008                 return(rc > 0 ? 0 : rc);
2009         }
2010
2011         inode = obj->oo_inode;
2012
2013         rc = osd_quota_transfer(inode, attr);
2014         if (rc)
2015                 return rc;
2016
2017         spin_lock(&obj->oo_guard);
2018         rc = osd_inode_setattr(env, inode, attr);
2019         spin_unlock(&obj->oo_guard);
2020
2021         if (!rc)
2022                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2023         return rc;
2024 }
2025
2026 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2027                                            struct osd_object *obj,
2028                                            const char *name, const int namelen)
2029 {
2030         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2031 }
2032
2033 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2034                       umode_t mode, struct dt_allocation_hint *hint,
2035                       struct thandle *th)
2036 {
2037         int result;
2038         struct osd_device  *osd = osd_obj2dev(obj);
2039         struct osd_thandle *oth;
2040         struct dt_object   *parent = NULL;
2041         struct inode       *inode;
2042
2043         LINVRNT(osd_invariant(obj));
2044         LASSERT(obj->oo_inode == NULL);
2045         LASSERT(obj->oo_hl_head == NULL);
2046
2047         if (S_ISDIR(mode) && ldiskfs_pdo) {
2048                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2049                 if (obj->oo_hl_head == NULL)
2050                         return -ENOMEM;
2051         }
2052
2053         oth = container_of(th, struct osd_thandle, ot_super);
2054         LASSERT(oth->ot_handle->h_transaction != NULL);
2055
2056         if (hint && hint->dah_parent)
2057                 parent = hint->dah_parent;
2058
2059         inode = ldiskfs_create_inode(oth->ot_handle,
2060                                      parent ? osd_dt_obj(parent)->oo_inode :
2061                                               osd_sb(osd)->s_root->d_inode,
2062                                      mode);
2063         if (!IS_ERR(inode)) {
2064                 /* Do not update file c/mtime in ldiskfs.
2065                  * NB: don't need any lock because no contention at this
2066                  * early stage */
2067                 inode->i_flags |= S_NOCMTIME;
2068
2069                 /* For new created object, it must be consistent,
2070                  * and it is unnecessary to scrub against it. */
2071                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2072                 obj->oo_inode = inode;
2073                 result = 0;
2074         } else {
2075                 if (obj->oo_hl_head != NULL) {
2076                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2077                         obj->oo_hl_head = NULL;
2078                 }
2079                 result = PTR_ERR(inode);
2080         }
2081         LINVRNT(osd_invariant(obj));
2082         return result;
2083 }
2084
2085 enum {
2086         OSD_NAME_LEN = 255
2087 };
2088
2089 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2090                      struct lu_attr *attr,
2091                      struct dt_allocation_hint *hint,
2092                      struct dt_object_format *dof,
2093                      struct thandle *th)
2094 {
2095         int result;
2096         struct osd_thandle *oth;
2097         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2098
2099         LASSERT(S_ISDIR(attr->la_mode));
2100
2101         oth = container_of(th, struct osd_thandle, ot_super);
2102         LASSERT(oth->ot_handle->h_transaction != NULL);
2103         result = osd_mkfile(info, obj, mode, hint, th);
2104
2105         return result;
2106 }
2107
2108 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2109                         struct lu_attr *attr,
2110                         struct dt_allocation_hint *hint,
2111                         struct dt_object_format *dof,
2112                         struct thandle *th)
2113 {
2114         int result;
2115         struct osd_thandle *oth;
2116         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2117
2118         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2119
2120         LASSERT(S_ISREG(attr->la_mode));
2121
2122         oth = container_of(th, struct osd_thandle, ot_super);
2123         LASSERT(oth->ot_handle->h_transaction != NULL);
2124
2125         result = osd_mkfile(info, obj, mode, hint, th);
2126         if (result == 0) {
2127                 LASSERT(obj->oo_inode != NULL);
2128                 if (feat->dif_flags & DT_IND_VARKEY)
2129                         result = iam_lvar_create(obj->oo_inode,
2130                                                  feat->dif_keysize_max,
2131                                                  feat->dif_ptrsize,
2132                                                  feat->dif_recsize_max,
2133                                                  oth->ot_handle);
2134                 else
2135                         result = iam_lfix_create(obj->oo_inode,
2136                                                  feat->dif_keysize_max,
2137                                                  feat->dif_ptrsize,
2138                                                  feat->dif_recsize_max,
2139                                                  oth->ot_handle);
2140
2141         }
2142         return result;
2143 }
2144
2145 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2146                      struct lu_attr *attr,
2147                      struct dt_allocation_hint *hint,
2148                      struct dt_object_format *dof,
2149                      struct thandle *th)
2150 {
2151         LASSERT(S_ISREG(attr->la_mode));
2152         return osd_mkfile(info, obj, (attr->la_mode &
2153                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2154 }
2155
2156 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2157                      struct lu_attr *attr,
2158                      struct dt_allocation_hint *hint,
2159                      struct dt_object_format *dof,
2160                      struct thandle *th)
2161 {
2162         LASSERT(S_ISLNK(attr->la_mode));
2163         return osd_mkfile(info, obj, (attr->la_mode &
2164                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2165 }
2166
2167 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2168                      struct lu_attr *attr,
2169                      struct dt_allocation_hint *hint,
2170                      struct dt_object_format *dof,
2171                      struct thandle *th)
2172 {
2173         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2174         int result;
2175
2176         LINVRNT(osd_invariant(obj));
2177         LASSERT(obj->oo_inode == NULL);
2178         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2179                 S_ISFIFO(mode) || S_ISSOCK(mode));
2180
2181         result = osd_mkfile(info, obj, mode, hint, th);
2182         if (result == 0) {
2183                 LASSERT(obj->oo_inode != NULL);
2184                 /*
2185                  * This inode should be marked dirty for i_rdev.  Currently
2186                  * that is done in the osd_attr_init().
2187                  */
2188                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2189                                    attr->la_rdev);
2190         }
2191         LINVRNT(osd_invariant(obj));
2192         return result;
2193 }
2194
2195 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2196                               struct lu_attr *,
2197                               struct dt_allocation_hint *hint,
2198                               struct dt_object_format *dof,
2199                               struct thandle *);
2200
2201 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2202 {
2203         osd_obj_type_f result;
2204
2205         switch (type) {
2206         case DFT_DIR:
2207                 result = osd_mkdir;
2208                 break;
2209         case DFT_REGULAR:
2210                 result = osd_mkreg;
2211                 break;
2212         case DFT_SYM:
2213                 result = osd_mksym;
2214                 break;
2215         case DFT_NODE:
2216                 result = osd_mknod;
2217                 break;
2218         case DFT_INDEX:
2219                 result = osd_mk_index;
2220                 break;
2221
2222         default:
2223                 LBUG();
2224                 break;
2225         }
2226         return result;
2227 }
2228
2229
2230 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2231                         struct dt_object *parent, struct dt_object *child,
2232                         umode_t child_mode)
2233 {
2234         LASSERT(ah);
2235
2236         ah->dah_parent = parent;
2237         ah->dah_mode = child_mode;
2238 }
2239
2240 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2241                           struct lu_attr *attr, struct dt_object_format *dof)
2242 {
2243         struct inode   *inode = obj->oo_inode;
2244         __u64           valid = attr->la_valid;
2245         int             result;
2246
2247         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2248
2249         if (dof->dof_type != DFT_NODE)
2250                 attr->la_valid &= ~LA_RDEV;
2251         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2252                 attr->la_valid &= ~LA_ATIME;
2253         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2254                 attr->la_valid &= ~LA_CTIME;
2255         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2256                 attr->la_valid &= ~LA_MTIME;
2257
2258         result = osd_quota_transfer(inode, attr);
2259         if (result)
2260                 return;
2261
2262         if (attr->la_valid != 0) {
2263                 result = osd_inode_setattr(info->oti_env, inode, attr);
2264                 /*
2265                  * The osd_inode_setattr() should always succeed here.  The
2266                  * only error that could be returned is EDQUOT when we are
2267                  * trying to change the UID or GID of the inode. However, this
2268                  * should not happen since quota enforcement is no longer
2269                  * enabled on ldiskfs (lquota takes care of it).
2270                  */
2271                 LASSERTF(result == 0, "%d\n", result);
2272                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2273         }
2274
2275         attr->la_valid = valid;
2276 }
2277
2278 /**
2279  * Helper function for osd_object_create()
2280  *
2281  * \retval 0, on success
2282  */
2283 static int __osd_object_create(struct osd_thread_info *info,
2284                                struct osd_object *obj, struct lu_attr *attr,
2285                                struct dt_allocation_hint *hint,
2286                                struct dt_object_format *dof,
2287                                struct thandle *th)
2288 {
2289         int     result;
2290         __u32   umask;
2291
2292         /* we drop umask so that permissions we pass are not affected */
2293         umask = current->fs->umask;
2294         current->fs->umask = 0;
2295
2296         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2297                                                   th);
2298         if (result == 0) {
2299                 osd_attr_init(info, obj, attr, dof);
2300                 osd_object_init0(obj);
2301                 /* bz 24037 */
2302                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2303                         unlock_new_inode(obj->oo_inode);
2304         }
2305
2306         /* restore previous umask value */
2307         current->fs->umask = umask;
2308
2309         return result;
2310 }
2311
2312 /**
2313  * Helper function for osd_object_create()
2314  *
2315  * \retval 0, on success
2316  */
2317 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2318                            const struct lu_fid *fid, struct thandle *th)
2319 {
2320         struct osd_thread_info *info = osd_oti_get(env);
2321         struct osd_inode_id    *id   = &info->oti_id;
2322         struct osd_device      *osd  = osd_obj2dev(obj);
2323         struct osd_thandle     *oh;
2324
2325         LASSERT(obj->oo_inode != NULL);
2326
2327         oh = container_of0(th, struct osd_thandle, ot_super);
2328         LASSERT(oh->ot_handle);
2329
2330         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2331         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2332 }
2333
2334 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2335                    obd_seq seq, struct lu_seq_range *range)
2336 {
2337         struct seq_server_site  *ss = osd_seq_site(osd);
2338
2339         if (fid_seq_is_idif(seq)) {
2340                 fld_range_set_ost(range);
2341                 range->lsr_index = idif_ost_idx(seq);
2342                 return 0;
2343         }
2344
2345         if (!fid_seq_in_fldb(seq)) {
2346                 fld_range_set_mdt(range);
2347                 if (ss != NULL)
2348                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2349                          * at all */
2350                         range->lsr_index = ss->ss_node_id;
2351                 return 0;
2352         }
2353
2354         LASSERT(ss != NULL);
2355         fld_range_set_any(range);
2356         /* OSD will only do local fld lookup */
2357         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2358 }
2359
2360 /*
2361  * Concurrency: no external locking is necessary.
2362  */
2363 static int osd_declare_object_create(const struct lu_env *env,
2364                                      struct dt_object *dt,
2365                                      struct lu_attr *attr,
2366                                      struct dt_allocation_hint *hint,
2367                                      struct dt_object_format *dof,
2368                                      struct thandle *handle)
2369 {
2370         struct osd_thandle      *oh;
2371         int                      rc;
2372         ENTRY;
2373
2374         LASSERT(handle != NULL);
2375
2376         oh = container_of0(handle, struct osd_thandle, ot_super);
2377         LASSERT(oh->ot_handle == NULL);
2378
2379         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2380                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2381         /* Reuse idle OI block may cause additional one OI block
2382          * to be changed. */
2383         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2384                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2385
2386         /* If this is directory, then we expect . and .. to be inserted as
2387          * well. The one directory block always needs to be created for the
2388          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2389          * block), there is no danger of needing a tree for the first block.
2390          */
2391         if (attr && S_ISDIR(attr->la_mode)) {
2392                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2393                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2394                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2395         }
2396
2397         if (!attr)
2398                 RETURN(0);
2399
2400         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2401                                    osd_dt_obj(dt), false, NULL, false);
2402         if (rc != 0)
2403                 RETURN(rc);
2404
2405         RETURN(rc);
2406 }
2407
2408 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2409                              struct lu_attr *attr,
2410                              struct dt_allocation_hint *hint,
2411                              struct dt_object_format *dof, struct thandle *th)
2412 {
2413         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2414         struct osd_object       *obj    = osd_dt_obj(dt);
2415         struct osd_thread_info  *info   = osd_oti_get(env);
2416         int result;
2417         ENTRY;
2418
2419         LINVRNT(osd_invariant(obj));
2420         LASSERT(!dt_object_exists(dt));
2421         LASSERT(!dt_object_remote(dt));
2422         LASSERT(osd_write_locked(env, obj));
2423         LASSERT(th != NULL);
2424
2425         if (unlikely(fid_is_acct(fid)))
2426                 /* Quota files can't be created from the kernel any more,
2427                  * 'tune2fs -O quota' will take care of creating them */
2428                 RETURN(-EPERM);
2429
2430         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2431         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2432
2433         result = __osd_object_create(info, obj, attr, hint, dof, th);
2434         if (result == 0)
2435                 result = __osd_oi_insert(env, obj, fid, th);
2436
2437         LASSERT(ergo(result == 0,
2438                      dt_object_exists(dt) && !dt_object_remote(dt)));
2439
2440         LASSERT(osd_invariant(obj));
2441         RETURN(result);
2442 }
2443
2444 /**
2445  * Called to destroy on-disk representation of the object
2446  *
2447  * Concurrency: must be locked
2448  */
2449 static int osd_declare_object_destroy(const struct lu_env *env,
2450                                       struct dt_object *dt,
2451                                       struct thandle *th)
2452 {
2453         struct osd_object  *obj = osd_dt_obj(dt);
2454         struct inode       *inode = obj->oo_inode;
2455         struct osd_thandle *oh;
2456         int                 rc;
2457         ENTRY;
2458
2459         oh = container_of0(th, struct osd_thandle, ot_super);
2460         LASSERT(oh->ot_handle == NULL);
2461         LASSERT(inode);
2462
2463         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2464                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2465         /* Recycle idle OI leaf may cause additional three OI blocks
2466          * to be changed. */
2467         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2468                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2469         /* one less inode */
2470         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2471                                    -1, oh, obj, false, NULL, false);
2472         if (rc)
2473                 RETURN(rc);
2474         /* data to be truncated */
2475         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2476                                    0, oh, obj, true, NULL, false);
2477         RETURN(rc);
2478 }
2479
2480 static int osd_object_destroy(const struct lu_env *env,
2481                               struct dt_object *dt,
2482                               struct thandle *th)
2483 {
2484         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2485         struct osd_object      *obj = osd_dt_obj(dt);
2486         struct inode           *inode = obj->oo_inode;
2487         struct osd_device      *osd = osd_obj2dev(obj);
2488         struct osd_thandle     *oh;
2489         int                     result;
2490         ENTRY;
2491
2492         oh = container_of0(th, struct osd_thandle, ot_super);
2493         LASSERT(oh->ot_handle);
2494         LASSERT(inode);
2495         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2496
2497         if (unlikely(fid_is_acct(fid)))
2498                 RETURN(-EPERM);
2499
2500         if (S_ISDIR(inode->i_mode)) {
2501                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2502                         inode->i_nlink == 2);
2503                 /* it will check/delete the inode from remote parent,
2504                  * how to optimize it? unlink performance impaction XXX */
2505                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2506                 if (result != 0 && result != -ENOENT) {
2507                         CERROR("%s: delete inode "DFID": rc = %d\n",
2508                                osd_name(osd), PFID(fid), result);
2509                 }
2510                 spin_lock(&obj->oo_guard);
2511                 clear_nlink(inode);
2512                 spin_unlock(&obj->oo_guard);
2513                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2514         }
2515
2516         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2517
2518         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2519                                OI_CHECK_FLD);
2520
2521         /* XXX: add to ext3 orphan list */
2522         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2523
2524         /* not needed in the cache anymore */
2525         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2526
2527         RETURN(0);
2528 }
2529
2530 /**
2531  * Put the fid into lustre_mdt_attrs, and then place the structure
2532  * inode's ea. This fid should not be altered during the life time
2533  * of the inode.
2534  *
2535  * \retval +ve, on success
2536  * \retval -ve, on error
2537  *
2538  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2539  */
2540 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2541                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2542 {
2543         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2544         int                      rc;
2545         ENTRY;
2546
2547         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2548                 RETURN(0);
2549
2550         lustre_lma_init(lma, fid, compat, incompat);
2551         lustre_lma_swab(lma);
2552
2553         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2554                              XATTR_CREATE);
2555         /* LMA may already exist, but we need to check that all the
2556          * desired compat/incompat flags have been added. */
2557         if (unlikely(rc == -EEXIST)) {
2558                 if (compat == 0 && incompat == 0)
2559                         RETURN(0);
2560
2561                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2562                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2563                                      LMA_OLD_SIZE);
2564                 if (rc <= 0)
2565                         RETURN(-EINVAL);
2566
2567                 lustre_lma_swab(lma);
2568                 if (!(~lma->lma_compat & compat) &&
2569                     !(~lma->lma_incompat & incompat))
2570                         RETURN(0);
2571
2572                 lma->lma_compat |= compat;
2573                 lma->lma_incompat |= incompat;
2574                 lustre_lma_swab(lma);
2575                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2576                                      sizeof(*lma), XATTR_REPLACE);
2577         }
2578
2579         RETURN(rc);
2580 }
2581
2582 /**
2583  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2584  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2585  * To have compatilibility with 1.8 ldiskfs driver we need to have
2586  * magic number at start of fid data.
2587  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2588  * its inmemory API.
2589  */
2590 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2591                                          const struct lu_fid *fid)
2592 {
2593         if (!fid_is_namespace_visible(fid) ||
2594             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2595                 param->edp_magic = 0;
2596                 return;
2597         }
2598
2599         param->edp_magic = LDISKFS_LUFID_MAGIC;
2600         param->edp_len =  sizeof(struct lu_fid) + 1;
2601         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2602 }
2603
2604 /**
2605  * Try to read the fid from inode ea into dt_rec.
2606  *
2607  * \param fid object fid.
2608  *
2609  * \retval 0 on success
2610  */
2611 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2612                           __u32 ino, struct lu_fid *fid,
2613                           struct osd_inode_id *id)
2614 {
2615         struct osd_thread_info *info  = osd_oti_get(env);
2616         struct inode           *inode;
2617         ENTRY;
2618
2619         osd_id_gen(id, ino, OSD_OII_NOGEN);
2620         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2621         if (IS_ERR(inode))
2622                 RETURN(PTR_ERR(inode));
2623
2624         iput(inode);
2625         RETURN(0);
2626 }
2627
2628 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2629                                         struct inode *dir,
2630                                         struct inode *parent_dir,
2631                                         const struct lu_fid *dot_fid,
2632                                         const struct lu_fid *dot_dot_fid,
2633                                         struct osd_thandle *oth)
2634 {
2635         struct ldiskfs_dentry_param *dot_ldp;
2636         struct ldiskfs_dentry_param *dot_dot_ldp;
2637
2638         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2639         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2640
2641         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2642         dot_ldp->edp_magic = 0;
2643         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2644                                         dir, dot_ldp, dot_dot_ldp);
2645 }
2646
2647 /**
2648  * Create an local agent inode for remote entry
2649  */
2650 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2651                                                   struct osd_device *osd,
2652                                                   struct osd_object *pobj,
2653                                                   const struct lu_fid *fid,
2654                                                   __u32 type,
2655                                                   struct thandle *th)
2656 {
2657         struct osd_thread_info  *info = osd_oti_get(env);
2658         struct inode            *local;
2659         struct osd_thandle      *oh;
2660         int                     rc;
2661         ENTRY;
2662
2663         LASSERT(th);
2664         oh = container_of(th, struct osd_thandle, ot_super);
2665         LASSERT(oh->ot_handle->h_transaction != NULL);
2666
2667         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2668         if (IS_ERR(local)) {
2669                 CERROR("%s: create local error %d\n", osd_name(osd),
2670                        (int)PTR_ERR(local));
2671                 RETURN(local);
2672         }
2673
2674         /* Set special LMA flag for local agent inode */
2675         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2676         if (rc != 0) {
2677                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2678                        osd_name(osd), PFID(fid), rc);
2679                 RETURN(ERR_PTR(rc));
2680         }
2681
2682         if (!S_ISDIR(type))
2683                 RETURN(local);
2684
2685         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2686                                          lu_object_fid(&pobj->oo_dt.do_lu),
2687                                          fid, oh);
2688         if (rc != 0) {
2689                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2690                         osd_name(osd), PFID(fid), rc);
2691                 RETURN(ERR_PTR(rc));
2692         }
2693
2694         RETURN(local);
2695 }
2696
2697 /**
2698  * Delete local agent inode for remote entry
2699  */
2700 static int osd_delete_local_agent_inode(const struct lu_env *env,
2701                                         struct osd_device *osd,
2702                                         const struct lu_fid *fid,
2703                                         __u32 ino, struct osd_thandle *oh)
2704 {
2705         struct osd_thread_info  *oti = osd_oti_get(env);
2706         struct osd_inode_id     *id = &oti->oti_id;
2707         struct inode            *inode;
2708         ENTRY;
2709
2710         id->oii_ino = le32_to_cpu(ino);
2711         id->oii_gen = OSD_OII_NOGEN;
2712         inode = osd_iget(oti, osd, id);
2713         if (IS_ERR(inode)) {
2714                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2715                        PFID(fid), id->oii_ino, id->oii_gen);
2716                 RETURN(PTR_ERR(inode));
2717         }
2718
2719         clear_nlink(inode);
2720         mark_inode_dirty(inode);
2721         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2722                 osd_name(osd), PFID(fid), inode->i_ino);
2723         iput(inode);
2724         RETURN(0);
2725 }
2726
2727 /**
2728  * OSD layer object create function for interoperability mode (b11826).
2729  * This is mostly similar to osd_object_create(). Only difference being, fid is
2730  * inserted into inode ea here.
2731  *
2732  * \retval   0, on success
2733  * \retval -ve, on error
2734  */
2735 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2736                                 struct lu_attr *attr,
2737                                 struct dt_allocation_hint *hint,
2738                                 struct dt_object_format *dof,
2739                                 struct thandle *th)
2740 {
2741         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2742         struct osd_object       *obj    = osd_dt_obj(dt);
2743         struct osd_thread_info  *info   = osd_oti_get(env);
2744         int                      result;
2745
2746         ENTRY;
2747
2748         LASSERT(osd_invariant(obj));
2749         LASSERT(!dt_object_exists(dt));
2750         LASSERT(!dt_object_remote(dt));
2751         LASSERT(osd_write_locked(env, obj));
2752         LASSERT(th != NULL);
2753
2754         if (unlikely(fid_is_acct(fid)))
2755                 /* Quota files can't be created from the kernel any more,
2756                  * 'tune2fs -O quota' will take care of creating them */
2757                 RETURN(-EPERM);
2758
2759         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2760         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2761
2762         result = __osd_object_create(info, obj, attr, hint, dof, th);
2763         if (result == 0)
2764                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2765                                 fid_is_on_ost(info, osd_obj2dev(obj),
2766                                               fid, OI_CHECK_FLD) ?
2767                                 LMAC_FID_ON_OST : 0, 0);
2768
2769         if (result == 0)
2770                 result = __osd_oi_insert(env, obj, fid, th);
2771
2772         LASSERT(ergo(result == 0,
2773                      dt_object_exists(dt) && !dt_object_remote(dt)));
2774         LINVRNT(osd_invariant(obj));
2775         RETURN(result);
2776 }
2777
2778 static int osd_declare_object_ref_add(const struct lu_env *env,
2779                                       struct dt_object *dt,
2780                                       struct thandle *handle)
2781 {
2782         struct osd_thandle       *oh;
2783
2784         /* it's possible that object doesn't exist yet */
2785         LASSERT(handle != NULL);
2786
2787         oh = container_of0(handle, struct osd_thandle, ot_super);
2788         LASSERT(oh->ot_handle == NULL);
2789
2790         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2791                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2792
2793         return 0;
2794 }
2795
2796 /*
2797  * Concurrency: @dt is write locked.
2798  */
2799 static int osd_object_ref_add(const struct lu_env *env,
2800                               struct dt_object *dt, struct thandle *th)
2801 {
2802         struct osd_object  *obj = osd_dt_obj(dt);
2803         struct inode       *inode = obj->oo_inode;
2804         struct osd_thandle *oh;
2805         int                 rc = 0;
2806
2807         LINVRNT(osd_invariant(obj));
2808         LASSERT(dt_object_exists(dt));
2809         LASSERT(!dt_object_remote(dt));
2810         LASSERT(osd_write_locked(env, obj));
2811         LASSERT(th != NULL);
2812
2813         oh = container_of0(th, struct osd_thandle, ot_super);
2814         LASSERT(oh->ot_handle != NULL);
2815
2816         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2817
2818         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2819                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2820         /*
2821          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2822          * (65000) subdirectories by storing "1" in i_nlink if the link count
2823          * would otherwise overflow. Directory tranversal tools understand
2824          * that (st_nlink == 1) indicates that the filesystem dose not track
2825          * hard links count on the directory, and will not abort subdirectory
2826          * scanning early once (st_nlink - 2) subdirs have been found.
2827          *
2828          * This also has to properly handle the case of inodes with nlink == 0
2829          * in case they are being linked into the PENDING directory
2830          */
2831         spin_lock(&obj->oo_guard);
2832         if (unlikely(inode->i_nlink == 0))
2833                 /* inc_nlink from 0 may cause WARN_ON */
2834                 set_nlink(inode, 1);
2835         else {
2836                 ldiskfs_inc_count(oh->ot_handle, inode);
2837                 if (!S_ISDIR(inode->i_mode))
2838                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2839         }
2840         spin_unlock(&obj->oo_guard);
2841
2842         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2843         LINVRNT(osd_invariant(obj));
2844
2845         return rc;
2846 }
2847
2848 static int osd_declare_object_ref_del(const struct lu_env *env,
2849                                       struct dt_object *dt,
2850                                       struct thandle *handle)
2851 {
2852         struct osd_thandle *oh;
2853
2854         LASSERT(dt_object_exists(dt));
2855         LASSERT(!dt_object_remote(dt));
2856         LASSERT(handle != NULL);
2857
2858         oh = container_of0(handle, struct osd_thandle, ot_super);
2859         LASSERT(oh->ot_handle == NULL);
2860
2861         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2862                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2863
2864         return 0;
2865 }
2866
2867 /*
2868  * Concurrency: @dt is write locked.
2869  */
2870 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2871                               struct thandle *th)
2872 {
2873         struct osd_object       *obj = osd_dt_obj(dt);
2874         struct inode            *inode = obj->oo_inode;
2875         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2876         struct osd_thandle      *oh;
2877
2878         LINVRNT(osd_invariant(obj));
2879         LASSERT(dt_object_exists(dt));
2880         LASSERT(!dt_object_remote(dt));
2881         LASSERT(osd_write_locked(env, obj));
2882         LASSERT(th != NULL);
2883
2884         oh = container_of0(th, struct osd_thandle, ot_super);
2885         LASSERT(oh->ot_handle != NULL);
2886
2887         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2888
2889         spin_lock(&obj->oo_guard);
2890         /* That can be result of upgrade from old Lustre version and
2891          * applied only to local files.  Just skip this ref_del call.
2892          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2893         if (inode->i_nlink == 0) {
2894                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2895                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2896                              ", maybe an upgraded file? (LU-3915)\n",
2897                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2898                 spin_unlock(&obj->oo_guard);
2899                 return 0;
2900         }
2901
2902         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2903                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2904
2905         ldiskfs_dec_count(oh->ot_handle, inode);
2906         spin_unlock(&obj->oo_guard);
2907
2908         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2909         LINVRNT(osd_invariant(obj));
2910
2911         return 0;
2912 }
2913
2914 /*
2915  * Get the 64-bit version for an inode.
2916  */
2917 static int osd_object_version_get(const struct lu_env *env,
2918                                   struct dt_object *dt, dt_obj_version_t *ver)
2919 {
2920         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2921
2922         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2923                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2924         *ver = LDISKFS_I(inode)->i_fs_version;
2925         return 0;
2926 }
2927
2928 /*
2929  * Concurrency: @dt is read locked.
2930  */
2931 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2932                          struct lu_buf *buf, const char *name,
2933                          struct lustre_capa *capa)
2934 {
2935         struct osd_object      *obj    = osd_dt_obj(dt);
2936         struct inode           *inode  = obj->oo_inode;
2937         struct osd_thread_info *info   = osd_oti_get(env);
2938         struct dentry          *dentry = &info->oti_obj_dentry;
2939
2940         /* version get is not real XATTR but uses xattr API */
2941         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2942                 /* for version we are just using xattr API but change inode
2943                  * field instead */
2944                 if (buf->lb_len == 0)
2945                         return sizeof(dt_obj_version_t);
2946
2947                 if (buf->lb_len < sizeof(dt_obj_version_t))
2948                         return -ERANGE;
2949
2950                 osd_object_version_get(env, dt, buf->lb_buf);
2951
2952                 return sizeof(dt_obj_version_t);
2953         }
2954
2955         LASSERT(dt_object_exists(dt));
2956         LASSERT(!dt_object_remote(dt));
2957         LASSERT(inode->i_op != NULL);
2958         LASSERT(inode->i_op->getxattr != NULL);
2959
2960         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2961                 return -EACCES;
2962
2963         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2964 }
2965
2966
2967 static int osd_declare_xattr_set(const struct lu_env *env,
2968                                  struct dt_object *dt,
2969                                  const struct lu_buf *buf, const char *name,
2970                                  int fl, struct thandle *handle)
2971 {
2972         struct osd_thandle *oh;
2973         int credits;
2974         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2975
2976         LASSERT(handle != NULL);
2977
2978         oh = container_of0(handle, struct osd_thandle, ot_super);
2979         LASSERT(oh->ot_handle == NULL);
2980
2981         /* optimistic optimization: LMA is set first and usually fit inode */
2982         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2983                 if (dt_object_exists(dt))
2984                         credits = 0;
2985                 else
2986                         credits = 1;
2987         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2988                 credits = 1;
2989         } else {
2990                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2991                 if (buf && buf->lb_len > sb->s_blocksize) {
2992                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2993                                         sb->s_blocksize_bits;
2994                 }
2995                 /*
2996                  * xattr set may involve inode quota change, reserve credits for
2997                  * dquot_initialize()
2998                  */
2999                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3000         }
3001
3002         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3003
3004         return 0;
3005 }
3006
3007 /*
3008  * Set the 64-bit version for object
3009  */
3010 static void osd_object_version_set(const struct lu_env *env,
3011                                    struct dt_object *dt,
3012                                    dt_obj_version_t *new_version)
3013 {
3014         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3015
3016         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3017                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3018
3019         LDISKFS_I(inode)->i_fs_version = *new_version;
3020         /** Version is set after all inode operations are finished,
3021          *  so we should mark it dirty here */
3022         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3023 }
3024
3025 /*
3026  * Concurrency: @dt is write locked.
3027  */
3028 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3029                          const struct lu_buf *buf, const char *name, int fl,
3030                          struct thandle *handle, struct lustre_capa *capa)
3031 {
3032         struct osd_object      *obj      = osd_dt_obj(dt);
3033         struct inode           *inode    = obj->oo_inode;
3034         struct osd_thread_info *info     = osd_oti_get(env);
3035         int                     fs_flags = 0;
3036         ENTRY;
3037
3038         LASSERT(handle != NULL);
3039
3040         /* version set is not real XATTR */
3041         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3042                 /* for version we are just using xattr API but change inode
3043                  * field instead */
3044                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3045                 osd_object_version_set(env, dt, buf->lb_buf);
3046                 return sizeof(dt_obj_version_t);
3047         }
3048
3049         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3050                 return -EACCES;
3051
3052         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3053                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3054
3055         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3056         if (fl & LU_XATTR_REPLACE)
3057                 fs_flags |= XATTR_REPLACE;
3058
3059         if (fl & LU_XATTR_CREATE)
3060                 fs_flags |= XATTR_CREATE;
3061
3062         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3063                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3064                 int                      rc;
3065
3066                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3067                 if (rc != 0)
3068                         RETURN(rc);
3069
3070                 lma->lma_incompat |= LMAI_STRIPED;
3071                 lustre_lma_swab(lma);
3072                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3073                                      sizeof(*lma), XATTR_REPLACE);
3074                 if (rc != 0)
3075                         RETURN(rc);
3076         }
3077
3078         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3079             strcmp(name, XATTR_NAME_LINK) == 0)
3080                 return -ENOSPC;
3081
3082         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3083                                fs_flags);
3084 }
3085
3086 /*
3087  * Concurrency: @dt is read locked.
3088  */
3089 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3090                           struct lu_buf *buf, struct lustre_capa *capa)
3091 {
3092         struct osd_object      *obj    = osd_dt_obj(dt);
3093         struct inode           *inode  = obj->oo_inode;
3094         struct osd_thread_info *info   = osd_oti_get(env);
3095         struct dentry          *dentry = &info->oti_obj_dentry;
3096
3097         LASSERT(dt_object_exists(dt));
3098         LASSERT(!dt_object_remote(dt));
3099         LASSERT(inode->i_op != NULL);
3100         LASSERT(inode->i_op->listxattr != NULL);
3101
3102         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3103                 return -EACCES;
3104
3105         dentry->d_inode = inode;
3106         dentry->d_sb = inode->i_sb;
3107         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3108 }
3109
3110 static int osd_declare_xattr_del(const struct lu_env *env,
3111                                  struct dt_object *dt, const char *name,
3112                                  struct thandle *handle)
3113 {
3114         struct osd_thandle *oh;
3115         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3116
3117         LASSERT(dt_object_exists(dt));
3118         LASSERT(!dt_object_remote(dt));
3119         LASSERT(handle != NULL);
3120
3121         oh = container_of0(handle, struct osd_thandle, ot_super);
3122         LASSERT(oh->ot_handle == NULL);
3123
3124         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3125                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3126         /*
3127          * xattr del may involve inode quota change, reserve credits for
3128          * dquot_initialize()
3129          */
3130         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3131
3132         return 0;
3133 }
3134
3135 /*
3136  * Concurrency: @dt is write locked.
3137  */
3138 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3139                          const char *name, struct thandle *handle,
3140                          struct lustre_capa *capa)
3141 {
3142         struct osd_object      *obj    = osd_dt_obj(dt);
3143         struct inode           *inode  = obj->oo_inode;
3144         struct osd_thread_info *info   = osd_oti_get(env);
3145         struct dentry          *dentry = &info->oti_obj_dentry;
3146         int                     rc;
3147
3148         LASSERT(dt_object_exists(dt));
3149         LASSERT(!dt_object_remote(dt));
3150         LASSERT(inode->i_op != NULL);
3151         LASSERT(inode->i_op->removexattr != NULL);
3152         LASSERT(handle != NULL);
3153
3154         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3155                 return -EACCES;
3156
3157         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3158
3159         ll_vfs_dq_init(inode);
3160         dentry->d_inode = inode;
3161         dentry->d_sb = inode->i_sb;
3162         rc = inode->i_op->removexattr(dentry, name);
3163         return rc;
3164 }
3165
3166 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3167                                      struct dt_object *dt,
3168                                      struct lustre_capa *old, __u64 opc)
3169 {
3170         struct osd_thread_info *info = osd_oti_get(env);
3171         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3172         struct osd_object *obj = osd_dt_obj(d