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_MDS
  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
  77 int ldiskfs_pdo = 1;
  78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  79                 "ldiskfs with parallel directory operations");
  80
  81 int ldiskfs_track_declares_assert;
  82 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  83                 "LBUG during tracking of declares");
  84
  85 /* Slab to allocate dynlocks */
  86 struct kmem_cache *dynlock_cachep;
  87
  88 static struct lu_kmem_descr ldiskfs_caches[] = {
  89         {
  90                 .ckd_cache = &dynlock_cachep,
  91                 .ckd_name  = "dynlock_cache",
  92                 .ckd_size  = sizeof(struct dynlock_handle)
  93         },
  94         {
  95                 .ckd_cache = NULL
  96         }
  97 };
  98
  99 static const char dot[] = ".";
 100 static const char dotdot[] = "..";
 101 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 102
 103 static const struct lu_object_operations      osd_lu_obj_ops;
 104 static const struct dt_object_operations      osd_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ea_ops;
 106 static const struct dt_object_operations      osd_obj_otable_it_ops;
 107 static const struct dt_index_operations       osd_index_iam_ops;
 108 static const struct dt_index_operations       osd_index_ea_ops;
 109
 110 int osd_trans_declare_op2rb[] = {
 111         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 112         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 113         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 114         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 115         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 116         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 117         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 118         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 119         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 120         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 121         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 122         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 123 };
 124
 125 static int osd_has_index(const struct osd_object *obj)
 126 {
 127         return obj->oo_dt.do_index_ops != NULL;
 128 }
 129
 130 static int osd_object_invariant(const struct lu_object *l)
 131 {
 132         return osd_invariant(osd_obj(l));
 133 }
 134
 135 /*
 136  * Concurrency: doesn't matter
 137  */
 138
 139 /*
 140  * Concurrency: doesn't matter
 141  */
 142 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 143 {
 144         struct osd_thread_info *oti = osd_oti_get(env);
 145         return oti->oti_w_locks > 0 && o->oo_owner == env;
 146 }
 147
 148 /*
 149  * Concurrency: doesn't access mutable data
 150  */
 151 static int osd_root_get(const struct lu_env *env,
 152                         struct dt_device *dev, struct lu_fid *f)
 153 {
 154         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 155         return 0;
 156 }
 157
 158 /*
 159  * OSD object methods.
 160  */
 161
 162 /*
 163  * Concurrency: no concurrent access is possible that early in object
 164  * life-cycle.
 165  */
 166 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 167                                           const struct lu_object_header *hdr,
 168                                           struct lu_device *d)
 169 {
 170         struct osd_object *mo;
 171
 172         OBD_ALLOC_PTR(mo);
 173         if (mo != NULL) {
 174                 struct lu_object *l;
 175
 176                 l = &mo->oo_dt.do_lu;
 177                 dt_object_init(&mo->oo_dt, NULL, d);
 178                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 179                 l->lo_ops = &osd_lu_obj_ops;
 180                 init_rwsem(&mo->oo_sem);
 181                 init_rwsem(&mo->oo_ext_idx_sem);
 182                 spin_lock_init(&mo->oo_guard);
 183                 return l;
 184         } else {
 185                 return NULL;
 186         }
 187 }
 188
 189 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 190                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 191 {
 192         int rc;
 193
 194         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 195         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 196                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 197         if (rc > 0) {
 198                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 199                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 200                 rc = 0;
 201                 lustre_lma_swab(lma);
 202                 /* Check LMA compatibility */
 203                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 204                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 205                               "for fid = "DFID", ino = %lu\n",
 206                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 207                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 208                               PFID(&lma->lma_self_fid), inode->i_ino);
 209                         rc = -EOPNOTSUPP;
 210                 }
 211         } else if (rc == 0) {
 212                 rc = -ENODATA;
 213         }
 214
 215         return rc;
 216 }
 217
 218 /*
 219  * retrieve object from backend ext fs.
 220  **/
 221 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 222                        struct osd_inode_id *id)
 223 {
 224         struct inode *inode = NULL;
 225
 226         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 227         if (IS_ERR(inode)) {
 228                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 229                        id->oii_ino, PTR_ERR(inode));
 230         } else if (id->oii_gen != OSD_OII_NOGEN &&
 231                    inode->i_generation != id->oii_gen) {
 232                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 233                        "i_generation = %u\n",
 234                        id->oii_ino, id->oii_gen, inode->i_generation);
 235                 iput(inode);
 236                 inode = ERR_PTR(-ESTALE);
 237         } else if (inode->i_nlink == 0) {
 238                 /* due to parallel readdir and unlink,
 239                 * we can have dead inode here. */
 240                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 241                 iput(inode);
 242                 inode = ERR_PTR(-ESTALE);
 243         } else if (is_bad_inode(inode)) {
 244                 CWARN("%.16s: bad inode: ino = %u\n",
 245                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 246                 iput(inode);
 247                 inode = ERR_PTR(-ENOENT);
 248         } else {
 249                 if (id->oii_gen == OSD_OII_NOGEN)
 250                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 251
 252                 /* Do not update file c/mtime in ldiskfs.
 253                  * NB: we don't have any lock to protect this because we don't
 254                  * have reference on osd_object now, but contention with
 255                  * another lookup + attr_set can't happen in the tiny window
 256                  * between if (...) and set S_NOCMTIME. */
 257                 if (!(inode->i_flags & S_NOCMTIME))
 258                         inode->i_flags |= S_NOCMTIME;
 259         }
 260         return inode;
 261 }
 262
 263 static struct inode *
 264 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 265              struct osd_inode_id *id, struct lu_fid *fid)
 266 {
 267         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 268         struct inode            *inode;
 269         int                      rc;
 270
 271         inode = osd_iget(info, dev, id);
 272         if (IS_ERR(inode))
 273                 return inode;
 274
 275         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 276         if (rc == 0) {
 277                 *fid = lma->lma_self_fid;
 278         } else if (rc == -ENODATA) {
 279                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 280                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 281                 else
 282                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 283         } else {
 284                 iput(inode);
 285                 inode = ERR_PTR(rc);
 286         }
 287         return inode;
 288 }
 289
 290 static struct inode *osd_iget_check(struct osd_thread_info *info,
 291                                     struct osd_device *dev,
 292                                     const struct lu_fid *fid,
 293                                     struct osd_inode_id *id,
 294                                     bool in_oi)
 295 {
 296         struct inode    *inode;
 297         int              rc     = 0;
 298         ENTRY;
 299
 300         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 301         if (IS_ERR(inode)) {
 302                 rc = PTR_ERR(inode);
 303                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 304                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 305                                id->oii_ino, rc);
 306
 307                         GOTO(put, rc);
 308                 }
 309
 310                 goto check_oi;
 311         }
 312
 313         if (is_bad_inode(inode)) {
 314                 rc = -ENOENT;
 315                 if (!in_oi) {
 316                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 317
 318                         GOTO(put, rc);
 319                 }
 320
 321                 goto check_oi;
 322         }
 323
 324         if (id->oii_gen != OSD_OII_NOGEN &&
 325             inode->i_generation != id->oii_gen) {
 326                 rc = -ESTALE;
 327                 if (!in_oi) {
 328                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 329                                "oii_gen = %u, i_generation = %u\n",
 330                                id->oii_ino, id->oii_gen, inode->i_generation);
 331
 332                         GOTO(put, rc);
 333                 }
 334
 335                 goto check_oi;
 336         }
 337
 338         if (inode->i_nlink == 0) {
 339                 rc = -ENOENT;
 340                 if (!in_oi) {
 341                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 342
 343                         GOTO(put, rc);
 344                 }
 345
 346                 goto check_oi;
 347         }
 348
 349 check_oi:
 350         if (rc != 0) {
 351                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 352
 353                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 354                 /* XXX: There are some possible cases:
 355                  *      1. rc = 0.
 356                  *         Backup/restore caused the OI invalid.
 357                  *      2. rc = 0.
 358                  *         Someone unlinked the object but NOT removed
 359                  *         the OI mapping, such as mount target device
 360                  *         as ldiskfs, and modify something directly.
 361                  *      3. rc = -ENOENT.
 362                  *         Someone just removed the object between the
 363                  *         former oi_lookup and the iget. It is normal.
 364                  *      4. Other failure cases.
 365                  *
 366                  *      Generally, when the device is mounted, it will
 367                  *      auto check whether the system is restored from
 368                  *      file-level backup or not. We trust such detect
 369                  *      to distinguish the 1st case from the 2nd case. */
 370                 if (rc == 0) {
 371                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 372                             inode->i_generation == id->oii_gen)
 373                                 rc = -ENOENT;
 374                         else
 375                                 rc = -EREMCHG;
 376                 }
 377         } else {
 378                 if (id->oii_gen == OSD_OII_NOGEN)
 379                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 380
 381                 /* Do not update file c/mtime in ldiskfs.
 382                  * NB: we don't have any lock to protect this because we don't
 383                  * have reference on osd_object now, but contention with
 384                  * another lookup + attr_set can't happen in the tiny window
 385                  * between if (...) and set S_NOCMTIME. */
 386                 if (!(inode->i_flags & S_NOCMTIME))
 387                         inode->i_flags |= S_NOCMTIME;
 388         }
 389
 390         GOTO(put, rc);
 391
 392 put:
 393         if (rc != 0) {
 394                 if (!IS_ERR(inode))
 395                         iput(inode);
 396
 397                 inode = ERR_PTR(rc);
 398         }
 399
 400         return inode;
 401 }
 402
 403 /**
 404  * \retval +v: new filter_fid, does not contain self-fid
 405  * \retval 0:  filter_fid_old, contains self-fid
 406  * \retval -v: other failure cases
 407  */
 408 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 409                  struct dentry *dentry, struct lu_fid *fid)
 410 {
 411         struct filter_fid_old   *ff     = &info->oti_ff;
 412         struct ost_id           *ostid  = &info->oti_ostid;
 413         int                      rc;
 414
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 rc = 0;
 418                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 419                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 420                 /* XXX: should use real OST index in the future. LU-3569 */
 421                 ostid_to_fid(fid, ostid, 0);
 422         } else if (rc == sizeof(struct filter_fid)) {
 423                 rc = 1;
 424         } else if (rc >= 0) {
 425                 rc = -EINVAL;
 426         }
 427
 428         return rc;
 429 }
 430
 431 static int osd_lma_self_repair(struct osd_thread_info *info,
 432                                struct osd_device *osd, struct inode *inode,
 433                                const struct lu_fid *fid, __u32 compat)
 434 {
 435         handle_t *jh;
 436         int       rc;
 437
 438         LASSERT(current->journal_info == NULL);
 439
 440         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 441                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 442         if (IS_ERR(jh)) {
 443                 rc = PTR_ERR(jh);
 444                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 445                       osd_name(osd), rc);
 446                 return rc;
 447         }
 448
 449         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 450         if (rc != 0)
 451                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 452                       osd_name(osd), rc);
 453         ldiskfs_journal_stop(jh);
 454         return rc;
 455 }
 456
 457 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 458 {
 459         struct osd_thread_info  *info   = osd_oti_get(env);
 460         struct osd_device       *osd    = osd_obj2dev(obj);
 461         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 462         struct inode            *inode  = obj->oo_inode;
 463         struct dentry           *dentry = &info->oti_obj_dentry;
 464         struct lu_fid           *fid    = NULL;
 465         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 466         int                      rc;
 467         ENTRY;
 468
 469         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 470         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 471                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 472         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 473                 fid = &lma->lma_self_fid;
 474                 rc = osd_get_idif(info, inode, dentry, fid);
 475                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 476                         /* For the given OST-object, if it has neither LMA nor
 477                          * FID in XATTR_NAME_FID, then the given FID (which is
 478                          * contained in the @obj, from client RPC for locating
 479                          * the OST-object) is trusted. We use it to generate
 480                          * the LMA. */
 481                         osd_lma_self_repair(info, osd, inode, rfid,
 482                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 483                                 LMAC_FID_ON_OST : 0);
 484                         RETURN(0);
 485                 }
 486         }
 487
 488         if (unlikely(rc == -ENODATA))
 489                 RETURN(0);
 490
 491         if (rc < 0)
 492                 RETURN(rc);
 493
 494         if (rc > 0) {
 495                 rc = 0;
 496                 lustre_lma_swab(lma);
 497                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 498                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 499                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 500                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 501                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 502                               PFID(rfid), inode->i_ino);
 503                         rc = -EOPNOTSUPP;
 504                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 505                         fid = &lma->lma_self_fid;
 506                 }
 507         }
 508
 509         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 510                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 511                         struct ost_id   *oi   = &info->oti_ostid;
 512                         struct lu_fid   *fid1 = &info->oti_fid3;
 513                         __u32            idx  = fid_idif_ost_idx(rfid);
 514
 515                         /* For old IDIF, the OST index is not part of the IDIF,
 516                          * Means that different OSTs may have the same IDIFs.
 517                          * Under such case, we need to make some compatible
 518                          * check to make sure to trigger OI scrub properly. */
 519                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 520                                 /* Given @rfid is new, LMA is old. */
 521                                 fid_to_ostid(fid, oi);
 522                                 ostid_to_fid(fid1, oi, idx);
 523                                 if (lu_fid_eq(fid1, rfid)) {
 524                                         if (osd->od_lma_self_repair)
 525                                                 osd_lma_self_repair(info, osd,
 526                                                         inode, rfid,
 527                                                         LMAC_FID_ON_OST);
 528                                         RETURN(0);
 529                                 }
 530                         }
 531                 }
 532
 533                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 534                        osd_name(osd), PFID(rfid), PFID(fid));
 535                 rc = -EREMCHG;
 536         }
 537
 538         RETURN(rc);
 539 }
 540
 541 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 542                           const struct lu_fid *fid,
 543                           const struct lu_object_conf *conf)
 544 {
 545         struct osd_thread_info *info;
 546         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 547         struct osd_device      *dev;
 548         struct osd_idmap_cache *oic;
 549         struct osd_inode_id    *id;
 550         struct inode           *inode;
 551         struct osd_scrub       *scrub;
 552         struct scrub_file      *sf;
 553         int                     result;
 554         int                     saved  = 0;
 555         bool                    in_oi  = false;
 556         bool                    triggered = false;
 557         ENTRY;
 558
 559         LINVRNT(osd_invariant(obj));
 560         LASSERT(obj->oo_inode == NULL);
 561         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 562
 563         dev = osd_dev(ldev);
 564         scrub = &dev->od_scrub;
 565         sf = &scrub->os_file;
 566         info = osd_oti_get(env);
 567         LASSERT(info);
 568         oic = &info->oti_cache;
 569
 570         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 571                 RETURN(-ENOENT);
 572
 573         /* For the object is created as locking anchor, or for the object to
 574          * be created on disk. No need to osd_oi_lookup() at here because FID
 575          * shouldn't never be re-used, if it's really a duplicate FID from
 576          * unexpected reason, we should be able to detect it later by calling
 577          * do_create->osd_oi_insert(). */
 578         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 579                 GOTO(out, result = 0);
 580
 581         /* Search order: 1. per-thread cache. */
 582         if (lu_fid_eq(fid, &oic->oic_fid) &&
 583             likely(oic->oic_dev == dev)) {
 584                 id = &oic->oic_lid;
 585                 goto iget;
 586         }
 587
 588         id = &info->oti_id;
 589         if (!list_empty(&scrub->os_inconsistent_items)) {
 590                 /* Search order: 2. OI scrub pending list. */
 591                 result = osd_oii_lookup(dev, fid, id);
 592                 if (result == 0)
 593                         goto iget;
 594         }
 595
 596         /* Search order: 3. OI files. */
 597         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 598         if (result == -ENOENT) {
 599                 if (!fid_is_norm(fid) ||
 600                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 601                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 602                                       sf->sf_oi_bitmap))
 603                         GOTO(out, result = 0);
 604
 605                 goto trigger;
 606         }
 607
 608         if (result != 0)
 609                 GOTO(out, result);
 610
 611         in_oi = true;
 612
 613 iget:
 614         inode = osd_iget_check(info, dev, fid, id, in_oi);
 615         if (IS_ERR(inode)) {
 616                 result = PTR_ERR(inode);
 617                 if (result == -ENOENT || result == -ESTALE) {
 618                         if (!in_oi)
 619                                 fid_zero(&oic->oic_fid);
 620
 621                         GOTO(out, result = -ENOENT);
 622                 } else if (result == -EREMCHG) {
 623
 624 trigger:
 625                         if (!in_oi)
 626                                 fid_zero(&oic->oic_fid);
 627
 628                         if (unlikely(triggered))
 629                                 GOTO(out, result = saved);
 630
 631                         triggered = true;
 632                         if (thread_is_running(&scrub->os_thread)) {
 633                                 result = -EINPROGRESS;
 634                         } else if (!dev->od_noscrub) {
 635                                 result = osd_scrub_start(dev);
 636                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 637                                               "for "DFID", rc = %d [1]\n",
 638                                               osd_name(dev), PFID(fid),result);
 639                                 if (result == 0 || result == -EALREADY)
 640                                         result = -EINPROGRESS;
 641                                 else
 642                                         result = -EREMCHG;
 643                         }
 644
 645                         /* We still have chance to get the valid inode: for the
 646                          * object which is referenced by remote name entry, the
 647                          * object on the local MDT will be linked under the dir
 648                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 649                          *
 650                          * We do not know whether the object for the given FID
 651                          * is referenced by some remote name entry or not, and
 652                          * especially for DNE II, a multiple-linked object may
 653                          * have many name entries reside on many MDTs.
 654                          *
 655                          * To simplify the operation, OSD will not distinguish
 656                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 657                          * only happened for the RPC from other MDT during the
 658                          * OI scrub, or for the client side RPC with FID only,
 659                          * such as FID to path, or from old connected client. */
 660                         saved = result;
 661                         result = osd_lookup_in_remote_parent(info, dev,
 662                                                              fid, id);
 663                         if (result == 0) {
 664                                 in_oi = false;
 665                                 goto iget;
 666                         }
 667
 668                         result = saved;
 669                 }
 670
 671                 GOTO(out, result);
 672         }
 673
 674         obj->oo_inode = inode;
 675         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 676
 677         result = osd_check_lma(env, obj);
 678         if (result != 0) {
 679                 iput(inode);
 680                 obj->oo_inode = NULL;
 681                 if (result == -EREMCHG) {
 682                         if (!in_oi) {
 683                                 result = osd_oi_lookup(info, dev, fid, id,
 684                                                        OI_CHECK_FLD);
 685                                 if (result != 0) {
 686                                         fid_zero(&oic->oic_fid);
 687                                         GOTO(out, result);
 688                                 }
 689                         }
 690
 691                         goto trigger;
 692                 }
 693
 694                 GOTO(out, result);
 695         }
 696
 697         obj->oo_compat_dot_created = 1;
 698         obj->oo_compat_dotdot_created = 1;
 699
 700         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 701                 GOTO(out, result = 0);
 702
 703         LASSERT(obj->oo_hl_head == NULL);
 704         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 705         if (obj->oo_hl_head == NULL) {
 706                 obj->oo_inode = NULL;
 707                 iput(inode);
 708                 GOTO(out, result = -ENOMEM);
 709         }
 710         GOTO(out, result = 0);
 711
 712 out:
 713         LINVRNT(osd_invariant(obj));
 714         return result;
 715 }
 716
 717 /*
 718  * Concurrency: shouldn't matter.
 719  */
 720 static void osd_object_init0(struct osd_object *obj)
 721 {
 722         LASSERT(obj->oo_inode != NULL);
 723         obj->oo_dt.do_body_ops = &osd_body_ops;
 724         obj->oo_dt.do_lu.lo_header->loh_attr |=
 725                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 726 }
 727
 728 /*
 729  * Concurrency: no concurrent access is possible that early in object
 730  * life-cycle.
 731  */
 732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 733                            const struct lu_object_conf *conf)
 734 {
 735         struct osd_object *obj = osd_obj(l);
 736         int result;
 737
 738         LINVRNT(osd_invariant(obj));
 739
 740         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 741                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 742                 l->lo_header->loh_attr |= LOHA_EXISTS;
 743                 return 0;
 744         }
 745
 746         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 747         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 748         if (result == 0 && obj->oo_inode != NULL)
 749                 osd_object_init0(obj);
 750
 751         LINVRNT(osd_invariant(obj));
 752         return result;
 753 }
 754
 755 /*
 756  * Concurrency: no concurrent access is possible that late in object
 757  * life-cycle.
 758  */
 759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 760 {
 761         struct osd_object *obj = osd_obj(l);
 762
 763         LINVRNT(osd_invariant(obj));
 764
 765         dt_object_fini(&obj->oo_dt);
 766         if (obj->oo_hl_head != NULL)
 767                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 768         OBD_FREE_PTR(obj);
 769 }
 770
 771 /*
 772  * Concurrency: no concurrent access is possible that late in object
 773  * life-cycle.
 774  */
 775 static void osd_index_fini(struct osd_object *o)
 776 {
 777         struct iam_container *bag;
 778
 779         if (o->oo_dir != NULL) {
 780                 bag = &o->oo_dir->od_container;
 781                 if (o->oo_inode != NULL) {
 782                         if (bag->ic_object == o->oo_inode)
 783                                 iam_container_fini(bag);
 784                 }
 785                 OBD_FREE_PTR(o->oo_dir);
 786                 o->oo_dir = NULL;
 787         }
 788 }
 789
 790 /*
 791  * Concurrency: no concurrent access is possible that late in object
 792  * life-cycle (for all existing callers, that is. New callers have to provide
 793  * their own locking.)
 794  */
 795 static int osd_inode_unlinked(const struct inode *inode)
 796 {
 797         return inode->i_nlink == 0;
 798 }
 799
 800 enum {
 801         OSD_TXN_OI_DELETE_CREDITS    = 20,
 802         OSD_TXN_INODE_DELETE_CREDITS = 20
 803 };
 804
 805 /*
 806  * Journal
 807  */
 808
 809 #if OSD_THANDLE_STATS
 810 /**
 811  * Set time when the handle is allocated
 812  */
 813 static void osd_th_alloced(struct osd_thandle *oth)
 814 {
 815         oth->oth_alloced = cfs_time_current();
 816 }
 817
 818 /**
 819  * Set time when the handle started
 820  */
 821 static void osd_th_started(struct osd_thandle *oth)
 822 {
 823         oth->oth_started = cfs_time_current();
 824 }
 825
 826 /**
 827  * Helper function to convert time interval to microseconds packed in
 828  * long int.
 829  */
 830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 831 {
 832         struct timeval val;
 833
 834         cfs_duration_usec(cfs_time_sub(end, start), &val);
 835         return val.tv_sec * 1000000 + val.tv_usec;
 836 }
 837
 838 /**
 839  * Check whether the we deal with this handle for too long.
 840  */
 841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 842                                 cfs_time_t alloced, cfs_time_t started,
 843                                 cfs_time_t closed)
 844 {
 845         cfs_time_t now = cfs_time_current();
 846
 847         LASSERT(dev != NULL);
 848
 849         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 850                             interval_to_usec(alloced, started));
 851         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 852                             interval_to_usec(started, closed));
 853         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 854                             interval_to_usec(closed, now));
 855
 856         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 857                 CWARN("transaction handle %p was open for too long: "
 858                       "now "CFS_TIME_T" ,"
 859                       "alloced "CFS_TIME_T" ,"
 860                       "started "CFS_TIME_T" ,"
 861                       "closed "CFS_TIME_T"\n",
 862                       oth, now, alloced, started, closed);
 863                 libcfs_debug_dumpstack(NULL);
 864         }
 865 }
 866
 867 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 868 {                                                                       \
 869         cfs_time_t __closed = cfs_time_current();                       \
 870         cfs_time_t __alloced = oth->oth_alloced;                        \
 871         cfs_time_t __started = oth->oth_started;                        \
 872                                                                         \
 873         expr;                                                           \
 874         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 875 }
 876
 877 #else /* OSD_THANDLE_STATS */
 878
 879 #define osd_th_alloced(h)                  do {} while(0)
 880 #define osd_th_started(h)                  do {} while(0)
 881 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 882
 883 #endif /* OSD_THANDLE_STATS */
 884
 885 /*
 886  * Concurrency: doesn't access mutable data.
 887  */
 888 static int osd_param_is_not_sane(const struct osd_device *dev,
 889                                  const struct thandle *th)
 890 {
 891         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 892
 893         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 894 }
 895
 896 /*
 897  * Concurrency: shouldn't matter.
 898  */
 899 static void osd_trans_commit_cb(struct super_block *sb,
 900                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 901 {
 902         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 903         struct thandle     *th  = &oh->ot_super;
 904         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 905         struct dt_txn_commit_cb *dcb, *tmp;
 906
 907         LASSERT(oh->ot_handle == NULL);
 908
 909         if (error)
 910                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 911
 912         dt_txn_hook_commit(th);
 913
 914         /* call per-transaction callbacks if any */
 915         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 916                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 917                          "commit callback entry: magic=%x name='%s'\n",
 918                          dcb->dcb_magic, dcb->dcb_name);
 919                 list_del_init(&dcb->dcb_linkage);
 920                 dcb->dcb_func(NULL, th, dcb, error);
 921         }
 922
 923         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 924         lu_device_put(lud);
 925         th->th_dev = NULL;
 926
 927         lu_context_exit(&th->th_ctx);
 928         lu_context_fini(&th->th_ctx);
 929         thandle_put(th);
 930 }
 931
 932 static struct thandle *osd_trans_create(const struct lu_env *env,
 933                                         struct dt_device *d)
 934 {
 935         struct osd_thread_info  *oti = osd_oti_get(env);
 936         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 937         struct osd_thandle      *oh;
 938         struct thandle          *th;
 939         ENTRY;
 940
 941         /* on pending IO in this thread should left from prev. request */
 942         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 943
 944         th = ERR_PTR(-ENOMEM);
 945         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 946         if (oh != NULL) {
 947                 oh->ot_quota_trans = &oti->oti_quota_trans;
 948                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 949                 th = &oh->ot_super;
 950                 th->th_dev = d;
 951                 th->th_result = 0;
 952                 th->th_tags = LCT_TX_HANDLE;
 953                 oh->ot_credits = 0;
 954                 atomic_set(&th->th_refc, 1);
 955                 th->th_alloc_size = sizeof(*oh);
 956                 oti->oti_dev = osd_dt_dev(d);
 957                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 958                 osd_th_alloced(oh);
 959
 960                 memset(oti->oti_declare_ops, 0,
 961                                         sizeof(oti->oti_declare_ops));
 962                 memset(oti->oti_declare_ops_rb, 0,
 963                                         sizeof(oti->oti_declare_ops_rb));
 964                 memset(oti->oti_declare_ops_cred, 0,
 965                                         sizeof(oti->oti_declare_ops_cred));
 966                 oti->oti_rollback = false;
 967         }
 968         RETURN(th);
 969 }
 970
 971 /*
 972  * Concurrency: shouldn't matter.
 973  */
 974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 975                     struct thandle *th)
 976 {
 977         struct osd_thread_info *oti = osd_oti_get(env);
 978         struct osd_device  *dev = osd_dt_dev(d);
 979         handle_t           *jh;
 980         struct osd_thandle *oh;
 981         int rc;
 982
 983         ENTRY;
 984
 985         LASSERT(current->journal_info == NULL);
 986
 987         oh = container_of0(th, struct osd_thandle, ot_super);
 988         LASSERT(oh != NULL);
 989         LASSERT(oh->ot_handle == NULL);
 990
 991         rc = dt_txn_hook_start(env, d, th);
 992         if (rc != 0)
 993                 GOTO(out, rc);
 994
 995         if (unlikely(osd_param_is_not_sane(dev, th))) {
 996                 static unsigned long last_printed;
 997                 static int last_credits;
 998
 999                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1001                       oh->ot_credits,
1002                       osd_journal(dev)->j_max_transaction_buffers);
1003                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1004                       oti->oti_declare_ops[OSD_OT_CREATE],
1005                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006                       oti->oti_declare_ops[OSD_OT_DESTROY],
1007                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_WRITE],
1015                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016                       oti->oti_declare_ops[OSD_OT_PUNCH],
1017                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops[OSD_OT_QUOTA],
1019                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1021                       oti->oti_declare_ops[OSD_OT_INSERT],
1022                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023                       oti->oti_declare_ops[OSD_OT_DELETE],
1024                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1027                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1029                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1030
1031                 if (last_credits != oh->ot_credits &&
1032                     time_after(jiffies, last_printed + 60 * HZ)) {
1033                         libcfs_debug_dumpstack(NULL);
1034                         last_credits = oh->ot_credits;
1035                         last_printed = jiffies;
1036                 }
1037                 /* XXX Limit the credits to 'max_transaction_buffers', and
1038                  *     let the underlying filesystem to catch the error if
1039                  *     we really need so many credits.
1040                  *
1041                  *     This should be removed when we can calculate the
1042                  *     credits precisely. */
1043                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1044         }
1045
1046         /*
1047          * XXX temporary stuff. Some abstraction layer should
1048          * be used.
1049          */
1050         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1051         osd_th_started(oh);
1052         if (!IS_ERR(jh)) {
1053                 oh->ot_handle = jh;
1054                 LASSERT(oti->oti_txns == 0);
1055                 lu_context_init(&th->th_ctx, th->th_tags);
1056                 lu_context_enter(&th->th_ctx);
1057
1058                 lu_device_get(&d->dd_lu_dev);
1059                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1060                               "osd-tx", th);
1061                 oti->oti_txns++;
1062                 rc = 0;
1063         } else {
1064                 rc = PTR_ERR(jh);
1065         }
1066 out:
1067         RETURN(rc);
1068 }
1069
1070 static int osd_seq_exists(const struct lu_env *env,
1071                               struct osd_device *osd, obd_seq seq)
1072 {
1073         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1074         struct seq_server_site  *ss = osd_seq_site(osd);
1075         int                     rc;
1076         ENTRY;
1077
1078         if (ss == NULL)
1079                 RETURN(1);
1080
1081         rc = osd_fld_lookup(env, osd, seq, range);
1082         if (rc != 0) {
1083                 if (rc != -ENOENT)
1084                         CERROR("%s: can't lookup FLD sequence "LPX64
1085                                ": rc = %d\n", osd_name(osd), seq, rc);
1086                 RETURN(0);
1087         }
1088
1089         RETURN(ss->ss_node_id == range->lsr_index);
1090 }
1091
1092 /*
1093  * Concurrency: shouldn't matter.
1094  */
1095 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1096                           struct thandle *th)
1097 {
1098         int                     rc = 0;
1099         struct osd_thandle     *oh;
1100         struct osd_thread_info *oti = osd_oti_get(env);
1101         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1102         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1103         struct lquota_trans    *qtrans;
1104         ENTRY;
1105
1106         oh = container_of0(th, struct osd_thandle, ot_super);
1107
1108         qtrans = oh->ot_quota_trans;
1109         oh->ot_quota_trans = NULL;
1110
1111         if (oh->ot_handle != NULL) {
1112                 handle_t *hdl = oh->ot_handle;
1113
1114                 /*
1115                  * add commit callback
1116                  * notice we don't do this in osd_trans_start()
1117                  * as underlying transaction can change during truncate
1118                  */
1119                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1120                                          &oh->ot_jcb);
1121
1122                 LASSERT(oti->oti_txns == 1);
1123                 oti->oti_txns--;
1124                 rc = dt_txn_hook_stop(env, th);
1125                 if (rc != 0)
1126                         CERROR("Failure in transaction hook: %d\n", rc);
1127
1128                 /* hook functions might modify th_sync */
1129                 hdl->h_sync = th->th_sync;
1130
1131                 oh->ot_handle = NULL;
1132                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1133                                   rc = ldiskfs_journal_stop(hdl));
1134                 if (rc != 0)
1135                         CERROR("Failure to stop transaction: %d\n", rc);
1136         } else {
1137                 thandle_put(&oh->ot_super);
1138         }
1139
1140         /* inform the quota slave device that the transaction is stopping */
1141         qsd_op_end(env, qsd, qtrans);
1142
1143         /* as we want IO to journal and data IO be concurrent, we don't block
1144          * awaiting data IO completion in osd_do_bio(), instead we wait here
1145          * once transaction is submitted to the journal. all reqular requests
1146          * don't do direct IO (except read/write), thus this wait_event becomes
1147          * no-op for them.
1148          *
1149          * IMPORTANT: we have to wait till any IO submited by the thread is
1150          * completed otherwise iobuf may be corrupted by different request
1151          */
1152         wait_event(iobuf->dr_wait,
1153                        atomic_read(&iobuf->dr_numreqs) == 0);
1154         osd_fini_iobuf(oti->oti_dev, iobuf);
1155         if (!rc)
1156                 rc = iobuf->dr_error;
1157
1158         RETURN(rc);
1159 }
1160
1161 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 {
1163         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1164                                                ot_super);
1165
1166         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1167         LASSERT(&dcb->dcb_func != NULL);
1168         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1169
1170         return 0;
1171 }
1172
1173 /*
1174  * Called just before object is freed. Releases all resources except for
1175  * object itself (that is released by osd_object_free()).
1176  *
1177  * Concurrency: no concurrent access is possible that late in object
1178  * life-cycle.
1179  */
1180 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 {
1182         struct osd_object *obj   = osd_obj(l);
1183         struct inode      *inode = obj->oo_inode;
1184
1185         LINVRNT(osd_invariant(obj));
1186
1187         /*
1188          * If object is unlinked remove fid->ino mapping from object index.
1189          */
1190
1191         osd_index_fini(obj);
1192         if (inode != NULL) {
1193                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1194                 qid_t                    uid = i_uid_read(inode);
1195                 qid_t                    gid = i_gid_read(inode);
1196
1197                 iput(inode);
1198                 obj->oo_inode = NULL;
1199
1200                 if (qsd != NULL) {
1201                         struct osd_thread_info  *info = osd_oti_get(env);
1202                         struct lquota_id_info   *qi = &info->oti_qi;
1203
1204                         /* Release granted quota to master if necessary */
1205                         qi->lqi_id.qid_uid = uid;
1206                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207
1208                         qi->lqi_id.qid_uid = gid;
1209                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1210                 }
1211         }
1212 }
1213
1214 /*
1215  * Concurrency: ->loo_object_release() is called under site spin-lock.
1216  */
1217 static void osd_object_release(const struct lu_env *env,
1218                                struct lu_object *l)
1219 {
1220 }
1221
1222 /*
1223  * Concurrency: shouldn't matter.
1224  */
1225 static int osd_object_print(const struct lu_env *env, void *cookie,
1226                             lu_printer_t p, const struct lu_object *l)
1227 {
1228         struct osd_object *o = osd_obj(l);
1229         struct iam_descr  *d;
1230
1231         if (o->oo_dir != NULL)
1232                 d = o->oo_dir->od_container.ic_descr;
1233         else
1234                 d = NULL;
1235         return (*p)(env, cookie,
1236                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237                     o, o->oo_inode,
1238                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1239                     o->oo_inode ? o->oo_inode->i_generation : 0,
1240                     d ? d->id_ops->id_name : "plain");
1241 }
1242
1243 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1244
1245 /*
1246  * Concurrency: shouldn't matter.
1247  */
1248 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1249                struct obd_statfs *sfs)
1250 {
1251         struct osd_device  *osd = osd_dt_dev(d);
1252         struct super_block *sb = osd_sb(osd);
1253         struct kstatfs     *ksfs;
1254         int result = 0;
1255
1256         if (unlikely(osd->od_mnt == NULL))
1257                 return -EINPROGRESS;
1258
1259         /* osd_lproc.c call this without env, allocate ksfs for that case */
1260         if (unlikely(env == NULL)) {
1261                 OBD_ALLOC_PTR(ksfs);
1262                 if (ksfs == NULL)
1263                         return -ENOMEM;
1264         } else {
1265                 ksfs = &osd_oti_get(env)->oti_ksfs;
1266         }
1267
1268         spin_lock(&osd->od_osfs_lock);
1269         result = sb->s_op->statfs(sb->s_root, ksfs);
1270         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1271                 statfs_pack(sfs, ksfs);
1272                 if (sb->s_flags & MS_RDONLY)
1273                         sfs->os_state = OS_STATE_READONLY;
1274         }
1275
1276         spin_unlock(&osd->od_osfs_lock);
1277
1278         if (unlikely(env == NULL))
1279                 OBD_FREE_PTR(ksfs);
1280
1281         /* Reserve a small amount of space for local objects like last_rcvd,
1282          * llog, quota files, ... */
1283         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1284                 sfs->os_bavail = 0;
1285         } else {
1286                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1287                 /** Take out metadata overhead for indirect blocks */
1288                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1289         }
1290
1291         return result;
1292 }
1293
1294 /**
1295  * Estimate space needed for file creations. We assume the largest filename
1296  * which is 2^64 - 1, hence a filename of 20 chars.
1297  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298  */
1299 #ifdef __LDISKFS_DIR_REC_LEN
1300 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #else
1302 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1303 #endif
1304
1305 /*
1306  * Concurrency: doesn't access mutable data.
1307  */
1308 static void osd_conf_get(const struct lu_env *env,
1309                          const struct dt_device *dev,
1310                          struct dt_device_param *param)
1311 {
1312         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1313         int                ea_overhead;
1314
1315         /*
1316          * XXX should be taken from not-yet-existing fs abstraction layer.
1317          */
1318         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1319         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1320         param->ddp_block_shift  = sb->s_blocksize_bits;
1321         param->ddp_mount_type     = LDD_MT_LDISKFS;
1322         param->ddp_maxbytes       = sb->s_maxbytes;
1323         /* Overhead estimate should be fairly accurate, so we really take a tiny
1324          * error margin which also avoids fragmenting the filesystem too much */
1325         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1326         /* inode are statically allocated, so per-inode space consumption
1327          * is the space consumed by the directory entry */
1328         param->ddp_inodespace     = PER_OBJ_USAGE;
1329         /* per-fragment overhead to be used by the client code */
1330         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1331         param->ddp_mntopts      = 0;
1332         if (test_opt(sb, XATTR_USER))
1333                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1334         if (test_opt(sb, POSIX_ACL))
1335                 param->ddp_mntopts |= MNTOPT_ACL;
1336
1337         /* LOD might calculate the max stripe count based on max_ea_size,
1338          * so we need take account in the overhead as well,
1339          * xattr_header + magic + xattr_entry_head */
1340         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1341                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342
1343 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1344         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1345                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1346                                                                 ea_overhead;
1347         else
1348 #endif
1349                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1350 }
1351
1352 /*
1353  * Concurrency: shouldn't matter.
1354  */
1355 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 {
1357         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1358         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1359 }
1360
1361 /**
1362  * Start commit for OSD device.
1363  *
1364  * An implementation of dt_commit_async method for OSD device.
1365  * Asychronously starts underlayng fs sync and thereby a transaction
1366  * commit.
1367  *
1368  * \param env environment
1369  * \param d dt device
1370  *
1371  * \see dt_device_operations
1372  */
1373 static int osd_commit_async(const struct lu_env *env,
1374                             struct dt_device *d)
1375 {
1376         struct super_block *s = osd_sb(osd_dt_dev(d));
1377         ENTRY;
1378
1379         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1380         RETURN(s->s_op->sync_fs(s, 0));
1381 }
1382
1383 /*
1384  * Concurrency: shouldn't matter.
1385  */
1386
1387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 {
1389         struct super_block *sb = osd_sb(osd_dt_dev(d));
1390         struct block_device *dev = sb->s_bdev;
1391 #ifdef HAVE_DEV_SET_RDONLY
1392         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1393         int rc = 0;
1394 #else
1395         int rc = -EOPNOTSUPP;
1396 #endif
1397         ENTRY;
1398
1399 #ifdef HAVE_DEV_SET_RDONLY
1400         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401
1402         if (jdev && (jdev != dev)) {
1403                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404                        (long)jdev);
1405                 dev_set_rdonly(jdev);
1406         }
1407         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1408         dev_set_rdonly(dev);
1409 #else
1410         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1411                osd_dt_dev(d)->od_svname, (long)dev, rc);
1412 #endif
1413         RETURN(rc);
1414 }
1415
1416 /*
1417  * Concurrency: serialization provided by callers.
1418  */
1419 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1420                               int mode, unsigned long timeout, __u32 alg,
1421                               struct lustre_capa_key *keys)
1422 {
1423         struct osd_device *dev = osd_dt_dev(d);
1424         ENTRY;
1425
1426         dev->od_fl_capa = mode;
1427         dev->od_capa_timeout = timeout;
1428         dev->od_capa_alg = alg;
1429         dev->od_capa_keys = keys;
1430         RETURN(0);
1431 }
1432
1433 /**
1434  * Note: we do not count into QUOTA here.
1435  * If we mount with --data_journal we may need more.
1436  */
1437 const int osd_dto_credits_noquota[DTO_NR] = {
1438         /**
1439          * Insert.
1440          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1441          * SINGLEDATA_TRANS_BLOCKS(8)
1442          * XXX Note: maybe iam need more, since iam have more level than
1443          *           EXT3 htree.
1444          */
1445         [DTO_INDEX_INSERT]  = 16,
1446         /**
1447          * Delete
1448          * just modify a single entry, probably merge few within a block
1449          */
1450         [DTO_INDEX_DELETE]  = 1,
1451         /**
1452          * Used for OI scrub
1453          */
1454         [DTO_INDEX_UPDATE]  = 16,
1455         /**
1456          * 4(inode, inode bits, groups, GDT)
1457          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1458          */
1459         [DTO_OBJECT_CREATE] = 4,
1460         /**
1461          * 4(inode, inode bits, groups, GDT)
1462          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1463          */
1464         [DTO_OBJECT_DELETE] = 4,
1465         /**
1466          * Attr set credits (inode)
1467          */
1468         [DTO_ATTR_SET_BASE] = 1,
1469         /**
1470          * Xattr set. The same as xattr of EXT3.
1471          * DATA_TRANS_BLOCKS(14)
1472          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1473          * are also counted in. Do not know why?
1474          */
1475         [DTO_XATTR_SET]     = 14,
1476         /**
1477          * credits for inode change during write.
1478          */
1479         [DTO_WRITE_BASE]    = 3,
1480         /**
1481          * credits for single block write.
1482          */
1483         [DTO_WRITE_BLOCK]   = 14,
1484         /**
1485          * Attr set credits for chown.
1486          * This is extra credits for setattr, and it is null without quota
1487          */
1488         [DTO_ATTR_SET_CHOWN] = 0
1489 };
1490
1491 static const struct dt_device_operations osd_dt_ops = {
1492         .dt_root_get       = osd_root_get,
1493         .dt_statfs         = osd_statfs,
1494         .dt_trans_create   = osd_trans_create,
1495         .dt_trans_start    = osd_trans_start,
1496         .dt_trans_stop     = osd_trans_stop,
1497         .dt_trans_cb_add   = osd_trans_cb_add,
1498         .dt_conf_get       = osd_conf_get,
1499         .dt_sync           = osd_sync,
1500         .dt_ro             = osd_ro,
1501         .dt_commit_async   = osd_commit_async,
1502         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1503 };
1504
1505 static void osd_object_read_lock(const struct lu_env *env,
1506                                  struct dt_object *dt, unsigned role)
1507 {
1508         struct osd_object *obj = osd_dt_obj(dt);
1509         struct osd_thread_info *oti = osd_oti_get(env);
1510
1511         LINVRNT(osd_invariant(obj));
1512
1513         LASSERT(obj->oo_owner != env);
1514         down_read_nested(&obj->oo_sem, role);
1515
1516         LASSERT(obj->oo_owner == NULL);
1517         oti->oti_r_locks++;
1518 }
1519
1520 static void osd_object_write_lock(const struct lu_env *env,
1521                                   struct dt_object *dt, unsigned role)
1522 {
1523         struct osd_object *obj = osd_dt_obj(dt);
1524         struct osd_thread_info *oti = osd_oti_get(env);
1525
1526         LINVRNT(osd_invariant(obj));
1527
1528         LASSERT(obj->oo_owner != env);
1529         down_write_nested(&obj->oo_sem, role);
1530
1531         LASSERT(obj->oo_owner == NULL);
1532         obj->oo_owner = env;
1533         oti->oti_w_locks++;
1534 }
1535
1536 static void osd_object_read_unlock(const struct lu_env *env,
1537                                    struct dt_object *dt)
1538 {
1539         struct osd_object *obj = osd_dt_obj(dt);
1540         struct osd_thread_info *oti = osd_oti_get(env);
1541
1542         LINVRNT(osd_invariant(obj));
1543
1544         LASSERT(oti->oti_r_locks > 0);
1545         oti->oti_r_locks--;
1546         up_read(&obj->oo_sem);
1547 }
1548
1549 static void osd_object_write_unlock(const struct lu_env *env,
1550                                     struct dt_object *dt)
1551 {
1552         struct osd_object *obj = osd_dt_obj(dt);
1553         struct osd_thread_info *oti = osd_oti_get(env);
1554
1555         LINVRNT(osd_invariant(obj));
1556
1557         LASSERT(obj->oo_owner == env);
1558         LASSERT(oti->oti_w_locks > 0);
1559         oti->oti_w_locks--;
1560         obj->oo_owner = NULL;
1561         up_write(&obj->oo_sem);
1562 }
1563
1564 static int osd_object_write_locked(const struct lu_env *env,
1565                                    struct dt_object *dt)
1566 {
1567         struct osd_object *obj = osd_dt_obj(dt);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         return obj->oo_owner == env;
1572 }
1573
1574 static int capa_is_sane(const struct lu_env *env,
1575                         struct osd_device *dev,
1576                         struct lustre_capa *capa,
1577                         struct lustre_capa_key *keys)
1578 {
1579         struct osd_thread_info *oti = osd_oti_get(env);
1580         struct lustre_capa *tcapa = &oti->oti_capa;
1581         struct obd_capa *oc;
1582         int i, rc = 0;
1583         ENTRY;
1584
1585         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586         if (oc) {
1587                 if (capa_is_expired(oc)) {
1588                         DEBUG_CAPA(D_ERROR, capa, "expired");
1589                         rc = -ESTALE;
1590                 }
1591                 capa_put(oc);
1592                 RETURN(rc);
1593         }
1594
1595         if (capa_is_expired_sec(capa)) {
1596                 DEBUG_CAPA(D_ERROR, capa, "expired");
1597                 RETURN(-ESTALE);
1598         }
1599
1600         spin_lock(&capa_lock);
1601         for (i = 0; i < 2; i++) {
1602                 if (keys[i].lk_keyid == capa->lc_keyid) {
1603                         oti->oti_capa_key = keys[i];
1604                         break;
1605                 }
1606         }
1607         spin_unlock(&capa_lock);
1608
1609         if (i == 2) {
1610                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1619                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1620                 RETURN(-EACCES);
1621         }
1622
1623         oc = capa_add(dev->od_capa_hash, capa);
1624         capa_put(oc);
1625
1626         RETURN(0);
1627 }
1628
1629 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1630                     struct lustre_capa *capa, __u64 opc)
1631 {
1632         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1633         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1634         struct lu_capainfo *lci;
1635         int rc;
1636
1637         if (!osd->od_fl_capa)
1638                 return 0;
1639
1640         if (capa == BYPASS_CAPA)
1641                 return 0;
1642
1643         lci = lu_capainfo_get(env);
1644         if (unlikely(lci == NULL))
1645                 return 0;
1646
1647         if (lci->lci_auth == LC_ID_NONE)
1648                 return 0;
1649
1650         if (capa == NULL) {
1651                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1652                        osd_name(osd), PFID(fid), -EACCES);
1653                 return -EACCES;
1654         }
1655
1656         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1657                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1658                            PFID(fid));
1659                 return -EACCES;
1660         }
1661
1662         if (!capa_opc_supported(capa, opc)) {
1663                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1664                 return -EACCES;
1665         }
1666
1667         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668         if (rc != 0) {
1669                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1670                 return -EACCES;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static struct timespec *osd_inode_time(const struct lu_env *env,
1677                                        struct inode *inode, __u64 seconds)
1678 {
1679         struct osd_thread_info  *oti = osd_oti_get(env);
1680         struct timespec         *t   = &oti->oti_time;
1681
1682         t->tv_sec = seconds;
1683         t->tv_nsec = 0;
1684         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1685         return t;
1686 }
1687
1688
1689 static void osd_inode_getattr(const struct lu_env *env,
1690                               struct inode *inode, struct lu_attr *attr)
1691 {
1692         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1693                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1694                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1695                            LA_TYPE;
1696
1697         attr->la_atime   = LTIME_S(inode->i_atime);
1698         attr->la_mtime   = LTIME_S(inode->i_mtime);
1699         attr->la_ctime   = LTIME_S(inode->i_ctime);
1700         attr->la_mode    = inode->i_mode;
1701         attr->la_size    = i_size_read(inode);
1702         attr->la_blocks  = inode->i_blocks;
1703         attr->la_uid     = i_uid_read(inode);
1704         attr->la_gid     = i_gid_read(inode);
1705         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1706         attr->la_nlink   = inode->i_nlink;
1707         attr->la_rdev    = inode->i_rdev;
1708         attr->la_blksize = 1 << inode->i_blkbits;
1709         attr->la_blkbits = inode->i_blkbits;
1710 }
1711
1712 static int osd_attr_get(const struct lu_env *env,
1713                         struct dt_object *dt,
1714                         struct lu_attr *attr,
1715                         struct lustre_capa *capa)
1716 {
1717         struct osd_object *obj = osd_dt_obj(dt);
1718
1719         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1720         LINVRNT(osd_invariant(obj));
1721
1722         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1723                 return -EACCES;
1724
1725         spin_lock(&obj->oo_guard);
1726         osd_inode_getattr(env, obj->oo_inode, attr);
1727         spin_unlock(&obj->oo_guard);
1728         return 0;
1729 }
1730
1731 static int osd_declare_attr_set(const struct lu_env *env,
1732                                 struct dt_object *dt,
1733                                 const struct lu_attr *attr,
1734                                 struct thandle *handle)
1735 {
1736         struct osd_thandle     *oh;
1737         struct osd_object      *obj;
1738         struct osd_thread_info *info = osd_oti_get(env);
1739         struct lquota_id_info  *qi = &info->oti_qi;
1740         qid_t                   uid;
1741         qid_t                   gid;
1742         long long               bspace;
1743         int                     rc = 0;
1744         bool                    allocated;
1745         ENTRY;
1746
1747         LASSERT(dt != NULL);
1748         LASSERT(handle != NULL);
1749
1750         obj = osd_dt_obj(dt);
1751         LASSERT(osd_invariant(obj));
1752
1753         oh = container_of0(handle, struct osd_thandle, ot_super);
1754         LASSERT(oh->ot_handle == NULL);
1755
1756         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1757                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1758
1759         if (attr == NULL || obj->oo_inode == NULL)
1760                 RETURN(rc);
1761
1762         bspace   = obj->oo_inode->i_blocks;
1763         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1764         bspace   = toqb(bspace);
1765
1766         /* Changing ownership is always preformed by super user, it should not
1767          * fail with EDQUOT.
1768          *
1769          * We still need to call the osd_declare_qid() to calculate the journal
1770          * credits for updating quota accounting files and to trigger quota
1771          * space adjustment once the operation is completed.*/
1772         if ((attr->la_valid & LA_UID) != 0 &&
1773              attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
1774                 qi->lqi_type = USRQUOTA;
1775
1776                 /* inode accounting */
1777                 qi->lqi_is_blk = false;
1778
1779                 /* one more inode for the new owner ... */
1780                 qi->lqi_id.qid_uid = attr->la_uid;
1781                 qi->lqi_space      = 1;
1782                 allocated = (attr->la_uid == 0) ? true : false;
1783                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1784                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1785                         rc = 0;
1786                 if (rc)
1787                         RETURN(rc);
1788
1789                 /* and one less inode for the current uid */
1790                 qi->lqi_id.qid_uid = uid;
1791                 qi->lqi_space      = -1;
1792                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1793                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1794                         rc = 0;
1795                 if (rc)
1796                         RETURN(rc);
1797
1798                 /* block accounting */
1799                 qi->lqi_is_blk = true;
1800
1801                 /* more blocks for the new owner ... */
1802                 qi->lqi_id.qid_uid = attr->la_uid;
1803                 qi->lqi_space      = bspace;
1804                 allocated = (attr->la_uid == 0) ? true : false;
1805                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1806                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1807                         rc = 0;
1808                 if (rc)
1809                         RETURN(rc);
1810
1811                 /* and finally less blocks for the current owner */
1812                 qi->lqi_id.qid_uid = uid;
1813                 qi->lqi_space      = -bspace;
1814                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1815                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1816                         rc = 0;
1817                 if (rc)
1818                         RETURN(rc);
1819         }
1820
1821         if (attr->la_valid & LA_GID &&
1822             attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
1823                 qi->lqi_type = GRPQUOTA;
1824
1825                 /* inode accounting */
1826                 qi->lqi_is_blk = false;
1827
1828                 /* one more inode for the new group owner ... */
1829                 qi->lqi_id.qid_gid = attr->la_gid;
1830                 qi->lqi_space      = 1;
1831                 allocated = (attr->la_gid == 0) ? true : false;
1832                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1833                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1834                         rc = 0;
1835                 if (rc)
1836                         RETURN(rc);
1837
1838                 /* and one less inode for the current gid */
1839                 qi->lqi_id.qid_gid = gid;
1840                 qi->lqi_space      = -1;
1841                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1842                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1843                         rc = 0;
1844                 if (rc)
1845                         RETURN(rc);
1846
1847                 /* block accounting */
1848                 qi->lqi_is_blk = true;
1849
1850                 /* more blocks for the new owner ... */
1851                 qi->lqi_id.qid_gid = attr->la_gid;
1852                 qi->lqi_space      = bspace;
1853                 allocated = (attr->la_gid == 0) ? true : false;
1854                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1855                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1856                         rc = 0;
1857                 if (rc)
1858                         RETURN(rc);
1859
1860                 /* and finally less blocks for the current owner */
1861                 qi->lqi_id.qid_gid = gid;
1862                 qi->lqi_space      = -bspace;
1863                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1864                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1865                         rc = 0;
1866                 if (rc)
1867                         RETURN(rc);
1868         }
1869
1870         RETURN(rc);
1871 }
1872
1873 static int osd_inode_setattr(const struct lu_env *env,
1874                              struct inode *inode, const struct lu_attr *attr)
1875 {
1876         __u64 bits;
1877
1878         bits = attr->la_valid;
1879
1880         if (bits & LA_ATIME)
1881                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1882         if (bits & LA_CTIME)
1883                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1884         if (bits & LA_MTIME)
1885                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1886         if (bits & LA_SIZE) {
1887                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1888                 i_size_write(inode, attr->la_size);
1889         }
1890
1891 #if 0
1892         /* OSD should not change "i_blocks" which is used by quota.
1893          * "i_blocks" should be changed by ldiskfs only. */
1894         if (bits & LA_BLOCKS)
1895                 inode->i_blocks = attr->la_blocks;
1896 #endif
1897         if (bits & LA_MODE)
1898                 inode->i_mode = (inode->i_mode & S_IFMT) |
1899                                 (attr->la_mode & ~S_IFMT);
1900         if (bits & LA_UID)
1901                 i_uid_write(inode, attr->la_uid);
1902         if (bits & LA_GID)
1903                 i_gid_write(inode, attr->la_gid);
1904         if (bits & LA_NLINK)
1905                 set_nlink(inode, attr->la_nlink);
1906         if (bits & LA_RDEV)
1907                 inode->i_rdev = attr->la_rdev;
1908
1909         if (bits & LA_FLAGS) {
1910                 /* always keep S_NOCMTIME */
1911                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1912                                  S_NOCMTIME;
1913         }
1914         return 0;
1915 }
1916
1917 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1918 {
1919         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1920             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1921                 struct iattr    iattr;
1922                 int             rc;
1923
1924                 iattr.ia_valid = 0;
1925                 if (attr->la_valid & LA_UID)
1926                         iattr.ia_valid |= ATTR_UID;
1927                 if (attr->la_valid & LA_GID)
1928                         iattr.ia_valid |= ATTR_GID;
1929                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1930                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1931
1932                 rc = ll_vfs_dq_transfer(inode, &iattr);
1933                 if (rc) {
1934                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1935                                "enforcement enabled on the ldiskfs filesystem?",
1936                                inode->i_sb->s_id, rc);
1937                         return rc;
1938                 }
1939         }
1940         return 0;
1941 }
1942
1943 static int osd_attr_set(const struct lu_env *env,
1944                         struct dt_object *dt,
1945                         const struct lu_attr *attr,
1946                         struct thandle *handle,
1947                         struct lustre_capa *capa)
1948 {
1949         struct osd_object *obj = osd_dt_obj(dt);
1950         struct inode      *inode;
1951         int rc;
1952
1953         LASSERT(handle != NULL);
1954         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1955         LASSERT(osd_invariant(obj));
1956
1957         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1958                 return -EACCES;
1959
1960         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1961
1962         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1963                 struct osd_thread_info  *oti  = osd_oti_get(env);
1964                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1965                 struct lu_fid           *fid1 = &oti->oti_fid;
1966                 struct osd_inode_id     *id   = &oti->oti_id;
1967                 struct iam_path_descr   *ipd;
1968                 struct iam_container    *bag;
1969                 struct osd_thandle      *oh;
1970                 int                      rc;
1971
1972                 fid_cpu_to_be(fid1, fid0);
1973                 memset(id, 1, sizeof(*id));
1974                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1975                                   fid0)->oi_dir.od_container;
1976                 ipd = osd_idx_ipd_get(env, bag);
1977                 if (unlikely(ipd == NULL))
1978                         RETURN(-ENOMEM);
1979
1980                 oh = container_of0(handle, struct osd_thandle, ot_super);
1981                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1982                                 (const struct iam_rec *)id, ipd);
1983                 osd_ipd_put(env, bag, ipd);
1984                 return(rc > 0 ? 0 : rc);
1985         }
1986
1987         inode = obj->oo_inode;
1988         ll_vfs_dq_init(inode);
1989
1990         rc = osd_quota_transfer(inode, attr);
1991         if (rc)
1992                 return rc;
1993
1994         spin_lock(&obj->oo_guard);
1995         rc = osd_inode_setattr(env, inode, attr);
1996         spin_unlock(&obj->oo_guard);
1997
1998         if (!rc)
1999                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2000         return rc;
2001 }
2002
2003 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2004                                     struct osd_object *obj,
2005                                     const char *name, const int namelen)
2006 {
2007         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2008 }
2009
2010 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2011                       umode_t mode, struct dt_allocation_hint *hint,
2012                       struct thandle *th)
2013 {
2014         int result;
2015         struct osd_device  *osd = osd_obj2dev(obj);
2016         struct osd_thandle *oth;
2017         struct dt_object   *parent = NULL;
2018         struct inode       *inode;
2019
2020         LINVRNT(osd_invariant(obj));
2021         LASSERT(obj->oo_inode == NULL);
2022         LASSERT(obj->oo_hl_head == NULL);
2023
2024         if (S_ISDIR(mode) && ldiskfs_pdo) {
2025                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2026                 if (obj->oo_hl_head == NULL)
2027                         return -ENOMEM;
2028         }
2029
2030         oth = container_of(th, struct osd_thandle, ot_super);
2031         LASSERT(oth->ot_handle->h_transaction != NULL);
2032
2033         if (hint && hint->dah_parent)
2034                 parent = hint->dah_parent;
2035
2036         inode = ldiskfs_create_inode(oth->ot_handle,
2037                                      parent ? osd_dt_obj(parent)->oo_inode :
2038                                               osd_sb(osd)->s_root->d_inode,
2039                                      mode);
2040         if (!IS_ERR(inode)) {
2041                 /* Do not update file c/mtime in ldiskfs.
2042                  * NB: don't need any lock because no contention at this
2043                  * early stage */
2044                 inode->i_flags |= S_NOCMTIME;
2045
2046                 /* For new created object, it must be consistent,
2047                  * and it is unnecessary to scrub against it. */
2048                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2049                 obj->oo_inode = inode;
2050                 result = 0;
2051         } else {
2052                 if (obj->oo_hl_head != NULL) {
2053                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2054                         obj->oo_hl_head = NULL;
2055                 }
2056                 result = PTR_ERR(inode);
2057         }
2058         LINVRNT(osd_invariant(obj));
2059         return result;
2060 }
2061
2062 enum {
2063         OSD_NAME_LEN = 255
2064 };
2065
2066 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2067                      struct lu_attr *attr,
2068                      struct dt_allocation_hint *hint,
2069                      struct dt_object_format *dof,
2070                      struct thandle *th)
2071 {
2072         int result;
2073         struct osd_thandle *oth;
2074         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2075
2076         LASSERT(S_ISDIR(attr->la_mode));
2077
2078         oth = container_of(th, struct osd_thandle, ot_super);
2079         LASSERT(oth->ot_handle->h_transaction != NULL);
2080         result = osd_mkfile(info, obj, mode, hint, th);
2081
2082         return result;
2083 }
2084
2085 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2086                         struct lu_attr *attr,
2087                         struct dt_allocation_hint *hint,
2088                         struct dt_object_format *dof,
2089                         struct thandle *th)
2090 {
2091         int result;
2092         struct osd_thandle *oth;
2093         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2094
2095         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2096
2097         LASSERT(S_ISREG(attr->la_mode));
2098
2099         oth = container_of(th, struct osd_thandle, ot_super);
2100         LASSERT(oth->ot_handle->h_transaction != NULL);
2101
2102         result = osd_mkfile(info, obj, mode, hint, th);
2103         if (result == 0) {
2104                 LASSERT(obj->oo_inode != NULL);
2105                 if (feat->dif_flags & DT_IND_VARKEY)
2106                         result = iam_lvar_create(obj->oo_inode,
2107                                                  feat->dif_keysize_max,
2108                                                  feat->dif_ptrsize,
2109                                                  feat->dif_recsize_max,
2110                                                  oth->ot_handle);
2111                 else
2112                         result = iam_lfix_create(obj->oo_inode,
2113                                                  feat->dif_keysize_max,
2114                                                  feat->dif_ptrsize,
2115                                                  feat->dif_recsize_max,
2116                                                  oth->ot_handle);
2117
2118         }
2119         return result;
2120 }
2121
2122 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2123                      struct lu_attr *attr,
2124                      struct dt_allocation_hint *hint,
2125                      struct dt_object_format *dof,
2126                      struct thandle *th)
2127 {
2128         LASSERT(S_ISREG(attr->la_mode));
2129         return osd_mkfile(info, obj, (attr->la_mode &
2130                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2131 }
2132
2133 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2134                      struct lu_attr *attr,
2135                      struct dt_allocation_hint *hint,
2136                      struct dt_object_format *dof,
2137                      struct thandle *th)
2138 {
2139         LASSERT(S_ISLNK(attr->la_mode));
2140         return osd_mkfile(info, obj, (attr->la_mode &
2141                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2142 }
2143
2144 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2145                      struct lu_attr *attr,
2146                      struct dt_allocation_hint *hint,
2147                      struct dt_object_format *dof,
2148                      struct thandle *th)
2149 {
2150         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2151         int result;
2152
2153         LINVRNT(osd_invariant(obj));
2154         LASSERT(obj->oo_inode == NULL);
2155         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2156                 S_ISFIFO(mode) || S_ISSOCK(mode));
2157
2158         result = osd_mkfile(info, obj, mode, hint, th);
2159         if (result == 0) {
2160                 LASSERT(obj->oo_inode != NULL);
2161                 /*
2162                  * This inode should be marked dirty for i_rdev.  Currently
2163                  * that is done in the osd_attr_init().
2164                  */
2165                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2166                                    attr->la_rdev);
2167         }
2168         LINVRNT(osd_invariant(obj));
2169         return result;
2170 }
2171
2172 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2173                               struct lu_attr *,
2174                               struct dt_allocation_hint *hint,
2175                               struct dt_object_format *dof,
2176                               struct thandle *);
2177
2178 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2179 {
2180         osd_obj_type_f result;
2181
2182         switch (type) {
2183         case DFT_DIR:
2184                 result = osd_mkdir;
2185                 break;
2186         case DFT_REGULAR:
2187                 result = osd_mkreg;
2188                 break;
2189         case DFT_SYM:
2190                 result = osd_mksym;
2191                 break;
2192         case DFT_NODE:
2193                 result = osd_mknod;
2194                 break;
2195         case DFT_INDEX:
2196                 result = osd_mk_index;
2197                 break;
2198
2199         default:
2200                 LBUG();
2201                 break;
2202         }
2203         return result;
2204 }
2205
2206
2207 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2208                         struct dt_object *parent, struct dt_object *child,
2209                         umode_t child_mode)
2210 {
2211         LASSERT(ah);
2212
2213         ah->dah_parent = parent;
2214         ah->dah_mode = child_mode;
2215 }
2216
2217 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2218                           struct lu_attr *attr, struct dt_object_format *dof)
2219 {
2220         struct inode   *inode = obj->oo_inode;
2221         __u64           valid = attr->la_valid;
2222         int             result;
2223
2224         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2225
2226         if (dof->dof_type != DFT_NODE)
2227                 attr->la_valid &= ~LA_RDEV;
2228         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2229                 attr->la_valid &= ~LA_ATIME;
2230         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2231                 attr->la_valid &= ~LA_CTIME;
2232         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2233                 attr->la_valid &= ~LA_MTIME;
2234
2235         result = osd_quota_transfer(inode, attr);
2236         if (result)
2237                 return;
2238
2239         if (attr->la_valid != 0) {
2240                 result = osd_inode_setattr(info->oti_env, inode, attr);
2241                 /*
2242                  * The osd_inode_setattr() should always succeed here.  The
2243                  * only error that could be returned is EDQUOT when we are
2244                  * trying to change the UID or GID of the inode. However, this
2245                  * should not happen since quota enforcement is no longer
2246                  * enabled on ldiskfs (lquota takes care of it).
2247                  */
2248                 LASSERTF(result == 0, "%d", result);
2249                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2250         }
2251
2252         attr->la_valid = valid;
2253 }
2254
2255 /**
2256  * Helper function for osd_object_create()
2257  *
2258  * \retval 0, on success
2259  */
2260 static int __osd_object_create(struct osd_thread_info *info,
2261                                struct osd_object *obj, struct lu_attr *attr,
2262                                struct dt_allocation_hint *hint,
2263                                struct dt_object_format *dof,
2264                                struct thandle *th)
2265 {
2266         int     result;
2267         __u32   umask;
2268
2269         /* we drop umask so that permissions we pass are not affected */
2270         umask = current->fs->umask;
2271         current->fs->umask = 0;
2272
2273         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2274                                                   th);
2275         if (result == 0) {
2276                 osd_attr_init(info, obj, attr, dof);
2277                 osd_object_init0(obj);
2278                 /* bz 24037 */
2279                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2280                         unlock_new_inode(obj->oo_inode);
2281         }
2282
2283         /* restore previous umask value */
2284         current->fs->umask = umask;
2285
2286         return result;
2287 }
2288
2289 /**
2290  * Helper function for osd_object_create()
2291  *
2292  * \retval 0, on success
2293  */
2294 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2295                            const struct lu_fid *fid, struct thandle *th)
2296 {
2297         struct osd_thread_info *info = osd_oti_get(env);
2298         struct osd_inode_id    *id   = &info->oti_id;
2299         struct osd_device      *osd  = osd_obj2dev(obj);
2300         struct osd_thandle     *oh;
2301
2302         LASSERT(obj->oo_inode != NULL);
2303
2304         oh = container_of0(th, struct osd_thandle, ot_super);
2305         LASSERT(oh->ot_handle);
2306
2307         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2308         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2309 }
2310
2311 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2312                    obd_seq seq, struct lu_seq_range *range)
2313 {
2314         struct seq_server_site  *ss = osd_seq_site(osd);
2315
2316         if (fid_seq_is_idif(seq)) {
2317                 fld_range_set_ost(range);
2318                 range->lsr_index = idif_ost_idx(seq);
2319                 return 0;
2320         }
2321
2322         if (!fid_seq_in_fldb(seq)) {
2323                 fld_range_set_mdt(range);
2324                 if (ss != NULL)
2325                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2326                          * at all */
2327                         range->lsr_index = ss->ss_node_id;
2328                 return 0;
2329         }
2330
2331         LASSERT(ss != NULL);
2332         fld_range_set_any(range);
2333         /* OSD will only do local fld lookup */
2334         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2335 }
2336
2337 /*
2338  * Concurrency: no external locking is necessary.
2339  */
2340 static int osd_declare_object_create(const struct lu_env *env,
2341                                      struct dt_object *dt,
2342                                      struct lu_attr *attr,
2343                                      struct dt_allocation_hint *hint,
2344                                      struct dt_object_format *dof,
2345                                      struct thandle *handle)
2346 {
2347         struct osd_thandle      *oh;
2348         int                      rc;
2349         ENTRY;
2350
2351         LASSERT(handle != NULL);
2352
2353         oh = container_of0(handle, struct osd_thandle, ot_super);
2354         LASSERT(oh->ot_handle == NULL);
2355
2356         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2357                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2358         /* Reuse idle OI block may cause additional one OI block
2359          * to be changed. */
2360         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2361                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2362
2363         /* If this is directory, then we expect . and .. to be inserted as
2364          * well. The one directory block always needs to be created for the
2365          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2366          * block), there is no danger of needing a tree for the first block.
2367          */
2368         if (attr && S_ISDIR(attr->la_mode)) {
2369                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2370                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2371                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2372         }
2373
2374         if (!attr)
2375                 RETURN(0);
2376
2377         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2378                                    false, false, NULL, false);
2379         if (rc != 0)
2380                 RETURN(rc);
2381
2382         RETURN(rc);
2383 }
2384
2385 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2386                              struct lu_attr *attr,
2387                              struct dt_allocation_hint *hint,
2388                              struct dt_object_format *dof,
2389                              struct thandle *th)
2390 {
2391         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2392         struct osd_object      *obj    = osd_dt_obj(dt);
2393         struct osd_thread_info *info   = osd_oti_get(env);
2394         int result;
2395
2396         ENTRY;
2397
2398         LINVRNT(osd_invariant(obj));
2399         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2400         LASSERT(osd_write_locked(env, obj));
2401         LASSERT(th != NULL);
2402
2403         if (unlikely(fid_is_acct(fid)))
2404                 /* Quota files can't be created from the kernel any more,
2405                  * 'tune2fs -O quota' will take care of creating them */
2406                 RETURN(-EPERM);
2407
2408         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2409         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2410
2411         result = __osd_object_create(info, obj, attr, hint, dof, th);
2412         if (result == 0)
2413                 result = __osd_oi_insert(env, obj, fid, th);
2414
2415         LASSERT(ergo(result == 0,
2416                      dt_object_exists(dt) && !dt_object_remote(dt)));
2417
2418         LASSERT(osd_invariant(obj));
2419         RETURN(result);
2420 }
2421
2422 /**
2423  * Called to destroy on-disk representation of the object
2424  *
2425  * Concurrency: must be locked
2426  */
2427 static int osd_declare_object_destroy(const struct lu_env *env,
2428                                       struct dt_object *dt,
2429                                       struct thandle *th)
2430 {
2431         struct osd_object  *obj = osd_dt_obj(dt);
2432         struct inode       *inode = obj->oo_inode;
2433         struct osd_thandle *oh;
2434         int                 rc;
2435         ENTRY;
2436
2437         oh = container_of0(th, struct osd_thandle, ot_super);
2438         LASSERT(oh->ot_handle == NULL);
2439         LASSERT(inode);
2440
2441         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2442                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2443         /* Recycle idle OI leaf may cause additional three OI blocks
2444          * to be changed. */
2445         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2446                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2447         /* one less inode */
2448         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2449                                    -1, oh, false, true, NULL, false);
2450         if (rc)
2451                 RETURN(rc);
2452         /* data to be truncated */
2453         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2454                                    0, oh, true, true, NULL, false);
2455         RETURN(rc);
2456 }
2457
2458 static int osd_object_destroy(const struct lu_env *env,
2459                               struct dt_object *dt,
2460                               struct thandle *th)
2461 {
2462         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2463         struct osd_object      *obj = osd_dt_obj(dt);
2464         struct inode           *inode = obj->oo_inode;
2465         struct osd_device      *osd = osd_obj2dev(obj);
2466         struct osd_thandle     *oh;
2467         int                     result;
2468         ENTRY;
2469
2470         oh = container_of0(th, struct osd_thandle, ot_super);
2471         LASSERT(oh->ot_handle);
2472         LASSERT(inode);
2473         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2474
2475         if (unlikely(fid_is_acct(fid)))
2476                 RETURN(-EPERM);
2477
2478         if (S_ISDIR(inode->i_mode)) {
2479                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2480                         inode->i_nlink == 2);
2481                 /* it will check/delete the inode from remote parent,
2482                  * how to optimize it? unlink performance impaction XXX */
2483                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2484                 if (result != 0 && result != -ENOENT) {
2485                         CERROR("%s: delete inode "DFID": rc = %d\n",
2486                                osd_name(osd), PFID(fid), result);
2487                 }
2488                 spin_lock(&obj->oo_guard);
2489                 clear_nlink(inode);
2490                 spin_unlock(&obj->oo_guard);
2491                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2492         }
2493
2494         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2495
2496         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2497                                OI_CHECK_FLD);
2498
2499         /* XXX: add to ext3 orphan list */
2500         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2501
2502         /* not needed in the cache anymore */
2503         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2504
2505         RETURN(0);
2506 }
2507
2508 /**
2509  * Put the fid into lustre_mdt_attrs, and then place the structure
2510  * inode's ea. This fid should not be altered during the life time
2511  * of the inode.
2512  *
2513  * \retval +ve, on success
2514  * \retval -ve, on error
2515  *
2516  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2517  */
2518 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2519                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2520 {
2521         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2522         int                      rc;
2523         ENTRY;
2524
2525         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2526                 RETURN(0);
2527
2528         lustre_lma_init(lma, fid, compat, incompat);
2529         lustre_lma_swab(lma);
2530
2531         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2532                              XATTR_CREATE);
2533         /* LMA may already exist, but we need to check that all the
2534          * desired compat/incompat flags have been added. */
2535         if (unlikely(rc == -EEXIST)) {
2536                 if (compat == 0 && incompat == 0)
2537                         RETURN(0);
2538
2539                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2540                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2541                                      LMA_OLD_SIZE);
2542                 if (rc <= 0)
2543                         RETURN(-EINVAL);
2544
2545                 lustre_lma_swab(lma);
2546                 if (!(~lma->lma_compat & compat) &&
2547                     !(~lma->lma_incompat & incompat))
2548                         RETURN(0);
2549
2550                 lma->lma_compat |= compat;
2551                 lma->lma_incompat |= incompat;
2552                 lustre_lma_swab(lma);
2553                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2554                                      sizeof(*lma), XATTR_REPLACE);
2555         }
2556
2557         RETURN(rc);
2558 }
2559
2560 /**
2561  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2562  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2563  * To have compatilibility with 1.8 ldiskfs driver we need to have
2564  * magic number at start of fid data.
2565  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2566  * its inmemory API.
2567  */
2568 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2569                                   const struct dt_rec *fid)
2570 {
2571         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2572             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2573                 param->edp_magic = 0;
2574                 return;
2575         }
2576
2577         param->edp_magic = LDISKFS_LUFID_MAGIC;
2578         param->edp_len =  sizeof(struct lu_fid) + 1;
2579         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2580 }
2581
2582 /**
2583  * Try to read the fid from inode ea into dt_rec.
2584  *
2585  * \param fid object fid.
2586  *
2587  * \retval 0 on success
2588  */
2589 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2590                           __u32 ino, struct lu_fid *fid,
2591                           struct osd_inode_id *id)
2592 {
2593         struct osd_thread_info *info  = osd_oti_get(env);
2594         struct inode           *inode;
2595         ENTRY;
2596
2597         osd_id_gen(id, ino, OSD_OII_NOGEN);
2598         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2599         if (IS_ERR(inode))
2600                 RETURN(PTR_ERR(inode));
2601
2602         iput(inode);
2603         RETURN(0);
2604 }
2605
2606 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2607                                         struct inode *dir,
2608                                         struct inode  *parent_dir,
2609                                         const struct dt_rec *dot_fid,
2610                                         const struct dt_rec *dot_dot_fid,
2611                                         struct osd_thandle *oth)
2612 {
2613         struct ldiskfs_dentry_param *dot_ldp;
2614         struct ldiskfs_dentry_param *dot_dot_ldp;
2615
2616         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2617         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2618
2619         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2620         dot_ldp->edp_magic = 0;
2621         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2622                                         dir, dot_ldp, dot_dot_ldp);
2623 }
2624
2625 /**
2626  * Create an local agent inode for remote entry
2627  */
2628 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2629                                                   struct osd_device *osd,
2630                                                   struct osd_object *pobj,
2631                                                   const struct lu_fid *fid,
2632                                                   struct thandle *th)
2633 {
2634         struct osd_thread_info  *info = osd_oti_get(env);
2635         struct inode            *local;
2636         struct osd_thandle      *oh;
2637         int                     rc;
2638         ENTRY;
2639
2640         LASSERT(th);
2641         oh = container_of(th, struct osd_thandle, ot_super);
2642         LASSERT(oh->ot_handle->h_transaction != NULL);
2643
2644         /* FIXME: Insert index api needs to know the mode of
2645          * the remote object. Just use S_IFDIR for now */
2646         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2647         if (IS_ERR(local)) {
2648                 CERROR("%s: create local error %d\n", osd_name(osd),
2649                        (int)PTR_ERR(local));
2650                 RETURN(local);
2651         }
2652
2653         /* Set special LMA flag for local agent inode */
2654         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2655         if (rc != 0) {
2656                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2657                        osd_name(osd), PFID(fid), rc);
2658                 RETURN(ERR_PTR(rc));
2659         }
2660
2661         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2662                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2663                 (const struct dt_rec *)fid, oh);
2664         if (rc != 0) {
2665                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2666                         osd_name(osd), PFID(fid), rc);
2667                 RETURN(ERR_PTR(rc));
2668         }
2669
2670         RETURN(local);
2671 }
2672
2673 /**
2674  * Delete local agent inode for remote entry
2675  */
2676 static int osd_delete_local_agent_inode(const struct lu_env *env,
2677                                         struct osd_device *osd,
2678                                         const struct lu_fid *fid,
2679                                         __u32 ino, struct osd_thandle *oh)
2680 {
2681         struct osd_thread_info  *oti = osd_oti_get(env);
2682         struct osd_inode_id     *id = &oti->oti_id;
2683         struct inode            *inode;
2684         ENTRY;
2685
2686         id->oii_ino = le32_to_cpu(ino);
2687         id->oii_gen = OSD_OII_NOGEN;
2688         inode = osd_iget(oti, osd, id);
2689         if (IS_ERR(inode)) {
2690                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2691                        PFID(fid), id->oii_ino, id->oii_gen);
2692                 RETURN(PTR_ERR(inode));
2693         }
2694
2695         clear_nlink(inode);
2696         mark_inode_dirty(inode);
2697         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2698                 osd_name(osd), PFID(fid), inode->i_ino);
2699         iput(inode);
2700         RETURN(0);
2701 }
2702
2703 /**
2704  * OSD layer object create function for interoperability mode (b11826).
2705  * This is mostly similar to osd_object_create(). Only difference being, fid is
2706  * inserted into inode ea here.
2707  *
2708  * \retval   0, on success
2709  * \retval -ve, on error
2710  */
2711 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2712                                 struct lu_attr *attr,
2713                                 struct dt_allocation_hint *hint,
2714                                 struct dt_object_format *dof,
2715                                 struct thandle *th)
2716 {
2717         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2718         struct osd_object      *obj    = osd_dt_obj(dt);
2719         struct osd_thread_info *info   = osd_oti_get(env);
2720         int                     result;
2721
2722         ENTRY;
2723
2724         LASSERT(osd_invariant(obj));
2725         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2726         LASSERT(osd_write_locked(env, obj));
2727         LASSERT(th != NULL);
2728
2729         if (unlikely(fid_is_acct(fid)))
2730                 /* Quota files can't be created from the kernel any more,
2731                  * 'tune2fs -O quota' will take care of creating them */
2732                 RETURN(-EPERM);
2733
2734         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2735         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2736
2737         result = __osd_object_create(info, obj, attr, hint, dof, th);
2738         if (result == 0)
2739                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2740                                 fid_is_on_ost(info, osd_obj2dev(obj),
2741                                               fid, OI_CHECK_FLD) ?
2742                                 LMAC_FID_ON_OST : 0, 0);
2743
2744         if (result == 0)
2745                 result = __osd_oi_insert(env, obj, fid, th);
2746
2747         LASSERT(ergo(result == 0,
2748                      dt_object_exists(dt) && !dt_object_remote(dt)));
2749         LINVRNT(osd_invariant(obj));
2750         RETURN(result);
2751 }
2752
2753 static int osd_declare_object_ref_add(const struct lu_env *env,
2754                                       struct dt_object *dt,
2755                                       struct thandle *handle)
2756 {
2757         struct osd_thandle       *oh;
2758
2759         /* it's possible that object doesn't exist yet */
2760         LASSERT(handle != NULL);
2761
2762         oh = container_of0(handle, struct osd_thandle, ot_super);
2763         LASSERT(oh->ot_handle == NULL);
2764
2765         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2766                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2767
2768         return 0;
2769 }
2770
2771 /*
2772  * Concurrency: @dt is write locked.
2773  */
2774 static int osd_object_ref_add(const struct lu_env *env,
2775                               struct dt_object *dt, struct thandle *th)
2776 {
2777         struct osd_object  *obj = osd_dt_obj(dt);
2778         struct inode       *inode = obj->oo_inode;
2779         struct osd_thandle *oh;
2780         int                 rc = 0;
2781
2782         LINVRNT(osd_invariant(obj));
2783         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2784         LASSERT(osd_write_locked(env, obj));
2785         LASSERT(th != NULL);
2786
2787         oh = container_of0(th, struct osd_thandle, ot_super);
2788         LASSERT(oh->ot_handle != NULL);
2789
2790         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2791
2792         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2793                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2794         /*
2795          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2796          * (65000) subdirectories by storing "1" in i_nlink if the link count
2797          * would otherwise overflow. Directory tranversal tools understand
2798          * that (st_nlink == 1) indicates that the filesystem dose not track
2799          * hard links count on the directory, and will not abort subdirectory
2800          * scanning early once (st_nlink - 2) subdirs have been found.
2801          *
2802          * This also has to properly handle the case of inodes with nlink == 0
2803          * in case they are being linked into the PENDING directory
2804          */
2805         spin_lock(&obj->oo_guard);
2806         ldiskfs_inc_count(oh->ot_handle, inode);
2807         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2808         spin_unlock(&obj->oo_guard);
2809
2810         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2811         LINVRNT(osd_invariant(obj));
2812
2813         return rc;
2814 }
2815
2816 static int osd_declare_object_ref_del(const struct lu_env *env,
2817                                       struct dt_object *dt,
2818                                       struct thandle *handle)
2819 {
2820         struct osd_thandle *oh;
2821
2822         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2823         LASSERT(handle != NULL);
2824
2825         oh = container_of0(handle, struct osd_thandle, ot_super);
2826         LASSERT(oh->ot_handle == NULL);
2827
2828         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2829                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2830
2831         return 0;
2832 }
2833
2834 /*
2835  * Concurrency: @dt is write locked.
2836  */
2837 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2838                               struct thandle *th)
2839 {
2840         struct osd_object       *obj = osd_dt_obj(dt);
2841         struct inode            *inode = obj->oo_inode;
2842         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2843         struct osd_thandle      *oh;
2844
2845         LINVRNT(osd_invariant(obj));
2846         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2847         LASSERT(osd_write_locked(env, obj));
2848         LASSERT(th != NULL);
2849
2850         oh = container_of0(th, struct osd_thandle, ot_super);
2851         LASSERT(oh->ot_handle != NULL);
2852
2853         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2854
2855         spin_lock(&obj->oo_guard);
2856         /* That can be result of upgrade from old Lustre version and
2857          * applied only to local files.  Just skip this ref_del call.
2858          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2859         if (inode->i_nlink == 0) {
2860                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2861                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2862                              ", maybe an upgraded file? (LU-3915)\n",
2863                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2864                 spin_unlock(&obj->oo_guard);
2865                 return 0;
2866         }
2867
2868         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2869                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2870
2871         ldiskfs_dec_count(oh->ot_handle, inode);
2872         spin_unlock(&obj->oo_guard);
2873
2874         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2875         LINVRNT(osd_invariant(obj));
2876
2877         return 0;
2878 }
2879
2880 /*
2881  * Get the 64-bit version for an inode.
2882  */
2883 static int osd_object_version_get(const struct lu_env *env,
2884                                   struct dt_object *dt, dt_obj_version_t *ver)
2885 {
2886         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2887
2888         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2889                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2890         *ver = LDISKFS_I(inode)->i_fs_version;
2891         return 0;
2892 }
2893
2894 /*
2895  * Concurrency: @dt is read locked.
2896  */
2897 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2898                          struct lu_buf *buf, const char *name,
2899                          struct lustre_capa *capa)
2900 {
2901         struct osd_object      *obj    = osd_dt_obj(dt);
2902         struct inode           *inode  = obj->oo_inode;
2903         struct osd_thread_info *info   = osd_oti_get(env);
2904         struct dentry          *dentry = &info->oti_obj_dentry;
2905
2906         /* version get is not real XATTR but uses xattr API */
2907         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2908                 /* for version we are just using xattr API but change inode
2909                  * field instead */
2910                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2911                 osd_object_version_get(env, dt, buf->lb_buf);
2912                 return sizeof(dt_obj_version_t);
2913         }
2914
2915         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2916         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2917
2918         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2919                 return -EACCES;
2920
2921         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2922 }
2923
2924
2925 static int osd_declare_xattr_set(const struct lu_env *env,
2926                                  struct dt_object *dt,
2927                                  const struct lu_buf *buf, const char *name,
2928                                  int fl, struct thandle *handle)
2929 {
2930         struct osd_thandle *oh;
2931         int credits;
2932
2933         LASSERT(handle != NULL);
2934
2935         oh = container_of0(handle, struct osd_thandle, ot_super);
2936         LASSERT(oh->ot_handle == NULL);
2937
2938         /* optimistic optimization: LMA is set first and usually fit inode */
2939         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2940                 if (dt_object_exists(dt))
2941                         credits = 0;
2942                 else
2943                         credits = 1;
2944         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2945                 credits = 1;
2946         } else {
2947                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2948                 struct super_block *sb = osd_sb(osd);
2949                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2950                 if (buf && buf->lb_len > sb->s_blocksize) {
2951                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2952                                         sb->s_blocksize_bits;
2953                 }
2954         }
2955
2956         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2957
2958         return 0;
2959 }
2960
2961 /*
2962  * Set the 64-bit version for object
2963  */
2964 static void osd_object_version_set(const struct lu_env *env,
2965                                    struct dt_object *dt,
2966                                    dt_obj_version_t *new_version)
2967 {
2968         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2969
2970         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2971                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2972
2973         LDISKFS_I(inode)->i_fs_version = *new_version;
2974         /** Version is set after all inode operations are finished,
2975          *  so we should mark it dirty here */
2976         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2977 }
2978
2979 /*
2980  * Concurrency: @dt is write locked.
2981  */
2982 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2983                          const struct lu_buf *buf, const char *name, int fl,
2984                          struct thandle *handle, struct lustre_capa *capa)
2985 {
2986         struct osd_object      *obj      = osd_dt_obj(dt);
2987         struct inode           *inode    = obj->oo_inode;
2988         struct osd_thread_info *info     = osd_oti_get(env);
2989         int                     fs_flags = 0;
2990         ENTRY;
2991
2992         LASSERT(handle != NULL);
2993
2994         /* version set is not real XATTR */
2995         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2996                 /* for version we are just using xattr API but change inode
2997                  * field instead */
2998                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2999                 osd_object_version_set(env, dt, buf->lb_buf);
3000                 return sizeof(dt_obj_version_t);
3001         }
3002
3003         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3004                 return -EACCES;
3005
3006         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3007                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3008
3009         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3010         if (fl & LU_XATTR_REPLACE)
3011                 fs_flags |= XATTR_REPLACE;
3012
3013         if (fl & LU_XATTR_CREATE)
3014                 fs_flags |= XATTR_CREATE;
3015
3016         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3017                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3018                 int                      rc;
3019
3020                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3021                 if (rc != 0)
3022                         RETURN(rc);
3023
3024                 lma->lma_incompat |= LMAI_STRIPED;
3025                 lustre_lma_swab(lma);
3026                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3027                                      sizeof(*lma), XATTR_REPLACE);
3028                 if (rc != 0)
3029                         RETURN(rc);
3030         }
3031
3032         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3033                                fs_flags);
3034 }
3035
3036 /*
3037  * Concurrency: @dt is read locked.
3038  */
3039 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3040                           struct lu_buf *buf, struct lustre_capa *capa)
3041 {
3042         struct osd_object      *obj    = osd_dt_obj(dt);
3043         struct inode           *inode  = obj->oo_inode;
3044         struct osd_thread_info *info   = osd_oti_get(env);
3045         struct dentry          *dentry = &info->oti_obj_dentry;
3046
3047         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3048         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3049
3050         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3051                 return -EACCES;
3052
3053         dentry->d_inode = inode;
3054         dentry->d_sb = inode->i_sb;
3055         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3056 }
3057
3058 static int osd_declare_xattr_del(const struct lu_env *env,
3059                                  struct dt_object *dt, const char *name,
3060                                  struct thandle *handle)
3061 {
3062         struct osd_thandle *oh;
3063
3064         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3065         LASSERT(handle != NULL);
3066
3067         oh = container_of0(handle, struct osd_thandle, ot_super);
3068         LASSERT(oh->ot_handle == NULL);
3069
3070         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3071                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3072
3073         return 0;
3074 }
3075
3076 /*
3077  * Concurrency: @dt is write locked.
3078  */
3079 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3080                          const char *name, struct thandle *handle,
3081                          struct lustre_capa *capa)
3082 {
3083         struct osd_object      *obj    = osd_dt_obj(dt);
3084         struct inode           *inode  = obj->oo_inode;
3085         struct osd_thread_info *info   = osd_oti_get(env);
3086         struct dentry          *dentry = &info->oti_obj_dentry;
3087         int                     rc;
3088
3089         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3090         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3091         LASSERT(handle != NULL);
3092
3093         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3094                 return -EACCES;
3095
3096         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3097
3098         ll_vfs_dq_init(inode);
3099         dentry->d_inode = inode;
3100         dentry->d_sb = inode->i_sb;
3101         rc = inode->i_op->removexattr(dentry, name);
3102         return rc;
3103 }
3104
3105 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3106                                      struct dt_object *dt,
3107                                      struct lustre_capa *old, __u64 opc)
3108 {
3109         struct osd_thread_info *info = osd_oti_get(env);
3110         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3111         struct osd_object *obj = osd_dt_obj(dt);
3112         struct osd_device *osd = osd_obj2dev(obj);
3113         struct lustre_capa_key *key = &info->oti_capa_key;
3114         struct lustre_capa *capa = &info->oti_capa;
3115         struct obd_capa *oc;
3116         struct lu_capainfo *lci;
3117         int rc;
3118         ENTRY;
3119
3120         if (!osd->od_fl_capa)
3121                 RETURN(ERR_PTR(-ENOENT));
3122
3123         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3124         LINVRNT(osd_invariant(obj));
3125
3126         /* renewal sanity check */
3127         if (old && osd_object_auth(env, dt, old, opc))
3128                 RETURN(ERR_PTR(-EACCES));
3129
3130         lci = lu_capainfo_get(env);
3131         if (unlikely(lci == NULL))
3132                 RETURN(ERR_PTR(-ENOENT));
3133
3134         switch (lci->lci_auth) {
3135         case LC_ID_NONE:
3136                 RETURN(NULL);
3137         case LC_ID_PLAIN:
3138                 capa->lc_uid = i_uid_read(obj->oo_inode);
3139                 capa->lc_gid = i_gid_read(obj->oo_inode);
3140                 capa->lc_flags = LC_ID_PLAIN;
3141                 break;
3142         case LC_ID_CONVERT: {
3143                 __u32 d[4], s[4];
3144
3145                 s[0] = i_uid_read(obj->oo_inode);
3146                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3147                 s[2] = i_uid_read(obj->oo_inode);
3148                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3149                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3150                 if (unlikely(rc))
3151                         RETURN(ERR_PTR(rc));
3152
3153                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3154                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3155                 capa->lc_flags = LC_ID_CONVERT;
3156                 break;
3157         }
3158         default:
3159                 RETURN(ERR_PTR(-EINVAL));
3160         }
3161
3162         capa->lc_fid = *fid;
3163         capa->lc_opc = opc;
3164         capa->lc_flags |= osd->od_capa_alg << 24;
3165         capa->lc_timeout = osd->od_capa_timeout;
3166         capa->lc_expiry = 0;
3167
3168         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3169         if (oc) {
3170                 LASSERT(!capa_is_expired(oc));
3171                 RETURN(oc);
3172         }
3173
3174         spin_lock(&capa_lock);
3175         *key = osd->od_capa_keys[1];
3176         spin_unlock(&capa_lock);
3177
3178         capa->lc_keyid = key->lk_keyid;
3179         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3180
3181         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3182         if (rc) {
3183                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3184                 RETURN(ERR_PTR(rc));
3185         }
3186
3187         oc = capa_add(osd->od_capa_hash, capa);
3188         RETURN(oc);
3189 }
3190
3191 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3192                            __u64 start, __u64 end)
3193 {
3194         struct osd_object       *obj    = osd_dt_obj(dt);
3195         struct inode            *inode  = obj->oo_inode;
3196         struct osd_thread_info  *info   = osd_oti_get(env);
3197         struct dentry           *dentry = &info->oti_obj_dentry;
3198         struct file             *file   = &info->oti_file;
3199         int                     rc;
3200
3201         ENTRY;
3202
3203         dentry->d_inode = inode;
3204         dentry->d_sb = inode->i_sb;
3205         file->f_dentry = dentry;
3206         file->f_mapping = inode->i_mapping;
3207         file->f_op = inode->i_fop;
3208         set_file_inode(file, inode);
3209
3210 #ifdef HAVE_FILE_FSYNC_4ARGS
3211         rc = file->f_op->fsync(file, start, end, 0);
3212 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3213         mutex_lock(&inode->i_mutex);
3214         rc = file->f_op->fsync(file, 0);
3215         mutex_unlock(&inode->i_mutex);
3216 #else
3217         mutex_lock(&inode->i_mutex);
3218         rc = file->f_op->fsync(file, dentry, 0);
3219         mutex_unlock(&inode->i_mutex);
3220 #endif
3221
3222         RETURN(rc);
3223 }
3224
3225 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3226                         void **data)
3227 {
3228         struct osd_object *obj = osd_dt_obj(dt);
3229         ENTRY;
3230
3231         *data = (void *)obj->oo_inode;
3232         RETURN(0);
3233 }
3234
3235 /*
3236  * Index operations.
3237  */
3238
3239 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3240                            const struct dt_index_features *feat)
3241 {
3242         struct iam_descr *descr;
3243
3244         if (osd_object_is_root(o))
3245                 return feat == &dt_directory_features;
3246
3247         LASSERT(o->oo_dir != NULL);
3248
3249         descr = o->oo_dir->od_container.ic_descr;
3250         if (feat == &dt_directory_features) {
3251                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3252                         return 1;
3253                 else
3254                         return 0;
3255         } else {
3256                 return
3257                         feat->dif_keysize_min <= descr->id_key_size &&
3258                         descr->id_key_size <= feat->dif_keysize_max &&
3259                         feat->dif_recsize_min <= descr->id_rec_size &&
3260                         descr->id_rec_size <= feat->dif_recsize_max &&
3261                         !(feat->dif_flags & (DT_IND_VARKEY |
3262                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3263                         ergo(feat->dif_flags & DT_IND_UPDATE,
3264                              1 /* XXX check that object (and file system) is
3265                                 * writable */);
3266         }
3267 }
3268
3269 static int osd_iam_container_init(const struct lu_env *env,
3270                                   struct osd_object *obj,
3271                                   struct osd_directory *dir)
3272 {
3273         struct iam_container *bag = &dir->od_container;
3274         int result;
3275
3276         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3277         if (result != 0)
3278                 return result;
3279
3280         result = iam_container_setup(bag);
3281         if (result == 0)
3282                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3283         else
3284                 iam_container_fini(bag);
3285
3286         return result;
3287 }
3288
3289
3290 /*
3291  * Concurrency: no external locking is necessary.
3292  */
3293 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3294                          const struct dt_index_features *feat)
3295 {
3296         int                      result;
3297         int                      skip_iam = 0;
3298         struct osd_object       *obj = osd_dt_obj(dt);
3299
3300         LINVRNT(osd_invariant(obj));
3301
3302         if (osd_object_is_root(obj)) {
3303                 dt->do_index_ops = &osd_index_ea_ops;
3304                 result = 0;
3305         } else if (feat == &dt_directory_features) {
3306                 dt->do_index_ops = &osd_index_ea_ops;
3307                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3308                         result = 0;
3309                 else
3310                         result = -ENOTDIR;
3311                 skip_iam = 1;
3312         } else if (unlikely(feat == &dt_otable_features)) {
3313                 dt->do_index_ops = &osd_otable_ops;
3314                 return 0;
3315         } else if (unlikely(feat == &dt_acct_features)) {
3316                 dt->do_index_ops = &osd_acct_index_ops;
3317                 result = 0;
3318                 skip_iam = 1;
3319         } else if (!osd_has_index(obj)) {
3320                 struct osd_directory *dir;
3321
3322                 OBD_ALLOC_PTR(dir);
3323                 if (dir != NULL) {
3324
3325                         spin_lock(&obj->oo_guard);
3326                         if (obj->oo_dir == NULL)
3327                                 obj->oo_dir = dir;
3328                         else
3329                                 /*
3330                                  * Concurrent thread allocated container data.
3331                                  */
3332                                 OBD_FREE_PTR(dir);
3333                         spin_unlock(&obj->oo_guard);
3334                         /*
3335                          * Now, that we have container data, serialize its
3336                          * initialization.
3337                          */
3338                         down_write(&obj->oo_ext_idx_sem);
3339                         /*
3340                          * recheck under lock.
3341                          */
3342                         if (!osd_has_index(obj))
3343                                 result = osd_iam_container_init(env, obj, dir);
3344                         else
3345                                 result = 0;
3346                         up_write(&obj->oo_ext_idx_sem);
3347                 } else {
3348                         result = -ENOMEM;
3349                 }
3350         } else {
3351                 result = 0;
3352         }
3353
3354         if (result == 0 && skip_iam == 0) {
3355                 if (!osd_iam_index_probe(env, obj, feat))
3356                         result = -ENOTDIR;
3357         }
3358         LINVRNT(osd_invariant(obj));
3359
3360         if (result == 0 && is_quota_glb_feat(feat) &&
3361             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3362                 result = osd_quota_migration(env, dt, feat);
3363
3364         return result;
3365 }
3366
3367 static int osd_otable_it_attr_get(const struct lu_env *env,
3368                                  struct dt_object *dt,
3369                                  struct lu_attr *attr,
3370                                  struct lustre_capa *capa)
3371 {
3372         attr->la_valid = 0;
3373         return 0;
3374 }
3375
3376 static const struct dt_object_operations osd_obj_ops = {
3377         .do_read_lock         = osd_object_read_lock,
3378         .do_write_lock        = osd_object_write_lock,
3379         .do_read_unlock       = osd_object_read_unlock,
3380         .do_write_unlock      = osd_object_write_unlock,
3381         .do_write_locked      = osd_object_write_locked,
3382         .do_attr_get          = osd_attr_get,
3383         .do_declare_attr_set  = osd_declare_attr_set,
3384         .do_attr_set          = osd_attr_set,
3385         .do_ah_init           = osd_ah_init,
3386         .do_declare_create    = osd_declare_object_create,
3387         .do_create            = osd_object_create,
3388         .do_declare_destroy   = osd_declare_object_destroy,
3389         .do_destroy           = osd_object_destroy,
3390         .do_index_try         = osd_index_try,
3391         .do_declare_ref_add   = osd_declare_object_ref_add,
3392         .do_ref_add           = osd_object_ref_add,
3393         .do_declare_ref_del   = osd_declare_object_ref_del,
3394         .do_ref_del           = osd_object_ref_del,
3395         .do_xattr_get         = osd_xattr_get,
3396         .do_declare_xattr_set = osd_declare_xattr_set,
3397         .do_xattr_set         = osd_xattr_set,
3398         .do_declare_xattr_del = osd_declare_xattr_del,
3399         .do_xattr_del         = osd_xattr_del,
3400         .do_xattr_list        = osd_xattr_list,
3401         .do_capa_get          = osd_capa_get,
3402         .do_object_sync       = osd_object_sync,
3403         .do_data_get          = osd_data_get,
3404 };
3405
3406 /**
3407  * dt_object_operations for interoperability mode
3408  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3409  */
3410 static const struct dt_object_operations osd_obj_ea_ops = {
3411         .do_read_lock         = osd_object_read_lock,
3412         .do_write_lock        = osd_object_write_lock,
3413         .do_read_unlock       = osd_object_read_unlock,
3414         .do_write_unlock      = osd_object_write_unlock,
3415         .do_write_locked      = osd_object_write_locked,
3416         .do_attr_get          = osd_attr_get,
3417         .do_declare_attr_set  = osd_declare_attr_set,
3418         .do_attr_set          = osd_attr_set,
3419         .do_ah_init           = osd_ah_init,
3420         .do_declare_create    = osd_declare_object_create,
3421         .do_create            = osd_object_ea_create,
3422         .do_declare_destroy   = osd_declare_object_destroy,
3423         .do_destroy           = osd_object_destroy,
3424         .do_index_try         = osd_index_try,
3425         .do_declare_ref_add   = osd_declare_object_ref_add,
3426         .do_ref_add           = osd_object_ref_add,
3427         .do_declare_ref_del   = osd_declare_object_ref_del,
3428         .do_ref_del           = osd_object_ref_del,
3429         .do_xattr_get         = osd_xattr_get,
3430         .do_declare_xattr_set = osd_declare_xattr_set,
3431         .do_xattr_set         = osd_xattr_set,
3432         .do_declare_xattr_del = osd_declare_xattr_del,
3433         .do_xattr_del         = osd_xattr_del,
3434         .do_xattr_list        = osd_xattr_list,
3435         .do_capa_get          = osd_capa_get,
3436         .do_object_sync       = osd_object_sync,
3437         .do_data_get          = osd_data_get,
3438 };
3439
3440 static const struct dt_object_operations osd_obj_otable_it_ops = {
3441         .do_attr_get    = osd_otable_it_attr_get,
3442         .do_index_try   = osd_index_try,
3443 };
3444
3445 static int osd_index_declare_iam_delete(const struct lu_env *env,
3446                                         struct dt_object *dt,
3447                                         const struct dt_key *key,
3448                                         struct thandle *handle)
3449 {
3450         struct osd_thandle    *oh;
3451
3452         oh = container_of0(handle, struct osd_thandle, ot_super);
3453         LASSERT(oh->ot_handle == NULL);
3454
3455         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3456                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3457
3458         return 0;
3459 }
3460
3461 /**
3462  *      delete a (key, value) pair from index \a dt specified by \a key
3463  *
3464  *      \param  dt      osd index object
3465  *      \param  key     key for index
3466  *      \param  rec     record reference
3467  *      \param  handle  transaction handler
3468  *
3469  *      \retval  0  success
3470  *      \retval -ve   failure
3471  */
3472
3473 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3474                                 const struct dt_key *key,
3475                                 struct thandle *handle,
3476                                 struct lustre_capa *capa)
3477 {
3478         struct osd_thread_info *oti = osd_oti_get(env);
3479         struct osd_object      *obj = osd_dt_obj(dt);
3480         struct osd_thandle     *oh;
3481         struct iam_path_descr  *ipd;
3482         struct iam_container   *bag = &obj->oo_dir->od_container;
3483         int                     rc;
3484
3485         ENTRY;
3486
3487         LINVRNT(osd_invariant(obj));
3488         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3489         LASSERT(bag->ic_object == obj->oo_inode);
3490         LASSERT(handle != NULL);
3491
3492         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3493                 RETURN(-EACCES);
3494
3495         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3496
3497         ipd = osd_idx_ipd_get(env, bag);
3498         if (unlikely(ipd == NULL))
3499                 RETURN(-ENOMEM);
3500
3501         oh = container_of0(handle, struct osd_thandle, ot_super);
3502         LASSERT(oh->ot_handle != NULL);
3503         LASSERT(oh->ot_handle->h_transaction != NULL);
3504
3505         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3506                 /* swab quota uid/gid provided by caller */
3507                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3508                 key = (const struct dt_key *)&oti->oti_quota_id;
3509         }
3510
3511         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3512         osd_ipd_put(env, bag, ipd);
3513         LINVRNT(osd_invariant(obj));
3514         RETURN(rc);
3515 }
3516
3517 static int osd_index_declare_ea_delete(const struct lu_env *env,
3518                                        struct dt_object *dt,
3519                                        const struct dt_key *key,
3520                                        struct thandle *handle)
3521 {
3522         struct osd_thandle *oh;
3523         struct inode       *inode;
3524         int                 rc;
3525         ENTRY;
3526
3527         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3528         LASSERT(handle != NULL);
3529
3530         oh = container_of0(handle, struct osd_thandle, ot_super);
3531         LASSERT(oh->ot_handle == NULL);
3532
3533         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3534                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3535
3536         inode = osd_dt_obj(dt)->oo_inode;
3537         LASSERT(inode);
3538
3539         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3540                                    0, oh, true, true, NULL, false);
3541         RETURN(rc);
3542 }
3543
3544 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3545                                           struct dt_rec *fid)
3546 {
3547         struct osd_fid_pack *rec;
3548         int                  rc = -ENODATA;
3549
3550         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3551                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3552                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3553         }
3554         return rc;
3555 }
3556
3557 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3558                           struct lu_fid *fid)
3559 {
3560         ENTRY;
3561
3562         /* FID seqs not in FLDB, must be local seq */
3563         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3564                 RETURN(0);
3565
3566         if (osd_seq_exists(env, osd, fid_seq(fid)))
3567                 RETURN(0);
3568
3569         RETURN(1);
3570 }
3571
3572 /**
3573  * Index delete function for interoperability mode (b11826).
3574  * It will remove the directory entry added by osd_index_ea_insert().
3575  * This entry is needed to maintain name->fid mapping.
3576  *
3577  * \param key,  key i.e. file entry to be deleted
3578  *
3579  * \retval   0, on success
3580  * \retval -ve, on error
3581  */
3582 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3583                                const struct dt_key *key,
3584                                struct thandle *handle,
3585                                struct lustre_capa *capa)
3586 {
3587         struct osd_object          *obj    = osd_dt_obj(dt);
3588         struct inode               *dir    = obj->oo_inode;
3589         struct dentry              *dentry;
3590         struct osd_thandle         *oh;
3591         struct ldiskfs_dir_entry_2 *de = NULL;
3592         struct buffer_head         *bh;
3593         struct htree_lock          *hlock = NULL;
3594         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3595         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3596         int                        rc;
3597         ENTRY;
3598
3599         LINVRNT(osd_invariant(obj));
3600         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3601         LASSERT(handle != NULL);
3602
3603         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3604
3605         oh = container_of(handle, struct osd_thandle, ot_super);
3606         LASSERT(oh->ot_handle != NULL);
3607         LASSERT(oh->ot_handle->h_transaction != NULL);
3608
3609         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3610                 RETURN(-EACCES);
3611
3612         ll_vfs_dq_init(dir);
3613         dentry = osd_child_dentry_get(env, obj,
3614                                       (char *)key, strlen((char *)key));
3615
3616         if (obj->oo_hl_head != NULL) {
3617                 hlock = osd_oti_get(env)->oti_hlock;
3618                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3619                                    dir, LDISKFS_HLOCK_DEL);
3620         } else {
3621                 down_write(&obj->oo_ext_idx_sem);
3622         }
3623
3624         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3625         if (bh) {
3626                 __u32 ino = 0;
3627
3628                 /* If this is not the ".." entry, it might be a remote DNE
3629                  * entry and  we need to check if the FID is for a remote
3630                  * MDT.  If the FID is  not in the directory entry (e.g.
3631                  * upgraded 1.8 filesystem without dirdata enabled) then
3632                  * we need to get the FID from the LMA. For a remote directory
3633                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3634                  * case.
3635                  *
3636                  * Delete the entry before the agent inode in order to
3637                  * simplify error handling.  At worst an error after deleting
3638                  * the entry first might leak the agent inode afterward. The
3639                  * reverse would need filesystem abort in case of error deleting
3640                  * the entry after the agent had been removed, or leave a
3641                  * dangling entry pointing at a random inode. */
3642                 if (strcmp((char *)key, dotdot) != 0) {
3643                         LASSERT(de != NULL);
3644                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3645                         /* If Fid is not in dentry, try to get it from LMA */
3646                         if (rc == -ENODATA) {
3647                                 struct osd_inode_id *id;
3648                                 struct inode *inode;
3649
3650                                 /* Before trying to get fid from the inode,
3651                                  * check whether the inode is valid.
3652                                  *
3653                                  * If the inode has been deleted, do not go
3654                                  * ahead to do osd_ea_fid_get, which will set
3655                                  * the inode to bad inode, which might cause
3656                                  * the inode to be deleted uncorrectly */
3657                                 inode = ldiskfs_iget(osd_sb(osd),
3658                                                      le32_to_cpu(de->inode));
3659                                 if (IS_ERR(inode)) {
3660                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3661                                                "error.\n", osd_name(osd),
3662                                                PFID(fid));
3663                                         rc = PTR_ERR(inode);
3664                                 } else {
3665                                         if (likely(inode->i_nlink != 0)) {
3666                                                 id = &osd_oti_get(env)->oti_id;
3667                                                 rc = osd_ea_fid_get(env, obj,
3668                                                         le32_to_cpu(de->inode),
3669                                                                     fid, id);
3670                                         } else {
3671                                                 CDEBUG(D_INFO, "%s: %u "DFID
3672                                                        "deleted.\n",
3673                                                        osd_name(osd),
3674                                                        le32_to_cpu(de->inode),
3675                                                        PFID(fid));
3676                                                 rc = -ESTALE;
3677                                         }
3678                                         iput(inode);
3679                                 }
3680                         }
3681                         if (rc == 0 &&
3682                             unlikely(osd_remote_fid(env, osd, fid)))
3683                                 /* Need to delete agent inode */
3684                                 ino = le32_to_cpu(de->inode);
3685                 }
3686                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3687                 brelse(bh);
3688                 if (rc == 0 && unlikely(ino != 0)) {
3689                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3690                                                           oh);
3691                         if (rc != 0)
3692                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3693                                        osd_name(osd), PFID(fid), rc);
3694                 }
3695         } else {
3696                 rc = -ENOENT;
3697         }
3698         if (hlock != NULL)
3699                 ldiskfs_htree_unlock(hlock);
3700         else
3701                 up_write(&obj->oo_ext_idx_sem);
3702
3703         if (rc != 0)
3704                 GOTO(out, rc);
3705
3706         /* For inode on the remote MDT, .. will point to
3707          * /Agent directory, Check whether it needs to delete
3708          * from agent directory */
3709         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3710                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3711                                                    oh);
3712                 if (rc != 0 && rc != -ENOENT) {
3713                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3714                                osd_name(osd), PFID(fid), rc);
3715                 }
3716
3717                 if (rc == -ENOENT)
3718                         rc = 0;
3719
3720                 GOTO(out, rc);
3721         }
3722 out:
3723
3724         LASSERT(osd_invariant(obj));
3725         RETURN(rc);
3726 }
3727
3728 /**
3729  *      Lookup index for \a key and copy record to \a rec.
3730  *
3731  *      \param  dt      osd index object
3732  *      \param  key     key for index
3733  *      \param  rec     record reference
3734  *
3735  *      \retval  +ve  success : exact mach
3736  *      \retval  0    return record with key not greater than \a key
3737  *      \retval -ve   failure
3738  */
3739 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3740                                 struct dt_rec *rec, const struct dt_key *key,
3741                                 struct lustre_capa *capa)
3742 {
3743         struct osd_object      *obj = osd_dt_obj(dt);
3744         struct iam_path_descr  *ipd;
3745         struct iam_container   *bag = &obj->oo_dir->od_container;
3746         struct osd_thread_info *oti = osd_oti_get(env);
3747         struct iam_iterator    *it = &oti->oti_idx_it;
3748         struct iam_rec         *iam_rec;
3749         int                     rc;
3750
3751         ENTRY;
3752
3753         LASSERT(osd_invariant(obj));
3754         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3755         LASSERT(bag->ic_object == obj->oo_inode);
3756
3757         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3758                 RETURN(-EACCES);
3759
3760         ipd = osd_idx_ipd_get(env, bag);
3761         if (IS_ERR(ipd))
3762                 RETURN(-ENOMEM);
3763
3764         /* got ipd now we can start iterator. */
3765         iam_it_init(it, bag, 0, ipd);
3766
3767         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3768                 /* swab quota uid/gid provided by caller */
3769                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3770                 key = (const struct dt_key *)&oti->oti_quota_id;
3771         }
3772
3773         rc = iam_it_get(it, (struct iam_key *)key);
3774         if (rc >= 0) {
3775                 if (S_ISDIR(obj->oo_inode->i_mode))
3776                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3777                 else
3778                         iam_rec = (struct iam_rec *) rec;
3779
3780                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3781
3782                 if (S_ISDIR(obj->oo_inode->i_mode))
3783                         osd_fid_unpack((struct lu_fid *) rec,
3784                                        (struct osd_fid_pack *)iam_rec);
3785                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3786                         osd_quota_unpack(obj, rec);
3787         }
3788
3789         iam_it_put(it);
3790         iam_it_fini(it);
3791         osd_ipd_put(env, bag, ipd);
3792
3793         LINVRNT(osd_invariant(obj));
3794
3795         RETURN(rc);
3796 }
3797
3798 static int osd_index_declare_iam_insert(const struct lu_env *env,
3799                                         struct dt_object *dt,
3800                                         const struct dt_rec *rec,
3801                                         const struct dt_key *key,
3802                                         struct thandle *handle)
3803 {
3804         struct osd_thandle *oh;
3805
3806         LASSERT(handle != NULL);
3807
3808         oh = container_of0(handle, struct osd_thandle, ot_super);
3809         LASSERT(oh->ot_handle == NULL);
3810
3811         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3812                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3813
3814         return 0;
3815 }
3816
3817 /**
3818  *      Inserts (key, value) pair in \a dt index object.
3819  *
3820  *      \param  dt      osd index object
3821  *      \param  key     key for index
3822  *      \param  rec     record reference
3823  *      \param  th      transaction handler
3824  *
3825  *      \retval  0  success
3826  *      \retval -ve failure
3827  */
3828 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3829                                 const struct dt_rec *rec,
3830                                 const struct dt_key *key, struct thandle *th,
3831                                 struct lustre_capa *capa, int ignore_quota)
3832 {
3833         struct osd_object     *obj = osd_dt_obj(dt);
3834         struct iam_path_descr *ipd;
3835         struct osd_thandle    *oh;
3836         struct iam_container  *bag = &obj->oo_dir->od_container;
3837         struct osd_thread_info *oti = osd_oti_get(env);
3838         struct iam_rec         *iam_rec;
3839         int                     rc;
3840
3841         ENTRY;
3842
3843         LINVRNT(osd_invariant(obj));
3844         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3845         LASSERT(bag->ic_object == obj->oo_inode);
3846         LASSERT(th != NULL);
3847
3848         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3849                 RETURN(-EACCES);
3850
3851         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3852
3853         ipd = osd_idx_ipd_get(env, bag);
3854         if (unlikely(ipd == NULL))
3855                 RETURN(-ENOMEM);
3856
3857         oh = container_of0(th, struct osd_thandle, ot_super);
3858         LASSERT(oh->ot_handle != NULL);
3859         LASSERT(oh->ot_handle->h_transaction != NULL);
3860         if (S_ISDIR(obj->oo_inode->i_mode)) {
3861                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3862                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3863         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3864                 /* pack quota uid/gid */
3865                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3866                 key = (const struct dt_key *)&oti->oti_quota_id;
3867                 /* pack quota record */
3868                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3869                 iam_rec = (struct iam_rec *)rec;
3870         } else {
3871                 iam_rec = (struct iam_rec *)rec;
3872         }
3873
3874         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3875                         iam_rec, ipd);
3876         osd_ipd_put(env, bag, ipd);
3877         LINVRNT(osd_invariant(obj));
3878         RETURN(rc);
3879 }
3880
3881 /**
3882  * Calls ldiskfs_add_entry() to add directory entry
3883  * into the directory. This is required for
3884  * interoperability mode (b11826)
3885  *
3886  * \retval   0, on success
3887  * \retval -ve, on error
3888  */
3889 static int __osd_ea_add_rec(struct osd_thread_info *info,
3890                             struct osd_object *pobj, struct inode  *cinode,
3891                             const char *name, const struct dt_rec *fid,
3892                             struct htree_lock *hlock, struct thandle *th)
3893 {
3894         struct ldiskfs_dentry_param *ldp;
3895         struct dentry               *child;
3896         struct osd_thandle          *oth;
3897         int                          rc;
3898
3899         oth = container_of(th, struct osd_thandle, ot_super);
3900         LASSERT(oth->ot_handle != NULL);
3901         LASSERT(oth->ot_handle->h_transaction != NULL);
3902         LASSERT(pobj->oo_inode);
3903
3904         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3905         if (unlikely(pobj->oo_inode ==
3906                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3907                 ldp->edp_magic = 0;
3908         else
3909                 osd_get_ldiskfs_dirent_param(ldp, fid);
3910         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3911         child->d_fsdata = (void *)ldp;
3912         ll_vfs_dq_init(pobj->oo_inode);
3913         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3914
3915         RETURN(rc);
3916 }
3917
3918 /**
3919  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3920  * into the directory.Also sets flags into osd object to
3921  * indicate dot and dotdot are created. This is required for
3922  * interoperability mode (b11826)
3923  *
3924  * \param dir   directory for dot and dotdot fixup.
3925  * \param obj   child object for linking
3926  *
3927  * \retval   0, on success
3928  * \retval -ve, on error
3929  */
3930 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3931                               struct osd_object *dir,
3932                               struct inode  *parent_dir, const char *name,
3933                               const struct dt_rec *dot_fid,
3934                               const struct dt_rec *dot_dot_fid,
3935                               struct thandle *th)
3936 {
3937         struct inode                *inode = dir->oo_inode;
3938         struct osd_thandle          *oth;
3939         int result = 0;
3940
3941         oth = container_of(th, struct osd_thandle, ot_super);
3942         LASSERT(oth->ot_handle->h_transaction != NULL);
3943         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3944
3945         if (strcmp(name, dot) == 0) {
3946                 if (dir->oo_compat_dot_created) {
3947                         result = -EEXIST;
3948                 } else {
3949                         LASSERT(inode == parent_dir);
3950                         dir->oo_compat_dot_created = 1;
3951                         result = 0;
3952                 }
3953         } else if (strcmp(name, dotdot) == 0) {
3954                 if (!dir->oo_compat_dot_created)
3955                         return -EINVAL;
3956                 /* in case of rename, dotdot is already created */
3957                 if (dir->oo_compat_dotdot_created) {
3958                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3959                                                 dot_dot_fid, NULL, th);
3960                 }
3961
3962                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3963                                                 parent_dir, dot_fid,
3964                                                 dot_dot_fid, oth);
3965                 if (result == 0)
3966                         dir->oo_compat_dotdot_created = 1;
3967         }
3968
3969         return result;
3970 }
3971
3972
3973 /**
3974  * It will call the appropriate osd_add* function and return the
3975  * value, return by respective functions.
3976  */
3977 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3978                           struct inode *cinode, const char *name,
3979                           const struct dt_rec *fid, struct thandle *th)
3980 {
3981         struct osd_thread_info *info   = osd_oti_get(env);
3982         struct htree_lock      *hlock;
3983         int                     rc;
3984
3985         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3986
3987         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3988                                                    name[2] =='\0'))) {
3989                 if (hlock != NULL) {
3990                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3991                                            pobj->oo_inode, 0);
3992                 } else {
3993                         down_write(&pobj->oo_ext_idx_sem);
3994                 }
3995                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3996                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3997                                         fid, th);
3998         } else {
3999                 if (hlock != NULL) {
4000                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4001                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4002                 } else {
4003                         down_write(&pobj->oo_ext_idx_sem);
4004                 }
4005
4006                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4007                         struct lu_fid *tfid = &info->oti_fid;
4008
4009                         *tfid = *(const struct lu_fid *)fid;
4010                         tfid->f_ver = ~0;
4011                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4012                                               (const struct dt_rec *)tfid,
4013                                               hlock, th);
4014                 } else {
4015                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4016                                               hlock, th);
4017                 }
4018         }
4019         if (hlock != NULL)
4020                 ldiskfs_htree_unlock(hlock);
4021         else
4022                 up_write(&pobj->oo_ext_idx_sem);
4023
4024         return rc;
4025 }
4026
4027 static void
4028 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4029                       struct osd_idmap_cache *oic)
4030 {
4031         struct osd_scrub    *scrub = &dev->od_scrub;
4032         struct lu_fid       *fid   = &oic->oic_fid;
4033         struct osd_inode_id *id    = &oti->oti_id;
4034         int                  once  = 0;
4035         int                  rc;
4036         ENTRY;
4037
4038         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4039                 RETURN_EXIT;
4040
4041 again:
4042         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4043         if (rc != 0 && rc != -ENOENT)
4044                 RETURN_EXIT;
4045
4046         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4047                 RETURN_EXIT;
4048
4049         if (thread_is_running(&scrub->os_thread)) {
4050                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4051                 /* There is race condition between osd_oi_lookup and OI scrub.
4052                  * The OI scrub finished just after osd_oi_lookup() failure.
4053                  * Under such case, it is unnecessary to trigger OI scrub again,
4054                  * but try to call osd_oi_lookup() again. */
4055                 if (unlikely(rc == -EAGAIN))
4056                         goto again;
4057
4058                 RETURN_EXIT;
4059         }
4060
4061         if (!dev->od_noscrub && ++once == 1) {
4062                 rc = osd_scrub_start(dev);
4063                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4064                               ", rc = %d [2]\n",
4065                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4066                               PFID(fid), rc);
4067                 if (rc == 0)
4068                         goto again;
4069         }
4070
4071         EXIT;
4072 }
4073
4074 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4075                                struct osd_device *dev,
4076                                struct osd_idmap_cache *oic,
4077                                struct lu_fid *fid, __u32 ino)
4078 {
4079         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4080         struct inode            *inode;
4081         int                      rc;
4082
4083         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4084         inode = osd_iget(oti, dev, &oic->oic_lid);
4085         if (IS_ERR(inode)) {
4086                 fid_zero(&oic->oic_fid);
4087                 return PTR_ERR(inode);
4088         }
4089
4090         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4091         iput(inode);
4092         if (rc != 0)
4093                 fid_zero(&oic->oic_fid);
4094         else
4095                 *fid = oic->oic_fid = lma->lma_self_fid;
4096         return rc;
4097 }
4098
4099 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4100                      struct osd_inode_id *id, const struct lu_fid *fid)
4101 {
4102         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4103                id->oii_ino, id->oii_gen, info);
4104         info->oti_cache.oic_lid = *id;
4105         info->oti_cache.oic_fid = *fid;
4106         info->oti_cache.oic_dev = osd;
4107
4108         return 0;
4109 }
4110
4111 /**
4112  * Calls ->lookup() to find dentry. From dentry get inode and
4113  * read inode's ea to get fid. This is required for  interoperability
4114  * mode (b11826)
4115  *
4116  * \retval   0, on success
4117  * \retval -ve, on error
4118  */
4119 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4120                              struct dt_rec *rec, const struct dt_key *key)
4121 {
4122         struct inode               *dir    = obj->oo_inode;
4123         struct dentry              *dentry;
4124         struct ldiskfs_dir_entry_2 *de;
4125         struct buffer_head         *bh;
4126         struct lu_fid              *fid = (struct lu_fid *) rec;
4127         struct htree_lock          *hlock = NULL;
4128         int                         ino;
4129         int                         rc;
4130         ENTRY;
4131
4132         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4133
4134         dentry = osd_child_dentry_get(env, obj,
4135                                       (char *)key, strlen((char *)key));
4136
4137         if (obj->oo_hl_head != NULL) {
4138                 hlock = osd_oti_get(env)->oti_hlock;
4139                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4140                                    dir, LDISKFS_HLOCK_LOOKUP);
4141         } else {
4142                 down_read(&obj->oo_ext_idx_sem);
4143         }
4144
4145         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4146         if (bh) {
4147                 struct osd_thread_info *oti = osd_oti_get(env);
4148                 struct osd_inode_id *id = &oti->oti_id;
4149                 struct osd_idmap_cache *oic = &oti->oti_cache;
4150                 struct osd_device *dev = osd_obj2dev(obj);
4151                 struct osd_scrub *scrub = &dev->od_scrub;
4152                 struct scrub_file *sf = &scrub->os_file;
4153
4154                 ino = le32_to_cpu(de->inode);
4155                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4156                         brelse(bh);
4157                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4158                         GOTO(out, rc);
4159                 }
4160
4161                 rc = osd_get_fid_from_dentry(de, rec);
4162
4163                 /* done with de, release bh */
4164                 brelse(bh);
4165                 if (rc != 0)
4166                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
4167                 else
4168                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4169                 if (rc != 0) {
4170                         fid_zero(&oic->oic_fid);
4171                         GOTO(out, rc);
4172                 }
4173
4174                 if (osd_remote_fid(env, dev, fid))
4175                         GOTO(out, rc = 0);
4176
4177                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4178                                       fid);
4179                 if (rc != 0)
4180                         GOTO(out, rc);
4181                 if ((scrub->os_pos_current <= ino) &&
4182                     ((sf->sf_flags & SF_INCONSISTENT) ||
4183                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4184                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4185                                       sf->sf_oi_bitmap)))
4186                         osd_consistency_check(oti, dev, oic);
4187         } else {
4188                 rc = -ENOENT;
4189         }
4190
4191         GOTO(out, rc);
4192
4193 out:
4194         if (hlock != NULL)
4195                 ldiskfs_htree_unlock(hlock);
4196         else
4197                 up_read(&obj->oo_ext_idx_sem);
4198         return rc;
4199 }
4200
4201 /**
4202  * Find the osd object for given fid.
4203  *
4204  * \param fid need to find the osd object having this fid
4205  *
4206  * \retval osd_object on success
4207  * \retval        -ve on error
4208  */
4209 struct osd_object *osd_object_find(const struct lu_env *env,
4210                                    struct dt_object *dt,
4211                                    const struct lu_fid *fid)
4212 {
4213         struct lu_device  *ludev = dt->do_lu.lo_dev;
4214         struct osd_object *child = NULL;
4215         struct lu_object  *luch;
4216         struct lu_object  *lo;
4217
4218         /*
4219          * at this point topdev might not exist yet
4220          * (i.e. MGS is preparing profiles). so we can
4221          * not rely on topdev and instead lookup with
4222          * our device passed as topdev. this can't work
4223          * if the object isn't cached yet (as osd doesn't
4224          * allocate lu_header). IOW, the object must be
4225          * in the cache, otherwise lu_object_alloc() crashes
4226          * -bzzz
4227          */
4228         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4229                                  ludev : ludev->ld_site->ls_top_dev,
4230                                  fid, NULL);
4231         if (!IS_ERR(luch)) {
4232                 if (lu_object_exists(luch)) {
4233                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4234                         if (lo != NULL)
4235                                 child = osd_obj(lo);
4236                         else
4237                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4238                                                 "lu_object can't be located"
4239                                                 DFID"\n", PFID(fid));
4240
4241                         if (child == NULL) {
4242                                 lu_object_put(env, luch);
4243                                 CERROR("Unable to get osd_object\n");
4244                                 child = ERR_PTR(-ENOENT);
4245                         }
4246                 } else {
4247                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4248                                         "lu_object does not exists "DFID"\n",
4249                                         PFID(fid));
4250                         lu_object_put(env, luch);
4251                         child = ERR_PTR(-ENOENT);
4252                 }
4253         } else {
4254                 child = ERR_CAST(luch);
4255         }
4256
4257         return child;
4258 }
4259
4260 /**
4261  * Put the osd object once done with it.
4262  *
4263  * \param obj osd object that needs to be put
4264  */
4265 static inline void osd_object_put(const struct lu_env *env,
4266                                   struct osd_object *obj)
4267 {
4268         lu_object_put(env, &obj->oo_dt.do_lu);
4269 }
4270
4271 static int osd_index_declare_ea_insert(const struct lu_env *env,
4272                                        struct dt_object *dt,
4273                                        const struct dt_rec *rec,
4274                                        const struct dt_key *key,
4275                                        struct thandle *handle)
4276 {
4277         struct osd_thandle      *oh;
4278         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4279         struct lu_fid           *fid = (struct lu_fid *)rec;
4280         int                     rc;
4281         ENTRY;
4282
4283         LASSERT(!dt_object_remote(dt));
4284         LASSERT(handle != NULL);
4285
4286         oh = container_of0(handle, struct osd_thandle, ot_super);
4287         LASSERT(oh->ot_handle == NULL);
4288
4289         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4290                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4291
4292         if (osd_dt_obj(dt)->oo_inode != NULL) {
4293                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4294
4295                 /* We ignore block quota on meta pool (MDTs), so needn't
4296                  * calculate how many blocks will be consumed by this index
4297                  * insert */
4298                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4299                                            i_gid_read(inode), 0,
4300                                            oh, true, true, NULL, false);
4301         }
4302
4303         if (fid == NULL)
4304                 RETURN(0);
4305
4306         rc = osd_remote_fid(env, osd, fid);
4307         if (rc <= 0)
4308                 RETURN(rc);
4309
4310         rc = 0;
4311
4312         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4313                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4314         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4315                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4316         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4317                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4318
4319         RETURN(rc);
4320 }
4321
4322 /**
4323  * Index add function for interoperability mode (b11826).
4324  * It will add the directory entry.This entry is needed to
4325  * maintain name->fid mapping.
4326  *
4327  * \param key it is key i.e. file entry to be inserted
4328  * \param rec it is value of given key i.e. fid
4329  *
4330  * \retval   0, on success
4331  * \retval -ve, on error
4332  */
4333 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4334                                const struct dt_rec *rec,
4335                                const struct dt_key *key, struct thandle *th,
4336                                struct lustre_capa *capa, int ignore_quota)
4337 {
4338         struct osd_object       *obj = osd_dt_obj(dt);
4339         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4340         struct lu_fid           *fid = (struct lu_fid *) rec;
4341         const char              *name = (const char *)key;
4342         struct osd_thread_info  *oti   = osd_oti_get(env);
4343         struct osd_inode_id     *id    = &oti->oti_id;
4344         struct inode            *child_inode = NULL;
4345         struct osd_object       *child = NULL;
4346         int                     rc;
4347         ENTRY;
4348
4349         LASSERT(osd_invariant(obj));
4350         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4351         LASSERT(th != NULL);
4352
4353         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4354
4355         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4356                 RETURN(-EACCES);
4357
4358         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4359
4360         rc = osd_remote_fid(env, osd, fid);
4361         if (rc < 0) {
4362                 CERROR("%s: Can not find object "DFID" rc %d\n",
4363                        osd_name(osd), PFID(fid), rc);
4364                 RETURN(rc);
4365         }
4366
4367         if (rc == 1) {
4368                 /* Insert remote entry */
4369                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4370                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4371                         struct osd_thandle      *oh;
4372
4373                         /* If parent on remote MDT, we need put this object
4374                          * under AGENT */
4375                         oh = container_of(th, typeof(*oh), ot_super);
4376                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4377                         if (rc != 0) {
4378                                 CERROR("%s: add "DFID" error: rc = %d\n",
4379                                        osd_name(osd),
4380                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4381                                 RETURN(rc);
4382                         }
4383
4384                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4385                 } else {
4386                         child_inode = osd_create_local_agent_inode(env, osd,
4387                                                                    obj, fid,
4388                                                                    th);
4389                         if (IS_ERR(child_inode))
4390                                 RETURN(PTR_ERR(child_inode));
4391                 }
4392         } else {
4393                 /* Insert local entry */
4394                 child = osd_object_find(env, dt, fid);
4395                 if (IS_ERR(child)) {
4396                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4397                                osd_name(osd), PFID(fid),
4398                                id->oii_ino, id->oii_gen,
4399                                (int)PTR_ERR(child));
4400                         RETURN(PTR_ERR(child));
4401                 }
4402                 child_inode = igrab(child->oo_inode);
4403         }
4404
4405         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4406
4407         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4408                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4409
4410         iput(child_inode);
4411         if (child != NULL)
4412                 osd_object_put(env, child);
4413         LASSERT(osd_invariant(obj));
4414         RETURN(rc);
4415 }
4416
4417 /**
4418  *  Initialize osd Iterator for given osd index object.
4419  *
4420  *  \param  dt      osd index object
4421  */
4422
4423 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4424                                      struct dt_object *dt,
4425                                      __u32 unused,
4426                                      struct lustre_capa *capa)
4427 {
4428         struct osd_it_iam      *it;
4429         struct osd_thread_info *oti = osd_oti_get(env);
4430         struct osd_object      *obj = osd_dt_obj(dt);
4431         struct lu_object       *lo  = &dt->do_lu;
4432         struct iam_path_descr  *ipd;
4433         struct iam_container   *bag = &obj->oo_dir->od_container;
4434
4435         LASSERT(lu_object_exists(lo));
4436
4437         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4438                 return ERR_PTR(-EACCES);
4439
4440         it = &oti->oti_it;
4441         ipd = osd_it_ipd_get(env, bag);
4442         if (likely(ipd != NULL)) {
4443                 it->oi_obj = obj;
4444                 it->oi_ipd = ipd;
4445                 lu_object_get(lo);
4446                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4447                 return (struct dt_it *)it;
4448         }
4449         return ERR_PTR(-ENOMEM);
4450 }
4451
4452 /**
4453  * free given Iterator.
4454  */
4455
4456 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4457 {
4458         struct osd_it_iam *it = (struct osd_it_iam *)di;
4459         struct osd_object *obj = it->oi_obj;
4460
4461         iam_it_fini(&it->oi_it);
4462         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4463         lu_object_put(env, &obj->oo_dt.do_lu);
4464 }
4465
4466 /**
4467  *  Move Iterator to record specified by \a key
4468  *
4469  *  \param  di      osd iterator
4470  *  \param  key     key for index
4471  *
4472  *  \retval +ve  di points to record with least key not larger than key
4473  *  \retval  0   di points to exact matched key
4474  *  \retval -ve  failure
4475  */
4476
4477 static int osd_it_iam_get(const struct lu_env *env,
4478                           struct dt_it *di, const struct dt_key *key)
4479 {
4480         struct osd_thread_info  *oti = osd_oti_get(env);
4481         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4482
4483         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4484                 /* swab quota uid/gid */
4485                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4486                 key = (struct dt_key *)&oti->oti_quota_id;
4487         }
4488
4489         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4490 }
4491
4492 /**
4493  *  Release Iterator
4494  *
4495  *  \param  di      osd iterator
4496  */
4497 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4498 {
4499         struct osd_it_iam *it = (struct osd_it_iam *)di;
4500
4501         iam_it_put(&it->oi_it);
4502 }
4503
4504 /**
4505  *  Move iterator by one record
4506  *
4507  *  \param  di      osd iterator
4508  *
4509  *  \retval +1   end of container reached
4510  *  \retval  0   success
4511  *  \retval -ve  failure
4512  */
4513
4514 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4515 {
4516         struct osd_it_iam *it = (struct osd_it_iam *)di;
4517
4518         return iam_it_next(&it->oi_it);
4519 }
4520
4521 /**
4522  * Return pointer to the key under iterator.
4523  */
4524
4525 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4526                                  const struct dt_it *di)
4527 {
4528         struct osd_thread_info *oti = osd_oti_get(env);
4529         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4530         struct osd_object      *obj = it->oi_obj;
4531         struct dt_key          *key;
4532
4533         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4534
4535         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4536                 /* swab quota uid/gid */
4537                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4538                 key = (struct dt_key *)&oti->oti_quota_id;
4539         }
4540
4541         return key;
4542 }
4543
4544 /**
4545  * Return size of key under iterator (in bytes)
4546  */
4547
4548 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4549 {
4550         struct osd_it_iam *it = (struct osd_it_iam *)di;
4551
4552         return iam_it_key_size(&it->oi_it);
4553 }
4554
4555 static inline void
4556 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4557 {
4558         /* check if file type is required */
4559         if (ent->lde_attrs & LUDA_TYPE) {
4560                 struct luda_type *lt;
4561                 int align = sizeof(*lt) - 1;
4562
4563                 len = (len + align) & ~align;
4564                 lt = (struct luda_type *)(ent->lde_name + len);
4565                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4566         }
4567
4568         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4569 }
4570
4571 /**
4572  * build lu direct from backend fs dirent.
4573  */
4574
4575 static inline void
4576 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4577                    char *name, __u16 namelen, __u16 type, __u32 attr)
4578 {
4579         ent->lde_attrs = attr | LUDA_FID;
4580         fid_cpu_to_le(&ent->lde_fid, fid);
4581
4582         ent->lde_hash = cpu_to_le64(offset);
4583         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4584
4585         strncpy(ent->lde_name, name, namelen);
4586         ent->lde_name[namelen] = '\0';
4587         ent->lde_namelen = cpu_to_le16(namelen);
4588
4589         /* append lustre attributes */
4590         osd_it_append_attrs(ent, namelen, type);
4591 }
4592
4593 /**
4594  * Return pointer to the record under iterator.
4595  */
4596 static int osd_it_iam_rec(const struct lu_env *env,
4597                           const struct dt_it *di,
4598                           struct dt_rec *dtrec, __u32 attr)
4599 {
4600         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4601         struct osd_thread_info *info = osd_oti_get(env);
4602         ENTRY;
4603
4604         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4605                 const struct osd_fid_pack *rec;
4606                 struct lu_fid             *fid = &info->oti_fid;
4607                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4608                 char                      *name;
4609                 int                        namelen;
4610                 __u64                      hash;
4611                 int                        rc;
4612
4613                 name = (char *)iam_it_key_get(&it->oi_it);
4614                 if (IS_ERR(name))
4615                         RETURN(PTR_ERR(name));
4616
4617                 namelen = iam_it_key_size(&it->oi_it);
4618
4619                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4620                 if (IS_ERR(rec))
4621                         RETURN(PTR_ERR(rec));
4622
4623                 rc = osd_fid_unpack(fid, rec);
4624                 if (rc)
4625                         RETURN(rc);
4626
4627                 hash = iam_it_store(&it->oi_it);
4628
4629                 /* IAM does not store object type in IAM index (dir) */
4630                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4631                                    0, LUDA_FID);
4632         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4633                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4634                            (struct iam_rec *)dtrec);
4635                 osd_quota_unpack(it->oi_obj, dtrec);
4636         } else {
4637                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4638                            (struct iam_rec *)dtrec);
4639         }
4640
4641         RETURN(0);
4642 }
4643
4644 /**
4645  * Returns cookie for current Iterator position.
4646  */
4647 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4648 {
4649         struct osd_it_iam *it = (struct osd_it_iam *)di;
4650
4651         return iam_it_store(&it->oi_it);
4652 }
4653
4654 /**
4655  * Restore iterator from cookie.
4656  *
4657  * \param  di      osd iterator
4658  * \param  hash    Iterator location cookie
4659  *
4660  * \retval +ve  di points to record with least key not larger than key.
4661  * \retval  0   di points to exact matched key
4662  * \retval -ve  failure
4663  */
4664
4665 static int osd_it_iam_load(const struct lu_env *env,
4666                            const struct dt_it *di, __u64 hash)
4667 {
4668         struct osd_it_iam *it = (struct osd_it_iam *)di;
4669
4670         return iam_it_load(&it->oi_it, hash);
4671 }
4672
4673 static const struct dt_index_operations osd_index_iam_ops = {
4674         .dio_lookup         = osd_index_iam_lookup,
4675         .dio_declare_insert = osd_index_declare_iam_insert,
4676         .dio_insert         = osd_index_iam_insert,
4677         .dio_declare_delete = osd_index_declare_iam_delete,
4678         .dio_delete         = osd_index_iam_delete,
4679         .dio_it     = {
4680                 .init     = osd_it_iam_init,
4681                 .fini     = osd_it_iam_fini,
4682                 .get      = osd_it_iam_get,
4683                 .put      = osd_it_iam_put,
4684                 .next     = osd_it_iam_next,
4685                 .key      = osd_it_iam_key,
4686                 .key_size = osd_it_iam_key_size,
4687                 .rec      = osd_it_iam_rec,
4688                 .store    = osd_it_iam_store,
4689                 .load     = osd_it_iam_load
4690         }
4691 };
4692
4693
4694 /**
4695  * Creates or initializes iterator context.
4696  *
4697  * \retval struct osd_it_ea, iterator structure on success
4698  *
4699  */
4700 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4701                                     struct dt_object *dt,
4702                                     __u32 attr,
4703                                     struct lustre_capa *capa)
4704 {
4705         struct osd_object       *obj  = osd_dt_obj(dt);
4706         struct osd_thread_info  *info = osd_oti_get(env);
4707         struct osd_it_ea        *it   = &info->oti_it_ea;
4708         struct file             *file = &it->oie_file;
4709         struct lu_object        *lo   = &dt->do_lu;
4710         struct dentry           *obj_dentry = &info->oti_it_dentry;
4711         ENTRY;
4712         LASSERT(lu_object_exists(lo));
4713
4714         obj_dentry->d_inode = obj->oo_inode;
4715         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4716         obj_dentry->d_name.hash = 0;
4717
4718         it->oie_rd_dirent       = 0;
4719         it->oie_it_dirent       = 0;
4720         it->oie_dirent          = NULL;
4721         it->oie_buf             = info->oti_it_ea_buf;
4722         it->oie_obj             = obj;
4723
4724         /* Reset the "file" totally to avoid to reuse any old value from
4725          * former readdir handling, the "file->f_pos" should be zero. */
4726         memset(file, 0, sizeof(*file));
4727         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4728         if (attr & LUDA_64BITHASH)
4729                 file->f_mode    = FMODE_64BITHASH;
4730         else
4731                 file->f_mode    = FMODE_32BITHASH;
4732         file->f_dentry          = obj_dentry;
4733         file->f_mapping         = obj->oo_inode->i_mapping;
4734         file->f_op              = obj->oo_inode->i_fop;
4735         set_file_inode(file, obj->oo_inode);
4736
4737         lu_object_get(lo);
4738         RETURN((struct dt_it *) it);
4739 }
4740
4741 /**
4742  * Destroy or finishes iterator context.
4743  *
4744  * \param di iterator structure to be destroyed
4745  */
4746 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4747 {
4748         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4749         struct osd_object    *obj  = it->oie_obj;
4750         struct inode       *inode  = obj->oo_inode;
4751
4752         ENTRY;
4753         it->oie_file.f_op->release(inode, &it->oie_file);
4754         lu_object_put(env, &obj->oo_dt.do_lu);
4755         EXIT;
4756 }
4757
4758 /**
4759  * It position the iterator at given key, so that next lookup continues from
4760  * that key Or it is similar to dio_it->load() but based on a key,
4761  * rather than file position.
4762  *
4763  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4764  * to the beginning.
4765  *
4766  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4767  */
4768 static int osd_it_ea_get(const struct lu_env *env,
4769                          struct dt_it *di, const struct dt_key *key)
4770 {
4771         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4772
4773         ENTRY;
4774         LASSERT(((const char *)key)[0] == '\0');
4775         it->oie_file.f_pos      = 0;
4776         it->oie_rd_dirent       = 0;
4777         it->oie_it_dirent       = 0;
4778         it->oie_dirent          = NULL;
4779
4780         RETURN(+1);
4781 }
4782
4783 /**
4784  * Does nothing
4785  */
4786 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4787 {
4788 }
4789
4790 struct osd_filldir_cbs {
4791 #ifdef HAVE_DIR_CONTEXT
4792         struct dir_context ctx;
4793 #endif
4794         struct osd_it_ea  *it;
4795 };
4796 /**
4797  * It is called internally by ->readdir(). It fills the
4798  * iterator's in-memory data structure with required
4799  * information i.e. name, namelen, rec_size etc.
4800  *
4801  * \param buf in which information to be filled in.
4802  * \param name name of the file in given dir
4803  *
4804  * \retval 0 on success
4805  * \retval 1 on buffer full
4806  */
4807 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4808                                loff_t offset, __u64 ino,
4809                                unsigned d_type)
4810 {
4811         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4812         struct osd_object       *obj  = it->oie_obj;
4813         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4814         struct lu_fid           *fid  = &ent->oied_fid;
4815         struct osd_fid_pack     *rec;
4816         ENTRY;
4817
4818         /* this should never happen */
4819         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4820                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4821                 RETURN(-EIO);
4822         }
4823
4824         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4825             OSD_IT_EA_BUFSIZE)
4826                 RETURN(1);
4827
4828         /* "." is just the object itself. */
4829         if (namelen == 1 && name[0] == '.') {
4830                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4831         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4832                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4833                 if (osd_fid_unpack(fid, rec) != 0)
4834                         fid_zero(fid);
4835         } else {
4836                 fid_zero(fid);
4837         }
4838         d_type &= ~LDISKFS_DIRENT_LUFID;
4839
4840         /* NOT export local root. */
4841         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4842                 ino = obj->oo_inode->i_ino;
4843                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4844         }
4845
4846         ent->oied_ino     = ino;
4847         ent->oied_off     = offset;
4848         ent->oied_namelen = namelen;
4849         ent->oied_type    = d_type;
4850
4851         memcpy(ent->oied_name, name, namelen);
4852
4853         it->oie_rd_dirent++;
4854         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4855         RETURN(0);
4856 }
4857
4858 /**
4859  * Calls ->readdir() to load a directory entry at a time
4860  * and stored it in iterator's in-memory data structure.
4861  *
4862  * \param di iterator's in memory structure
4863  *
4864  * \retval   0 on success
4865  * \retval -ve on error
4866  * \retval +1 reach the end of entry
4867  */
4868 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4869                                const struct dt_it *di)
4870 {
4871         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4872         struct osd_object  *obj   = it->oie_obj;
4873         struct inode       *inode = obj->oo_inode;
4874         struct htree_lock  *hlock = NULL;
4875         struct file        *filp  = &it->oie_file;
4876         int                 rc = 0;
4877         struct osd_filldir_cbs buf = {
4878 #ifdef HAVE_DIR_CONTEXT
4879                 .ctx.actor = osd_ldiskfs_filldir,
4880 #endif
4881                 .it = it
4882         };
4883
4884         ENTRY;
4885         it->oie_dirent = it->oie_buf;
4886         it->oie_rd_dirent = 0;
4887
4888         if (obj->oo_hl_head != NULL) {
4889                 hlock = osd_oti_get(env)->oti_hlock;
4890                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4891                                    inode, LDISKFS_HLOCK_READDIR);
4892         } else {
4893                 down_read(&obj->oo_ext_idx_sem);
4894         }
4895
4896 #ifdef HAVE_DIR_CONTEXT
4897         buf.ctx.pos = filp->f_pos;
4898         rc = inode->i_fop->iterate(filp, &buf.ctx);
4899         filp->f_pos = buf.ctx.pos;
4900 #else
4901         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4902 #endif
4903
4904         if (hlock != NULL)
4905                 ldiskfs_htree_unlock(hlock);
4906         else
4907                 up_read(&obj->oo_ext_idx_sem);
4908
4909         if (it->oie_rd_dirent == 0) {
4910                 /*If it does not get any dirent, it means it has been reached
4911                  *to the end of the dir */
4912                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4913                 if (rc == 0)
4914                         rc = 1;
4915         } else {
4916                 it->oie_dirent = it->oie_buf;
4917                 it->oie_it_dirent = 1;
4918         }
4919
4920         RETURN(rc);
4921 }
4922
4923 /**
4924  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4925  * to load a directory entry at a time and stored it in
4926  * iterator's in-memory data structure.
4927  *
4928  * \param di iterator's in memory structure
4929  *
4930  * \retval +ve iterator reached to end
4931  * \retval   0 iterator not reached to end
4932  * \retval -ve on error
4933  */
4934 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4935 {
4936         struct osd_it_ea *it = (struct osd_it_ea *)di;
4937         int rc;
4938
4939         ENTRY;
4940
4941         if (it->oie_it_dirent < it->oie_rd_dirent) {
4942                 it->oie_dirent =
4943                         (void *) it->oie_dirent +
4944                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4945                                        it->oie_dirent->oied_namelen);
4946                 it->oie_it_dirent++;
4947                 RETURN(0);
4948         } else {
4949                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4950                         rc = +1;
4951                 else
4952                         rc = osd_ldiskfs_it_fill(env, di);
4953         }
4954
4955         RETURN(rc);
4956 }
4957
4958 /**
4959  * Returns the key at current position from iterator's in memory structure.
4960  *
4961  * \param di iterator's in memory structure
4962  *
4963  * \retval key i.e. struct dt_key on success
4964  */
4965 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4966                                     const struct dt_it *di)
4967 {
4968         struct osd_it_ea *it = (struct osd_it_ea *)di;
4969
4970         return (struct dt_key *)it->oie_dirent->oied_name;
4971 }
4972
4973 /**
4974  * Returns the key's size at current position from iterator's in memory structure.
4975  *
4976  * \param di iterator's in memory structure
4977  *
4978  * \retval key_size i.e. struct dt_key on success
4979  */
4980 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4981 {
4982         struct osd_it_ea *it = (struct osd_it_ea *)di;
4983
4984         return it->oie_dirent->oied_namelen;
4985 }
4986
4987 static int
4988 osd_dirent_update(handle_t *jh, struct super_block *sb,
4989                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4990                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4991 {
4992         struct osd_fid_pack *rec;
4993         int                  rc;
4994         ENTRY;
4995
4996         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4997         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4998
4999         rc = ldiskfs_journal_get_write_access(jh, bh);
5000         if (rc != 0) {
5001                 CERROR("%.16s: fail to write access for update dirent: "
5002                        "name = %.*s, rc = %d\n",
5003                        LDISKFS_SB(sb)->s_es->s_volume_name,
5004                        ent->oied_namelen, ent->oied_name, rc);
5005                 RETURN(rc);
5006         }
5007
5008         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5009         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5010         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5011         if (rc != 0)
5012                 CERROR("%.16s: fail to dirty metadata for update dirent: "
5013                        "name = %.*s, rc = %d\n",
5014                        LDISKFS_SB(sb)->s_es->s_volume_name,
5015                        ent->oied_namelen, ent->oied_name, rc);
5016
5017         RETURN(rc);
5018 }
5019
5020 static inline int
5021 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5022 {
5023         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5024             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5025                 return 1;
5026         else
5027                 return 0;
5028 }
5029
5030 static inline int
5031 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5032 {
5033         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5034                  "dot_dotdot = %d\n", dot_dotdot);
5035
5036         if (LDISKFS_DIR_REC_LEN(de) >=
5037             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5038                 return 1;
5039         else
5040                 return 0;
5041 }
5042
5043 static int
5044 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5045                     struct inode *dir, struct inode *inode,
5046                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5047                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5048                     struct htree_lock *hlock)
5049 {
5050         struct dentry               *dentry;
5051         struct osd_fid_pack         *rec;
5052         struct ldiskfs_dentry_param *ldp;
5053         int                          rc;
5054         ENTRY;
5055
5056         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5057                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5058                 RETURN(0);
5059
5060         /* There is enough space to hold the FID-in-dirent. */
5061         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5062                                  dir->i_sb->s_blocksize)) {
5063                 rc = ldiskfs_journal_get_write_access(jh, bh);
5064                 if (rc != 0) {
5065                         CERROR("%.16s: fail to write access for reinsert "
5066                                "dirent: name = %.*s, rc = %d\n",
5067                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5068                                ent->oied_namelen, ent->oied_name, rc);
5069                         RETURN(rc);
5070                 }
5071
5072                 de->name[de->name_len] = 0;
5073                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5074                 rec->fp_len = sizeof(struct lu_fid) + 1;
5075                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5076                 de->file_type |= LDISKFS_DIRENT_LUFID;
5077
5078                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5079                 if (rc != 0)
5080                         CERROR("%.16s: fail to dirty metadata for reinsert "
5081                                "dirent: name = %.*s, rc = %d\n",
5082                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5083                                ent->oied_namelen, ent->oied_name, rc);
5084
5085                 RETURN(rc);
5086         }
5087
5088         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5089         if (rc != 0) {
5090                 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5091                        "name = %.*s, rc = %d\n",
5092                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5093                        ent->oied_namelen, ent->oied_name, rc);
5094                 RETURN(rc);
5095         }
5096
5097         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5098                                            ent->oied_namelen);
5099         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5100         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5101         dentry->d_fsdata = (void *)ldp;
5102         ll_vfs_dq_init(dir);
5103         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5104         /* It is too bad, we cannot reinsert the name entry back.
5105          * That means we lose it! */
5106         if (rc != 0)
5107                 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5108                        "name = %.*s, rc = %d\n",
5109                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5110                        ent->oied_namelen, ent->oied_name, rc);
5111
5112         RETURN(rc);
5113 }
5114
5115 static int
5116 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5117                         struct osd_it_ea *it, struct lu_fid *fid,
5118                         struct osd_inode_id *id, __u32 *attr)
5119 {
5120         struct osd_thread_info     *info        = osd_oti_get(env);
5121         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5122         struct osd_device          *dev         = osd_obj2dev(obj);
5123         struct super_block         *sb          = osd_sb(dev);
5124         const char                 *devname     =
5125                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5126         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5127         struct inode               *dir         = obj->oo_inode;
5128         struct htree_lock          *hlock       = NULL;
5129         struct buffer_head         *bh          = NULL;
5130         handle_t                   *jh          = NULL;
5131         struct ldiskfs_dir_entry_2 *de;
5132         struct dentry              *dentry;
5133         struct inode               *inode;
5134         int                         credits;
5135         int                         rc;
5136         int                         dot_dotdot  = 0;
5137         bool                        dirty       = false;
5138         ENTRY;
5139
5140         if (ent->oied_name[0] == '.') {
5141                 if (ent->oied_namelen == 1)
5142                         dot_dotdot = 1;
5143                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5144                         dot_dotdot = 2;
5145         }
5146
5147         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5148                                       ent->oied_namelen);
5149
5150         /* We need to ensure that the name entry is still valid.
5151          * Because it may be removed or renamed by other already.
5152          *
5153          * The unlink or rename operation will start journal before PDO lock,
5154          * so to avoid deadlock, here we need to start journal handle before
5155          * related PDO lock also. But because we do not know whether there
5156          * will be something to be repaired before PDO lock, we just start
5157          * journal without conditions.
5158          *
5159          * We may need to remove the name entry firstly, then insert back.
5160          * One credit is for user quota file update.
5161          * One credit is for group quota file update.
5162          * Two credits are for dirty inode. */
5163         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5164                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5165
5166 again:
5167         if (dev->od_dirent_journal) {
5168                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5169                 if (IS_ERR(jh)) {
5170                         rc = PTR_ERR(jh);
5171                         CERROR("%.16s: fail to start trans for dirent "
5172                                "check_repair: credits %d, name %.*s, rc %d\n",
5173                                devname, credits, ent->oied_namelen,
5174                                ent->oied_name, rc);
5175                         RETURN(rc);
5176                 }
5177
5178                 if (obj->oo_hl_head != NULL) {
5179                         hlock = osd_oti_get(env)->oti_hlock;
5180                         /* "0" means exclusive lock for the whole directory.
5181                          * We need to prevent others access such name entry
5182                          * during the delete + insert. Neither HLOCK_ADD nor
5183                          * HLOCK_DEL cannot guarantee the atomicity. */
5184                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5185                 } else {
5186                         down_write(&obj->oo_ext_idx_sem);
5187                 }
5188         } else {
5189                 if (obj->oo_hl_head != NULL) {
5190                         hlock = osd_oti_get(env)->oti_hlock;
5191                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5192                                            LDISKFS_HLOCK_LOOKUP);
5193                 } else {
5194                         down_read(&obj->oo_ext_idx_sem);
5195                 }
5196         }
5197
5198         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5199         /* For dot/dotdot entry, if there is not enough space to hold the
5200          * FID-in-dirent, just keep them there. It only happens when the
5201          * device upgraded from 1.8 or restored from MDT file-level backup.
5202          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5203          * and needs to get FID from LMA when readdir, it will not affect the
5204          * performance much. */
5205         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5206             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5207                 *attr |= LUDA_IGNORE;
5208                 GOTO(out_journal, rc = 0);
5209         }
5210
5211         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5212         inode = osd_iget(info, dev, id);
5213         if (IS_ERR(inode)) {
5214                 rc = PTR_ERR(inode);
5215                 if (rc == -ENOENT || rc == -ESTALE) {
5216                         *attr |= LUDA_IGNORE;
5217                         rc = 0;
5218                 }
5219
5220                 GOTO(out_journal, rc);
5221         }
5222
5223         /* skip the REMOTE_PARENT_DIR. */
5224         if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5225                 GOTO(out_inode, rc = 0);
5226
5227         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5228         if (rc == 0) {
5229                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5230
5231                 if (fid_is_sane(fid)) {
5232                         /* FID-in-dirent is valid. */
5233                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5234                                 GOTO(out_inode, rc = 0);
5235
5236                         /* Do not repair under dryrun mode. */
5237                         if (*attr & LUDA_VERIFY_DRYRUN) {
5238                                 *attr |= LUDA_REPAIR;
5239                                 GOTO(out_inode, rc = 0);
5240                         }
5241
5242                         if (!dev->od_dirent_journal) {
5243                                 iput(inode);
5244                                 brelse(bh);
5245                                 if (hlock != NULL)
5246                                         ldiskfs_htree_unlock(hlock);
5247                                 else
5248                                         up_read(&obj->oo_ext_idx_sem);
5249                                 dev->od_dirent_journal = 1;
5250                                 goto again;
5251                         }
5252
5253                         *fid = lma->lma_self_fid;
5254                         dirty = true;
5255                         /* Update the FID-in-dirent. */
5256                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5257                         if (rc == 0)
5258                                 *attr |= LUDA_REPAIR;
5259                 } else {
5260                         /* Do not repair under dryrun mode. */
5261                         if (*attr & LUDA_VERIFY_DRYRUN) {
5262                                 *fid = lma->lma_self_fid;
5263                                 *attr |= LUDA_REPAIR;
5264                                 GOTO(out_inode, rc = 0);
5265                         }
5266
5267                         if (!dev->od_dirent_journal) {
5268                                 iput(inode);
5269                                 brelse(bh);
5270                                 if (hlock != NULL)
5271                                         ldiskfs_htree_unlock(hlock);
5272                                 else
5273                                         up_read(&obj->oo_ext_idx_sem);
5274                                 dev->od_dirent_journal = 1;
5275                                 goto again;
5276                         }
5277
5278                         *fid = lma->lma_self_fid;
5279                         dirty = true;
5280                         /* Append the FID-in-dirent. */
5281                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5282                                                  fid, bh, de, hlock);
5283                         if (rc == 0)
5284                                 *attr |= LUDA_REPAIR;
5285                 }
5286         } else if (rc == -ENODATA) {
5287                 /* Do not repair under dryrun mode. */
5288                 if (*attr & LUDA_VERIFY_DRYRUN) {
5289                         if (fid_is_sane(fid)) {
5290                                 *attr |= LUDA_REPAIR;
5291                         } else {
5292                                 lu_igif_build(fid, inode->i_ino,
5293                                               inode->i_generation);
5294                                 *attr |= LUDA_UPGRADE;
5295                         }
5296                         GOTO(out_inode, rc = 0);
5297                 }
5298
5299                 if (!dev->od_dirent_journal) {
5300                         iput(inode);
5301                         brelse(bh);
5302                         if (hlock != NULL)
5303                                 ldiskfs_htree_unlock(hlock);
5304                         else
5305                                 up_read(&obj->oo_ext_idx_sem);
5306                         dev->od_dirent_journal = 1;
5307                         goto again;
5308                 }
5309
5310                 dirty = true;
5311                 if (unlikely(fid_is_sane(fid))) {
5312                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5313                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5314                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5315                         if (rc == 0)
5316                                 *attr |= LUDA_REPAIR;
5317                 } else {
5318                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5319                         /* It is probably IGIF object. Only aappend the
5320                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5321                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5322                                                  fid, bh, de, hlock);
5323                         if (rc == 0)
5324                                 *attr |= LUDA_UPGRADE;
5325                 }
5326         }
5327
5328         GOTO(out_inode, rc);
5329
5330 out_inode:
5331         iput(inode);
5332
5333 out_journal:
5334         brelse(bh);
5335         if (hlock != NULL) {
5336                 ldiskfs_htree_unlock(hlock);
5337         } else {
5338                 if (dev->od_dirent_journal)
5339                         up_write(&obj->oo_ext_idx_sem);
5340                 else
5341                         up_read(&obj->oo_ext_idx_sem);
5342         }
5343         if (jh != NULL)
5344                 ldiskfs_journal_stop(jh);
5345         if (rc >= 0 && !dirty)
5346                 dev->od_dirent_journal = 0;
5347         return rc;
5348 }
5349
5350 /**
5351  * Returns the value at current position from iterator's in memory structure.
5352  *
5353  * \param di struct osd_it_ea, iterator's in memory structure
5354  * \param attr attr requested for dirent.
5355  * \param lde lustre dirent
5356  *
5357  * \retval   0 no error and \param lde has correct lustre dirent.
5358  * \retval -ve on error
5359  */
5360 static inline int osd_it_ea_rec(const struct lu_env *env,
5361                                 const struct dt_it *di,
5362                                 struct dt_rec *dtrec, __u32 attr)
5363 {
5364         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5365         struct osd_object      *obj   = it->oie_obj;
5366         struct osd_device      *dev   = osd_obj2dev(obj);
5367         struct osd_scrub       *scrub = &dev->od_scrub;
5368         struct scrub_file      *sf    = &scrub->os_file;
5369         struct osd_thread_info *oti   = osd_oti_get(env);
5370         struct osd_inode_id    *id    = &oti->oti_id;
5371         struct osd_idmap_cache *oic   = &oti->oti_cache;
5372         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5373         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5374         __u32                   ino   = it->oie_dirent->oied_ino;
5375         int                     rc    = 0;
5376         ENTRY;
5377
5378         if (attr & LUDA_VERIFY) {
5379                 attr |= LUDA_TYPE;
5380                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5381                         attr |= LUDA_IGNORE;
5382                         rc = 0;
5383                 } else {
5384                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5385                                                      &attr);
5386                 }
5387         } else {
5388                 attr &= ~LU_DIRENT_ATTRS_MASK;
5389                 if (!fid_is_sane(fid)) {
5390                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5391                             likely(it->oie_dirent->oied_namelen != 2 ||
5392                                    it->oie_dirent->oied_name[0] != '.' ||
5393                                    it->oie_dirent->oied_name[1] != '.'))
5394                                 RETURN(-ENOENT);
5395
5396                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
5397                 } else {
5398                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5399                 }
5400         }
5401
5402         /* Pack the entry anyway, at least the offset is right. */
5403         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5404                            it->oie_dirent->oied_name,
5405                            it->oie_dirent->oied_namelen,
5406                            it->oie_dirent->oied_type, attr);
5407
5408         if (rc < 0)
5409                 RETURN(rc);
5410
5411         if (osd_remote_fid(env, dev, fid))
5412                 RETURN(0);
5413
5414         if (likely(!(attr & LUDA_IGNORE)))
5415                 rc = osd_add_oi_cache(oti, dev, id, fid);
5416
5417         if (!(attr & LUDA_VERIFY) &&
5418             (scrub->os_pos_current <= ino) &&
5419             ((sf->sf_flags & SF_INCONSISTENT) ||
5420              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5421              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5422                 osd_consistency_check(oti, dev, oic);
5423
5424         RETURN(rc);
5425 }
5426
5427 /**
5428  * Returns the record size size at current position.
5429  *
5430  * This function will return record(lu_dirent) size in bytes.
5431  *
5432  * \param[in] env       execution environment
5433  * \param[in] di        iterator's in memory structure
5434  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5435  *                      calculate the lu_dirent size.
5436  *
5437  * \retval      record size(in bytes & in memory) of the current lu_dirent
5438  *              entry.
5439  */
5440 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5441                               __u32 attr)
5442 {
5443         struct osd_it_ea *it = (struct osd_it_ea *)di;
5444
5445         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5446 }
5447
5448 /**
5449  * Returns a cookie for current position of the iterator head, so that
5450  * user can use this cookie to load/start the iterator next time.
5451  *
5452  * \param di iterator's in memory structure
5453  *
5454  * \retval cookie for current position, on success
5455  */
5456 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5457 {
5458         struct osd_it_ea *it = (struct osd_it_ea *)di;
5459
5460         return it->oie_dirent->oied_off;
5461 }
5462
5463 /**
5464  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5465  * to load a directory entry at a time and stored it i inn,
5466  * in iterator's in-memory data structure.
5467  *
5468  * \param di struct osd_it_ea, iterator's in memory structure
5469  *
5470  * \retval +ve on success
5471  * \retval -ve on error
5472  */
5473 static int osd_it_ea_load(const struct lu_env *env,
5474                           const struct dt_it *di, __u64 hash)
5475 {
5476         struct osd_it_ea *it = (struct osd_it_ea *)di;
5477         int rc;
5478
5479         ENTRY;
5480         it->oie_file.f_pos = hash;
5481
5482         rc =  osd_ldiskfs_it_fill(env, di);
5483         if (rc > 0)
5484                 rc = -ENODATA;
5485
5486         if (rc == 0)
5487                 rc = +1;
5488
5489         RETURN(rc);
5490 }
5491
5492 /**
5493  * Index lookup function for interoperability mode (b11826).
5494  *
5495  * \param key,  key i.e. file name to be searched
5496  *
5497  * \retval +ve, on success
5498  * \retval -ve, on error
5499  */
5500 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5501                                struct dt_rec *rec, const struct dt_key *key,
5502                                struct lustre_capa *capa)
5503 {
5504         struct osd_object *obj = osd_dt_obj(dt);
5505         int rc = 0;
5506
5507         ENTRY;
5508
5509         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5510         LINVRNT(osd_invariant(obj));
5511
5512         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5513                 return -EACCES;
5514
5515         rc = osd_ea_lookup_rec(env, obj, rec, key);
5516         if (rc == 0)
5517                 rc = +1;
5518         RETURN(rc);
5519 }
5520
5521 /**
5522  * Index and Iterator operations for interoperability
5523  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5524  */
5525 static const struct dt_index_operations osd_index_ea_ops = {
5526         .dio_lookup         = osd_index_ea_lookup,
5527         .dio_declare_insert = osd_index_declare_ea_insert,
5528         .dio_insert         = osd_index_ea_insert,
5529         .dio_declare_delete = osd_index_declare_ea_delete,
5530         .dio_delete         = osd_index_ea_delete,
5531         .dio_it     = {
5532                 .init     = osd_it_ea_init,
5533                 .fini     = osd_it_ea_fini,
5534                 .get      = osd_it_ea_get,
5535                 .put      = osd_it_ea_put,
5536                 .next     = osd_it_ea_next,
5537                 .key      = osd_it_ea_key,
5538                 .key_size = osd_it_ea_key_size,
5539                 .rec      = osd_it_ea_rec,
5540                 .rec_size = osd_it_ea_rec_size,
5541                 .store    = osd_it_ea_store,
5542                 .load     = osd_it_ea_load
5543         }
5544 };
5545
5546 static void *osd_key_init(const struct lu_context *ctx,
5547                           struct lu_context_key *key)
5548 {
5549         struct osd_thread_info *info;
5550
5551         OBD_ALLOC_PTR(info);
5552         if (info == NULL)
5553                 return ERR_PTR(-ENOMEM);
5554
5555         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5556         if (info->oti_it_ea_buf == NULL)
5557                 goto out_free_info;
5558
5559         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5560
5561         info->oti_hlock = ldiskfs_htree_lock_alloc();
5562         if (info->oti_hlock == NULL)
5563                 goto out_free_ea;
5564
5565         return info;
5566
5567  out_free_ea:
5568         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5569  out_free_info:
5570         OBD_FREE_PTR(info);
5571         return ERR_PTR(-ENOMEM);
5572 }
5573
5574 static void osd_key_fini(const struct lu_context *ctx,
5575                          struct lu_context_key *key, void* data)
5576 {
5577         struct osd_thread_info *info = data;
5578
5579         if (info->oti_hlock != NULL)
5580                 ldiskfs_htree_lock_free(info->oti_hlock);
5581         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5582         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5583         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5584         OBD_FREE_PTR(info);
5585 }
5586
5587 static void osd_key_exit(const struct lu_context *ctx,
5588                          struct lu_context_key *key, void *data)
5589 {
5590         struct osd_thread_info *info = data;
5591
5592         LASSERT(info->oti_r_locks == 0);
5593         LASSERT(info->oti_w_locks == 0);
5594         LASSERT(info->oti_txns    == 0);
5595 }
5596
5597 /* type constructor/destructor: osd_type_init, osd_type_fini */
5598 LU_TYPE_INIT_FINI(osd, &osd_key);
5599
5600 struct lu_context_key osd_key = {
5601         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5602         .lct_init = osd_key_init,
5603         .lct_fini = osd_key_fini,
5604         .lct_exit = osd_key_exit
5605 };
5606
5607
5608 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5609                            const char *name, struct lu_device *next)
5610 {
5611         struct osd_device *osd = osd_dev(d);
5612
5613         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5614             >= sizeof(osd->od_svname))
5615                 return -E2BIG;
5616         return osd_procfs_init(osd, name);
5617 }
5618
5619 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5620 {
5621         struct seq_server_site  *ss = osd_seq_site(osd);
5622         int                     rc;
5623         ENTRY;
5624
5625         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5626                 RETURN(0);
5627
5628         if (unlikely(ss == NULL))
5629                 RETURN(-ENODEV);
5630
5631         OBD_ALLOC_PTR(osd->od_cl_seq);
5632         if (osd->od_cl_seq == NULL)
5633                 RETURN(-ENOMEM);
5634
5635         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5636                              osd->od_svname, ss->ss_server_seq);
5637
5638         if (rc != 0) {
5639                 OBD_FREE_PTR(osd->od_cl_seq);
5640                 osd->od_cl_seq = NULL;
5641         }
5642
5643         RETURN(rc);
5644 }
5645
5646 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5647 {
5648         if (osd->od_cl_seq == NULL)
5649                 return;
5650
5651         seq_client_fini(osd->od_cl_seq);
5652         OBD_FREE_PTR(osd->od_cl_seq);
5653         osd->od_cl_seq = NULL;
5654 }
5655
5656 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5657 {
5658         ENTRY;
5659
5660         /* shutdown quota slave instance associated with the device */
5661         if (o->od_quota_slave != NULL) {
5662                 qsd_fini(env, o->od_quota_slave);
5663                 o->od_quota_slave = NULL;
5664         }
5665
5666         osd_fid_fini(env, o);
5667
5668         RETURN(0);
5669 }
5670
5671 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5672 {
5673         ENTRY;
5674
5675         if (o->od_mnt != NULL) {
5676                 shrink_dcache_sb(osd_sb(o));
5677                 osd_sync(env, &o->od_dt_dev);
5678
5679                 mntput(o->od_mnt);
5680                 o->od_mnt = NULL;
5681         }
5682
5683         EXIT;
5684 }
5685
5686 static int osd_mount(const struct lu_env *env,
5687                      struct osd_device *o, struct lustre_cfg *cfg)
5688 {
5689         const char              *name  = lustre_cfg_string(cfg, 0);
5690         const char              *dev  = lustre_cfg_string(cfg, 1);
5691         const char              *opts;
5692         unsigned long            page, s_flags, lmd_flags = 0;
5693         struct page             *__page;
5694         struct file_system_type *type;
5695         char                    *options = NULL;
5696         char                    *str;
5697         struct osd_thread_info  *info = osd_oti_get(env);
5698         struct lu_fid           *fid = &info->oti_fid;
5699         struct inode            *inode;
5700         int                      rc = 0;
5701         ENTRY;
5702
5703         if (o->od_mnt != NULL)
5704                 RETURN(0);
5705
5706         if (strlen(dev) >= sizeof(o->od_mntdev))
5707                 RETURN(-E2BIG);
5708         strcpy(o->od_mntdev, dev);
5709
5710         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5711         if (__page == NULL)
5712                 GOTO(out, rc = -ENOMEM);
5713
5714         str = lustre_cfg_string(cfg, 2);
5715         s_flags = simple_strtoul(str, NULL, 0);
5716         str = strstr(str, ":");
5717         if (str)
5718                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5719         opts = lustre_cfg_string(cfg, 3);
5720         page = (unsigned long)page_address(__page);
5721         options = (char *)page;
5722         *options = '\0';
5723         if (opts == NULL)
5724                 strcat(options, "user_xattr,acl");
5725         else
5726                 strcat(options, opts);
5727
5728         /* Glom up mount options */
5729         if (*options != '\0')
5730                 strcat(options, ",");
5731         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5732
5733         type = get_fs_type("ldiskfs");
5734         if (!type) {
5735                 CERROR("%s: cannot find ldiskfs module\n", name);
5736                 GOTO(out, rc = -ENODEV);
5737         }
5738
5739         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5740         module_put(type->owner);
5741
5742         if (IS_ERR(o->od_mnt)) {
5743                 rc = PTR_ERR(o->od_mnt);
5744                 o->od_mnt = NULL;
5745                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5746                 GOTO(out, rc);
5747         }
5748
5749 #ifdef HAVE_DEV_SET_RDONLY
5750         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5751                 CERROR("%s: underlying device %s is marked as read-only. "
5752                        "Setup failed\n", name, dev);
5753                 GOTO(out_mnt, rc = -EROFS);
5754         }
5755 #endif
5756
5757         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5758                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5759                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5760                 GOTO(out_mnt, rc = -EINVAL);
5761         }
5762
5763 #ifdef LDISKFS_MOUNT_DIRDATA
5764         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5765                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5766                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5767         else if (!o->od_is_ost)
5768                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5769                       "enabling the dirdata feature. If you do not want to "
5770                       "downgrade to Lustre-1.x again, you can enable it via "
5771                       "'tune2fs -O dirdata device'\n", name, dev);
5772 #endif
5773         inode = osd_sb(o)->s_root->d_inode;
5774         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5775         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5776         if (rc != 0) {
5777                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5778                 GOTO(out_mnt, rc);
5779         }
5780
5781         if (lmd_flags & LMD_FLG_NOSCRUB)
5782                 o->od_noscrub = 1;
5783
5784         GOTO(out, rc = 0);
5785
5786 out_mnt:
5787         mntput(o->od_mnt);
5788         o->od_mnt = NULL;
5789
5790 out:
5791         if (__page)
5792                 OBD_PAGE_FREE(__page);
5793
5794         return rc;
5795 }
5796
5797 static struct lu_device *osd_device_fini(const struct lu_env *env,
5798                                          struct lu_device *d)
5799 {
5800         struct osd_device *o = osd_dev(d);
5801         ENTRY;
5802
5803         osd_shutdown(env, o);
5804         osd_procfs_fini(o);
5805         osd_scrub_cleanup(env, o);
5806         osd_obj_map_fini(o);
5807         osd_umount(env, o);
5808
5809         RETURN(NULL);
5810 }
5811
5812 static int osd_device_init0(const struct lu_env *env,
5813                             struct osd_device *o,
5814                             struct lustre_cfg *cfg)
5815 {
5816         struct lu_device        *l = osd2lu_dev(o);
5817         struct osd_thread_info *info;
5818         int                     rc;
5819         int                     cplen = 0;
5820
5821         /* if the module was re-loaded, env can loose its keys */
5822         rc = lu_env_refill((struct lu_env *) env);
5823         if (rc)
5824                 GOTO(out, rc);
5825         info = osd_oti_get(env);
5826         LASSERT(info);
5827
5828         l->ld_ops = &osd_lu_ops;
5829         o->od_dt_dev.dd_ops = &osd_dt_ops;
5830
5831         spin_lock_init(&o->od_osfs_lock);
5832         mutex_init(&o->od_otable_mutex);
5833
5834         o->od_capa_hash = init_capa_hash();
5835         if (o->od_capa_hash == NULL)
5836                 GOTO(out, rc = -ENOMEM);
5837
5838         o->od_read_cache = 1;
5839         o->od_writethrough_cache = 1;
5840         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5841
5842         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5843                         sizeof(o->od_svname));
5844         if (cplen >= sizeof(o->od_svname)) {
5845                 rc = -E2BIG;
5846                 GOTO(out_capa, rc);
5847         }
5848
5849         if (server_name_is_ost(o->od_svname))
5850                 o->od_is_ost = 1;
5851
5852         rc = osd_mount(env, o, cfg);
5853         if (rc != 0)
5854                 GOTO(out_capa, rc);
5855
5856         rc = osd_obj_map_init(env, o);
5857         if (rc != 0)
5858                 GOTO(out_mnt, rc);
5859
5860         rc = lu_site_init(&o->od_site, l);
5861         if (rc != 0)
5862                 GOTO(out_compat, rc);
5863         o->od_site.ls_bottom_dev = l;
5864
5865         rc = lu_site_init_finish(&o->od_site);
5866         if (rc != 0)
5867                 GOTO(out_site, rc);
5868
5869         /* self-repair LMA by default */
5870         o->od_lma_self_repair = 1;
5871
5872         INIT_LIST_HEAD(&o->od_ios_list);
5873         /* setup scrub, including OI files initialization */
5874         rc = osd_scrub_setup(env, o);
5875         if (rc < 0)
5876                 GOTO(out_site, rc);
5877
5878         rc = osd_procfs_init(o, o->od_svname);
5879         if (rc != 0) {
5880                 CERROR("%s: can't initialize procfs: rc = %d\n",
5881                        o->od_svname, rc);
5882                 GOTO(out_scrub, rc);
5883         }
5884
5885         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5886
5887         /* initialize quota slave instance */
5888         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5889                                      o->od_proc_entry);
5890         if (IS_ERR(o->od_quota_slave)) {
5891                 rc = PTR_ERR(o->od_quota_slave);
5892                 o->od_quota_slave = NULL;
5893                 GOTO(out_procfs, rc);
5894         }
5895
5896         RETURN(0);
5897
5898 out_procfs:
5899         osd_procfs_fini(o);
5900 out_scrub:
5901         osd_scrub_cleanup(env, o);
5902 out_site:
5903         lu_site_fini(&o->od_site);
5904 out_compat:
5905         osd_obj_map_fini(o);
5906 out_mnt:
5907         osd_umount(env, o);
5908 out_capa:
5909         cleanup_capa_hash(o->od_capa_hash);
5910 out:
5911         return rc;
5912 }
5913
5914 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5915                                           struct lu_device_type *t,
5916                                           struct lustre_cfg *cfg)
5917 {
5918         struct osd_device *o;
5919         int                rc;
5920
5921         OBD_ALLOC_PTR(o);
5922         if (o == NULL)
5923                 return ERR_PTR(-ENOMEM);
5924
5925         rc = dt_device_init(&o->od_dt_dev, t);
5926         if (rc == 0) {
5927                 /* Because the ctx might be revived in dt_device_init,
5928                  * refill the env here */
5929                 lu_env_refill((struct lu_env *)env);
5930                 rc = osd_device_init0(env, o, cfg);
5931                 if (rc)
5932                         dt_device_fini(&o->od_dt_dev);
5933         }
5934
5935         if (unlikely(rc != 0))
5936                 OBD_FREE_PTR(o);
5937
5938         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5939 }
5940
5941 static struct lu_device *osd_device_free(const struct lu_env *env,
5942                                          struct lu_device *d)
5943 {
5944         struct osd_device *o = osd_dev(d);
5945         ENTRY;
5946
5947         cleanup_capa_hash(o->od_capa_hash);
5948         /* XXX: make osd top device in order to release reference */
5949         d->ld_site->ls_top_dev = d;
5950         lu_site_purge(env, d->ld_site, -1);
5951         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5952                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5953                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5954         }
5955         lu_site_fini(&o->od_site);
5956         dt_device_fini(&o->od_dt_dev);
5957         OBD_FREE_PTR(o);
5958         RETURN(NULL);
5959 }
5960
5961 static int osd_process_config(const struct lu_env *env,
5962                               struct lu_device *d, struct lustre_cfg *cfg)
5963 {
5964         struct osd_device               *o = osd_dev(d);
5965         int                             rc;
5966         ENTRY;
5967
5968         switch (cfg->lcfg_command) {
5969         case LCFG_SETUP:
5970                 rc = osd_mount(env, o, cfg);
5971                 break;
5972         case LCFG_CLEANUP:
5973                 lu_dev_del_linkage(d->ld_site, d);
5974                 rc = osd_shutdown(env, o);
5975                 break;
5976         case LCFG_PARAM:
5977                 LASSERT(&o->od_dt_dev);
5978                 rc = class_process_proc_seq_param(PARAM_OSD,
5979                                                   lprocfs_osd_obd_vars,
5980                                                   cfg, &o->od_dt_dev);
5981                 if (rc > 0 || rc == -ENOSYS)
5982                         rc = class_process_proc_seq_param(PARAM_OST,
5983                                                           lprocfs_osd_obd_vars,
5984                                                           cfg, &o->od_dt_dev);
5985                 break;
5986         default:
5987                 rc = -ENOSYS;
5988         }
5989
5990         RETURN(rc);
5991 }
5992
5993 static int osd_recovery_complete(const struct lu_env *env,
5994                                  struct lu_device *d)
5995 {
5996         struct osd_device       *osd = osd_dev(d);
5997         int                      rc = 0;
5998         ENTRY;
5999
6000         if (osd->od_quota_slave == NULL)
6001                 RETURN(0);
6002
6003         /* start qsd instance on recovery completion, this notifies the quota
6004          * slave code that we are about to process new requests now */
6005         rc = qsd_start(env, osd->od_quota_slave);
6006         RETURN(rc);
6007 }
6008
6009 /*
6010  * we use exports to track all osd users
6011  */
6012 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6013                            struct obd_device *obd, struct obd_uuid *cluuid,
6014                            struct obd_connect_data *data, void *localdata)
6015 {
6016         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6017         struct lustre_handle  conn;
6018         int                   rc;
6019         ENTRY;
6020
6021         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6022
6023         rc = class_connect(&conn, obd, cluuid);
6024         if (rc)
6025                 RETURN(rc);
6026
6027         *exp = class_conn2export(&conn);
6028
6029         spin_lock(&osd->od_osfs_lock);
6030         osd->od_connects++;
6031         spin_unlock(&osd->od_osfs_lock);
6032
6033         RETURN(0);
6034 }
6035
6036 /*
6037  * once last export (we don't count self-export) disappeared
6038  * osd can be released
6039  */
6040 static int osd_obd_disconnect(struct obd_export *exp)
6041 {
6042         struct obd_device *obd = exp->exp_obd;
6043         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6044         int                rc, release = 0;
6045         ENTRY;
6046
6047         /* Only disconnect the underlying layers on the final disconnect. */
6048         spin_lock(&osd->od_osfs_lock);
6049         osd->od_connects--;
6050         if (osd->od_connects == 0)
6051                 release = 1;
6052         spin_unlock(&osd->od_osfs_lock);
6053
6054         rc = class_disconnect(exp); /* bz 9811 */
6055
6056         if (rc == 0 && release)
6057                 class_manual_cleanup(obd);
6058         RETURN(rc);
6059 }
6060
6061 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6062                        struct lu_device *dev)
6063 {
6064         struct osd_device *osd = osd_dev(dev);
6065         int                result = 0;
6066         ENTRY;
6067
6068         if (osd->od_quota_slave != NULL) {
6069                 /* set up quota slave objects */
6070                 result = qsd_prepare(env, osd->od_quota_slave);
6071                 if (result != 0)
6072                         RETURN(result);
6073         }
6074
6075         result = osd_fid_init(env, osd);
6076
6077         RETURN(result);
6078 }
6079
6080 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6081                   struct lu_fid *fid, struct md_op_data *op_data)
6082 {
6083         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6084
6085         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6086 }
6087
6088 static const struct lu_object_operations osd_lu_obj_ops = {
6089         .loo_object_init      = osd_object_init,
6090         .loo_object_delete    = osd_object_delete,
6091         .loo_object_release   = osd_object_release,
6092         .loo_object_free      = osd_object_free,
6093         .loo_object_print     = osd_object_print,
6094         .loo_object_invariant = osd_object_invariant
6095 };
6096
6097 const struct lu_device_operations osd_lu_ops = {
6098         .ldo_object_alloc      = osd_object_alloc,
6099         .ldo_process_config    = osd_process_config,
6100         .ldo_recovery_complete = osd_recovery_complete,
6101         .ldo_prepare           = osd_prepare,
6102 };
6103
6104 static const struct lu_device_type_operations osd_device_type_ops = {
6105         .ldto_init = osd_type_init,
6106         .ldto_fini = osd_type_fini,
6107
6108         .ldto_start = osd_type_start,
6109         .ldto_stop  = osd_type_stop,
6110
6111         .ldto_device_alloc = osd_device_alloc,
6112         .ldto_device_free  = osd_device_free,
6113
6114         .ldto_device_init    = osd_device_init,
6115         .ldto_device_fini    = osd_device_fini
6116 };
6117
6118 struct lu_device_type osd_device_type = {
6119         .ldt_tags     = LU_DEVICE_DT,
6120         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6121         .ldt_ops      = &osd_device_type_ops,
6122         .ldt_ctx_tags = LCT_LOCAL,
6123 };
6124
6125 /*
6126  * lprocfs legacy support.
6127  */
6128 static struct obd_ops osd_obd_device_ops = {
6129         .o_owner = THIS_MODULE,
6130         .o_connect      = osd_obd_connect,
6131         .o_disconnect   = osd_obd_disconnect,
6132         .o_fid_alloc    = osd_fid_alloc,
6133 };
6134
6135 static int __init osd_mod_init(void)
6136 {
6137         int rc;
6138
6139         osd_oi_mod_init();
6140
6141         rc = lu_kmem_init(ldiskfs_caches);
6142         if (rc)
6143                 return rc;
6144
6145         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6146                                  lprocfs_osd_module_vars,
6147 #ifndef HAVE_ONLY_PROCFS_SEQ
6148                                  NULL,
6149 #endif
6150                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6151         if (rc)
6152                 lu_kmem_fini(ldiskfs_caches);
6153         return rc;
6154 }
6155
6156 static void __exit osd_mod_exit(void)
6157 {
6158         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6159         lu_kmem_fini(ldiskfs_caches);
6160 }
6161
6162 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6163 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6164 MODULE_LICENSE("GPL");
6165
6166 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);