lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76
  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         LASSERT(ss != NULL);
1079         LASSERT(ss->ss_server_fld != NULL);
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 #ifdef I_LINKABLE
2806         /* This is necessary to increment from i_nlink == 0 */
2807         spin_lock(&inode->i_lock);
2808         inode->i_state |= I_LINKABLE;
2809         spin_unlock(&inode->i_lock);
2810 #endif
2811
2812         spin_lock(&obj->oo_guard);
2813         ldiskfs_inc_count(oh->ot_handle, inode);
2814         if (!S_ISDIR(inode->i_mode))
2815                 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2816         spin_unlock(&obj->oo_guard);
2817
2818         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2819         LINVRNT(osd_invariant(obj));
2820
2821         return rc;
2822 }
2823
2824 static int osd_declare_object_ref_del(const struct lu_env *env,
2825                                       struct dt_object *dt,
2826                                       struct thandle *handle)
2827 {
2828         struct osd_thandle *oh;
2829
2830         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2831         LASSERT(handle != NULL);
2832
2833         oh = container_of0(handle, struct osd_thandle, ot_super);
2834         LASSERT(oh->ot_handle == NULL);
2835
2836         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2837                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2838
2839         return 0;
2840 }
2841
2842 /*
2843  * Concurrency: @dt is write locked.
2844  */
2845 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2846                               struct thandle *th)
2847 {
2848         struct osd_object       *obj = osd_dt_obj(dt);
2849         struct inode            *inode = obj->oo_inode;
2850         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2851         struct osd_thandle      *oh;
2852
2853         LINVRNT(osd_invariant(obj));
2854         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2855         LASSERT(osd_write_locked(env, obj));
2856         LASSERT(th != NULL);
2857
2858         oh = container_of0(th, struct osd_thandle, ot_super);
2859         LASSERT(oh->ot_handle != NULL);
2860
2861         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2862
2863         spin_lock(&obj->oo_guard);
2864         /* That can be result of upgrade from old Lustre version and
2865          * applied only to local files.  Just skip this ref_del call.
2866          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2867         if (inode->i_nlink == 0) {
2868                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2869                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2870                              ", maybe an upgraded file? (LU-3915)\n",
2871                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2872                 spin_unlock(&obj->oo_guard);
2873                 return 0;
2874         }
2875
2876         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2877                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2878
2879         ldiskfs_dec_count(oh->ot_handle, inode);
2880         spin_unlock(&obj->oo_guard);
2881
2882         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2883         LINVRNT(osd_invariant(obj));
2884
2885         return 0;
2886 }
2887
2888 /*
2889  * Get the 64-bit version for an inode.
2890  */
2891 static int osd_object_version_get(const struct lu_env *env,
2892                                   struct dt_object *dt, dt_obj_version_t *ver)
2893 {
2894         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2895
2896         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2897                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2898         *ver = LDISKFS_I(inode)->i_fs_version;
2899         return 0;
2900 }
2901
2902 /*
2903  * Concurrency: @dt is read locked.
2904  */
2905 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2906                          struct lu_buf *buf, const char *name,
2907                          struct lustre_capa *capa)
2908 {
2909         struct osd_object      *obj    = osd_dt_obj(dt);
2910         struct inode           *inode  = obj->oo_inode;
2911         struct osd_thread_info *info   = osd_oti_get(env);
2912         struct dentry          *dentry = &info->oti_obj_dentry;
2913
2914         /* version get is not real XATTR but uses xattr API */
2915         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2916                 /* for version we are just using xattr API but change inode
2917                  * field instead */
2918                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2919                 osd_object_version_get(env, dt, buf->lb_buf);
2920                 return sizeof(dt_obj_version_t);
2921         }
2922
2923         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2924         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2925
2926         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2927                 return -EACCES;
2928
2929         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2930 }
2931
2932
2933 static int osd_declare_xattr_set(const struct lu_env *env,
2934                                  struct dt_object *dt,
2935                                  const struct lu_buf *buf, const char *name,
2936                                  int fl, struct thandle *handle)
2937 {
2938         struct osd_thandle *oh;
2939         int credits;
2940
2941         LASSERT(handle != NULL);
2942
2943         oh = container_of0(handle, struct osd_thandle, ot_super);
2944         LASSERT(oh->ot_handle == NULL);
2945
2946         /* optimistic optimization: LMA is set first and usually fit inode */
2947         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2948                 if (dt_object_exists(dt))
2949                         credits = 0;
2950                 else
2951                         credits = 1;
2952         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2953                 credits = 1;
2954         } else {
2955                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2956                 struct super_block *sb = osd_sb(osd);
2957                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2958                 if (buf && buf->lb_len > sb->s_blocksize) {
2959                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2960                                         sb->s_blocksize_bits;
2961                 }
2962         }
2963
2964         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2965
2966         return 0;
2967 }
2968
2969 /*
2970  * Set the 64-bit version for object
2971  */
2972 static void osd_object_version_set(const struct lu_env *env,
2973                                    struct dt_object *dt,
2974                                    dt_obj_version_t *new_version)
2975 {
2976         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2977
2978         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2979                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2980
2981         LDISKFS_I(inode)->i_fs_version = *new_version;
2982         /** Version is set after all inode operations are finished,
2983          *  so we should mark it dirty here */
2984         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2985 }
2986
2987 /*
2988  * Concurrency: @dt is write locked.
2989  */
2990 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2991                          const struct lu_buf *buf, const char *name, int fl,
2992                          struct thandle *handle, struct lustre_capa *capa)
2993 {
2994         struct osd_object      *obj      = osd_dt_obj(dt);
2995         struct inode           *inode    = obj->oo_inode;
2996         struct osd_thread_info *info     = osd_oti_get(env);
2997         int                     fs_flags = 0;
2998         ENTRY;
2999
3000         LASSERT(handle != NULL);
3001
3002         /* version set is not real XATTR */
3003         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3004                 /* for version we are just using xattr API but change inode
3005                  * field instead */
3006                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3007                 osd_object_version_set(env, dt, buf->lb_buf);
3008                 return sizeof(dt_obj_version_t);
3009         }
3010
3011         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3012                 return -EACCES;
3013
3014         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3015                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3016
3017         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3018         if (fl & LU_XATTR_REPLACE)
3019                 fs_flags |= XATTR_REPLACE;
3020
3021         if (fl & LU_XATTR_CREATE)
3022                 fs_flags |= XATTR_CREATE;
3023
3024         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3025                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3026                 int                      rc;
3027
3028                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3029                 if (rc != 0)
3030                         RETURN(rc);
3031
3032                 lma->lma_incompat |= LMAI_STRIPED;
3033                 lustre_lma_swab(lma);
3034                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3035                                      sizeof(*lma), XATTR_REPLACE);
3036                 if (rc != 0)
3037                         RETURN(rc);
3038         }
3039
3040         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3041                                fs_flags);
3042 }
3043
3044 /*
3045  * Concurrency: @dt is read locked.
3046  */
3047 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3048                           struct lu_buf *buf, struct lustre_capa *capa)
3049 {
3050         struct osd_object      *obj    = osd_dt_obj(dt);
3051         struct inode           *inode  = obj->oo_inode;
3052         struct osd_thread_info *info   = osd_oti_get(env);
3053         struct dentry          *dentry = &info->oti_obj_dentry;
3054
3055         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3056         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3057
3058         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3059                 return -EACCES;
3060
3061         dentry->d_inode = inode;
3062         dentry->d_sb = inode->i_sb;
3063         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3064 }
3065
3066 static int osd_declare_xattr_del(const struct lu_env *env,
3067                                  struct dt_object *dt, const char *name,
3068                                  struct thandle *handle)
3069 {
3070         struct osd_thandle *oh;
3071
3072         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3073         LASSERT(handle != NULL);
3074
3075         oh = container_of0(handle, struct osd_thandle, ot_super);
3076         LASSERT(oh->ot_handle == NULL);
3077
3078         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3079                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3080
3081         return 0;
3082 }
3083
3084 /*
3085  * Concurrency: @dt is write locked.
3086  */
3087 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3088                          const char *name, struct thandle *handle,
3089                          struct lustre_capa *capa)
3090 {
3091         struct osd_object      *obj    = osd_dt_obj(dt);
3092         struct inode           *inode  = obj->oo_inode;
3093         struct osd_thread_info *info   = osd_oti_get(env);
3094         struct dentry          *dentry = &info->oti_obj_dentry;
3095         int                     rc;
3096
3097         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3098         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3099         LASSERT(handle != NULL);
3100
3101         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3102                 return -EACCES;
3103
3104         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3105
3106         ll_vfs_dq_init(inode);
3107         dentry->d_inode = inode;
3108         dentry->d_sb = inode->i_sb;
3109         rc = inode->i_op->removexattr(dentry, name);
3110         return rc;
3111 }
3112
3113 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3114                                      struct dt_object *dt,
3115                                      struct lustre_capa *old, __u64 opc)
3116 {
3117         struct osd_thread_info *info = osd_oti_get(env);
3118         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3119         struct osd_object *obj = osd_dt_obj(dt);
3120         struct osd_device *osd = osd_obj2dev(obj);
3121         struct lustre_capa_key *key = &info->oti_capa_key;
3122         struct lustre_capa *capa = &info->oti_capa;
3123         struct obd_capa *oc;
3124         struct lu_capainfo *lci;
3125         int rc;
3126         ENTRY;
3127
3128         if (!osd->od_fl_capa)
3129                 RETURN(ERR_PTR(-ENOENT));
3130
3131         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3132         LINVRNT(osd_invariant(obj));
3133
3134         /* renewal sanity check */
3135         if (old && osd_object_auth(env, dt, old, opc))
3136                 RETURN(ERR_PTR(-EACCES));
3137
3138         lci = lu_capainfo_get(env);
3139         if (unlikely(lci == NULL))
3140                 RETURN(ERR_PTR(-ENOENT));
3141
3142         switch (lci->lci_auth) {
3143         case LC_ID_NONE:
3144                 RETURN(NULL);
3145         case LC_ID_PLAIN:
3146                 capa->lc_uid = i_uid_read(obj->oo_inode);
3147                 capa->lc_gid = i_gid_read(obj->oo_inode);
3148                 capa->lc_flags = LC_ID_PLAIN;
3149                 break;
3150         case LC_ID_CONVERT: {
3151                 __u32 d[4], s[4];
3152
3153                 s[0] = i_uid_read(obj->oo_inode);
3154                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3155                 s[2] = i_uid_read(obj->oo_inode);
3156                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3157                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3158                 if (unlikely(rc))
3159                         RETURN(ERR_PTR(rc));
3160
3161                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3162                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3163                 capa->lc_flags = LC_ID_CONVERT;
3164                 break;
3165         }
3166         default:
3167                 RETURN(ERR_PTR(-EINVAL));
3168         }
3169
3170         capa->lc_fid = *fid;
3171         capa->lc_opc = opc;
3172         capa->lc_flags |= osd->od_capa_alg << 24;
3173         capa->lc_timeout = osd->od_capa_timeout;
3174         capa->lc_expiry = 0;
3175
3176         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3177         if (oc) {
3178                 LASSERT(!capa_is_expired(oc));
3179                 RETURN(oc);
3180         }
3181
3182         spin_lock(&capa_lock);
3183         *key = osd->od_capa_keys[1];
3184         spin_unlock(&capa_lock);
3185
3186         capa->lc_keyid = key->lk_keyid;
3187         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3188
3189         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3190         if (rc) {
3191                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3192                 RETURN(ERR_PTR(rc));
3193         }
3194
3195         oc = capa_add(osd->od_capa_hash, capa);
3196         RETURN(oc);
3197 }
3198
3199 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3200                            __u64 start, __u64 end)
3201 {
3202         struct osd_object       *obj    = osd_dt_obj(dt);
3203         struct inode            *inode  = obj->oo_inode;
3204         struct osd_thread_info  *info   = osd_oti_get(env);
3205         struct dentry           *dentry = &info->oti_obj_dentry;
3206         struct file             *file   = &info->oti_file;
3207         int                     rc;
3208
3209         ENTRY;
3210
3211         dentry->d_inode = inode;
3212         dentry->d_sb = inode->i_sb;
3213         file->f_dentry = dentry;
3214         file->f_mapping = inode->i_mapping;
3215         file->f_op = inode->i_fop;
3216         set_file_inode(file, inode);
3217
3218 #ifdef HAVE_FILE_FSYNC_4ARGS
3219         rc = file->f_op->fsync(file, start, end, 0);
3220 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3221         mutex_lock(&inode->i_mutex);
3222         rc = file->f_op->fsync(file, 0);
3223         mutex_unlock(&inode->i_mutex);
3224 #else
3225         mutex_lock(&inode->i_mutex);
3226         rc = file->f_op->fsync(file, dentry, 0);
3227         mutex_unlock(&inode->i_mutex);
3228 #endif
3229
3230         RETURN(rc);
3231 }
3232
3233 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3234                         void **data)
3235 {
3236         struct osd_object *obj = osd_dt_obj(dt);
3237         ENTRY;
3238
3239         *data = (void *)obj->oo_inode;
3240         RETURN(0);
3241 }
3242
3243 /*
3244  * Index operations.
3245  */
3246
3247 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3248                            const struct dt_index_features *feat)
3249 {
3250         struct iam_descr *descr;
3251
3252         if (osd_object_is_root(o))
3253                 return feat == &dt_directory_features;
3254
3255         LASSERT(o->oo_dir != NULL);
3256
3257         descr = o->oo_dir->od_container.ic_descr;
3258         if (feat == &dt_directory_features) {
3259                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3260                         return 1;
3261                 else
3262                         return 0;
3263         } else {
3264                 return
3265                         feat->dif_keysize_min <= descr->id_key_size &&
3266                         descr->id_key_size <= feat->dif_keysize_max &&
3267                         feat->dif_recsize_min <= descr->id_rec_size &&
3268                         descr->id_rec_size <= feat->dif_recsize_max &&
3269                         !(feat->dif_flags & (DT_IND_VARKEY |
3270                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3271                         ergo(feat->dif_flags & DT_IND_UPDATE,
3272                              1 /* XXX check that object (and file system) is
3273                                 * writable */);
3274         }
3275 }
3276
3277 static int osd_iam_container_init(const struct lu_env *env,
3278                                   struct osd_object *obj,
3279                                   struct osd_directory *dir)
3280 {
3281         struct iam_container *bag = &dir->od_container;
3282         int result;
3283
3284         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3285         if (result != 0)
3286                 return result;
3287
3288         result = iam_container_setup(bag);
3289         if (result == 0)
3290                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3291         else
3292                 iam_container_fini(bag);
3293
3294         return result;
3295 }
3296
3297
3298 /*
3299  * Concurrency: no external locking is necessary.
3300  */
3301 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3302                          const struct dt_index_features *feat)
3303 {
3304         int                      result;
3305         int                      skip_iam = 0;
3306         struct osd_object       *obj = osd_dt_obj(dt);
3307
3308         LINVRNT(osd_invariant(obj));
3309
3310         if (osd_object_is_root(obj)) {
3311                 dt->do_index_ops = &osd_index_ea_ops;
3312                 result = 0;
3313         } else if (feat == &dt_directory_features) {
3314                 dt->do_index_ops = &osd_index_ea_ops;
3315                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3316                         result = 0;
3317                 else
3318                         result = -ENOTDIR;
3319                 skip_iam = 1;
3320         } else if (unlikely(feat == &dt_otable_features)) {
3321                 dt->do_index_ops = &osd_otable_ops;
3322                 return 0;
3323         } else if (unlikely(feat == &dt_acct_features)) {
3324                 dt->do_index_ops = &osd_acct_index_ops;
3325                 result = 0;
3326                 skip_iam = 1;
3327         } else if (!osd_has_index(obj)) {
3328                 struct osd_directory *dir;
3329
3330                 OBD_ALLOC_PTR(dir);
3331                 if (dir != NULL) {
3332
3333                         spin_lock(&obj->oo_guard);
3334                         if (obj->oo_dir == NULL)
3335                                 obj->oo_dir = dir;
3336                         else
3337                                 /*
3338                                  * Concurrent thread allocated container data.
3339                                  */
3340                                 OBD_FREE_PTR(dir);
3341                         spin_unlock(&obj->oo_guard);
3342                         /*
3343                          * Now, that we have container data, serialize its
3344                          * initialization.
3345                          */
3346                         down_write(&obj->oo_ext_idx_sem);
3347                         /*
3348                          * recheck under lock.
3349                          */
3350                         if (!osd_has_index(obj))
3351                                 result = osd_iam_container_init(env, obj,
3352                                                                 obj->oo_dir);
3353                         else
3354                                 result = 0;
3355                         up_write(&obj->oo_ext_idx_sem);
3356                 } else {
3357                         result = -ENOMEM;
3358                 }
3359         } else {
3360                 result = 0;
3361         }
3362
3363         if (result == 0 && skip_iam == 0) {
3364                 if (!osd_iam_index_probe(env, obj, feat))
3365                         result = -ENOTDIR;
3366         }
3367         LINVRNT(osd_invariant(obj));
3368
3369         if (result == 0 && is_quota_glb_feat(feat) &&
3370             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3371                 result = osd_quota_migration(env, dt, feat);
3372
3373         return result;
3374 }
3375
3376 static int osd_otable_it_attr_get(const struct lu_env *env,
3377                                  struct dt_object *dt,
3378                                  struct lu_attr *attr,
3379                                  struct lustre_capa *capa)
3380 {
3381         attr->la_valid = 0;
3382         return 0;
3383 }
3384
3385 static const struct dt_object_operations osd_obj_ops = {
3386         .do_read_lock         = osd_object_read_lock,
3387         .do_write_lock        = osd_object_write_lock,
3388         .do_read_unlock       = osd_object_read_unlock,
3389         .do_write_unlock      = osd_object_write_unlock,
3390         .do_write_locked      = osd_object_write_locked,
3391         .do_attr_get          = osd_attr_get,
3392         .do_declare_attr_set  = osd_declare_attr_set,
3393         .do_attr_set          = osd_attr_set,
3394         .do_ah_init           = osd_ah_init,
3395         .do_declare_create    = osd_declare_object_create,
3396         .do_create            = osd_object_create,
3397         .do_declare_destroy   = osd_declare_object_destroy,
3398         .do_destroy           = osd_object_destroy,
3399         .do_index_try         = osd_index_try,
3400         .do_declare_ref_add   = osd_declare_object_ref_add,
3401         .do_ref_add           = osd_object_ref_add,
3402         .do_declare_ref_del   = osd_declare_object_ref_del,
3403         .do_ref_del           = osd_object_ref_del,
3404         .do_xattr_get         = osd_xattr_get,
3405         .do_declare_xattr_set = osd_declare_xattr_set,
3406         .do_xattr_set         = osd_xattr_set,
3407         .do_declare_xattr_del = osd_declare_xattr_del,
3408         .do_xattr_del         = osd_xattr_del,
3409         .do_xattr_list        = osd_xattr_list,
3410         .do_capa_get          = osd_capa_get,
3411         .do_object_sync       = osd_object_sync,
3412         .do_data_get          = osd_data_get,
3413 };
3414
3415 /**
3416  * dt_object_operations for interoperability mode
3417  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3418  */
3419 static const struct dt_object_operations osd_obj_ea_ops = {
3420         .do_read_lock         = osd_object_read_lock,
3421         .do_write_lock        = osd_object_write_lock,
3422         .do_read_unlock       = osd_object_read_unlock,
3423         .do_write_unlock      = osd_object_write_unlock,
3424         .do_write_locked      = osd_object_write_locked,
3425         .do_attr_get          = osd_attr_get,
3426         .do_declare_attr_set  = osd_declare_attr_set,
3427         .do_attr_set          = osd_attr_set,
3428         .do_ah_init           = osd_ah_init,
3429         .do_declare_create    = osd_declare_object_create,
3430         .do_create            = osd_object_ea_create,
3431         .do_declare_destroy   = osd_declare_object_destroy,
3432         .do_destroy           = osd_object_destroy,
3433         .do_index_try         = osd_index_try,
3434         .do_declare_ref_add   = osd_declare_object_ref_add,
3435         .do_ref_add           = osd_object_ref_add,
3436         .do_declare_ref_del   = osd_declare_object_ref_del,
3437         .do_ref_del           = osd_object_ref_del,
3438         .do_xattr_get         = osd_xattr_get,
3439         .do_declare_xattr_set = osd_declare_xattr_set,
3440         .do_xattr_set         = osd_xattr_set,
3441         .do_declare_xattr_del = osd_declare_xattr_del,
3442         .do_xattr_del         = osd_xattr_del,
3443         .do_xattr_list        = osd_xattr_list,
3444         .do_capa_get          = osd_capa_get,
3445         .do_object_sync       = osd_object_sync,
3446         .do_data_get          = osd_data_get,
3447 };
3448
3449 static const struct dt_object_operations osd_obj_otable_it_ops = {
3450         .do_attr_get    = osd_otable_it_attr_get,
3451         .do_index_try   = osd_index_try,
3452 };
3453
3454 static int osd_index_declare_iam_delete(const struct lu_env *env,
3455                                         struct dt_object *dt,
3456                                         const struct dt_key *key,
3457                                         struct thandle *handle)
3458 {
3459         struct osd_thandle    *oh;
3460
3461         oh = container_of0(handle, struct osd_thandle, ot_super);
3462         LASSERT(oh->ot_handle == NULL);
3463
3464         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3465                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3466
3467         return 0;
3468 }
3469
3470 /**
3471  *      delete a (key, value) pair from index \a dt specified by \a key
3472  *
3473  *      \param  dt      osd index object
3474  *      \param  key     key for index
3475  *      \param  rec     record reference
3476  *      \param  handle  transaction handler
3477  *
3478  *      \retval  0  success
3479  *      \retval -ve   failure
3480  */
3481
3482 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3483                                 const struct dt_key *key,
3484                                 struct thandle *handle,
3485                                 struct lustre_capa *capa)
3486 {
3487         struct osd_thread_info *oti = osd_oti_get(env);
3488         struct osd_object      *obj = osd_dt_obj(dt);
3489         struct osd_thandle     *oh;
3490         struct iam_path_descr  *ipd;
3491         struct iam_container   *bag = &obj->oo_dir->od_container;
3492         int                     rc;
3493
3494         ENTRY;
3495
3496         LINVRNT(osd_invariant(obj));
3497         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3498         LASSERT(bag->ic_object == obj->oo_inode);
3499         LASSERT(handle != NULL);
3500
3501         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3502                 RETURN(-EACCES);
3503
3504         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3505
3506         ipd = osd_idx_ipd_get(env, bag);
3507         if (unlikely(ipd == NULL))
3508                 RETURN(-ENOMEM);
3509
3510         oh = container_of0(handle, struct osd_thandle, ot_super);
3511         LASSERT(oh->ot_handle != NULL);
3512         LASSERT(oh->ot_handle->h_transaction != NULL);
3513
3514         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3515                 /* swab quota uid/gid provided by caller */
3516                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3517                 key = (const struct dt_key *)&oti->oti_quota_id;
3518         }
3519
3520         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3521         osd_ipd_put(env, bag, ipd);
3522         LINVRNT(osd_invariant(obj));
3523         RETURN(rc);
3524 }
3525
3526 static int osd_index_declare_ea_delete(const struct lu_env *env,
3527                                        struct dt_object *dt,
3528                                        const struct dt_key *key,
3529                                        struct thandle *handle)
3530 {
3531         struct osd_thandle *oh;
3532         struct inode       *inode;
3533         int                 rc;
3534         ENTRY;
3535
3536         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3537         LASSERT(handle != NULL);
3538
3539         oh = container_of0(handle, struct osd_thandle, ot_super);
3540         LASSERT(oh->ot_handle == NULL);
3541
3542         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3543                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3544
3545         inode = osd_dt_obj(dt)->oo_inode;
3546         LASSERT(inode);
3547
3548         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3549                                    0, oh, true, true, NULL, false);
3550         RETURN(rc);
3551 }
3552
3553 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3554                                           struct dt_rec *fid)
3555 {
3556         struct osd_fid_pack *rec;
3557         int                  rc = -ENODATA;
3558
3559         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3560                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3561                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3562         }
3563         return rc;
3564 }
3565
3566 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3567                           struct lu_fid *fid)
3568 {
3569         struct seq_server_site  *ss = osd_seq_site(osd);
3570         ENTRY;
3571
3572         /* FID seqs not in FLDB, must be local seq */
3573         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3574                 RETURN(0);
3575
3576         /* If FLD is not being initialized yet, it only happens during the
3577          * initialization, likely during mgs initialization, and we assume
3578          * this is local FID. */
3579         if (ss == NULL || ss->ss_server_fld == NULL)
3580                 RETURN(0);
3581
3582         /* Only check the local FLDB here */
3583         if (osd_seq_exists(env, osd, fid_seq(fid)))
3584                 RETURN(0);
3585
3586         RETURN(1);
3587 }
3588
3589 /**
3590  * Index delete function for interoperability mode (b11826).
3591  * It will remove the directory entry added by osd_index_ea_insert().
3592  * This entry is needed to maintain name->fid mapping.
3593  *
3594  * \param key,  key i.e. file entry to be deleted
3595  *
3596  * \retval   0, on success
3597  * \retval -ve, on error
3598  */
3599 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3600                                const struct dt_key *key,
3601                                struct thandle *handle,
3602                                struct lustre_capa *capa)
3603 {
3604         struct osd_object          *obj    = osd_dt_obj(dt);
3605         struct inode               *dir    = obj->oo_inode;
3606         struct dentry              *dentry;
3607         struct osd_thandle         *oh;
3608         struct ldiskfs_dir_entry_2 *de = NULL;
3609         struct buffer_head         *bh;
3610         struct htree_lock          *hlock = NULL;
3611         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3612         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3613         int                        rc;
3614         ENTRY;
3615
3616         LINVRNT(osd_invariant(obj));
3617         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3618         LASSERT(handle != NULL);
3619
3620         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3621
3622         oh = container_of(handle, struct osd_thandle, ot_super);
3623         LASSERT(oh->ot_handle != NULL);
3624         LASSERT(oh->ot_handle->h_transaction != NULL);
3625
3626         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3627                 RETURN(-EACCES);
3628
3629         ll_vfs_dq_init(dir);
3630         dentry = osd_child_dentry_get(env, obj,
3631                                       (char *)key, strlen((char *)key));
3632
3633         if (obj->oo_hl_head != NULL) {
3634                 hlock = osd_oti_get(env)->oti_hlock;
3635                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3636                                    dir, LDISKFS_HLOCK_DEL);
3637         } else {
3638                 down_write(&obj->oo_ext_idx_sem);
3639         }
3640
3641         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3642         if (bh) {
3643                 __u32 ino = 0;
3644
3645                 /* If this is not the ".." entry, it might be a remote DNE
3646                  * entry and  we need to check if the FID is for a remote
3647                  * MDT.  If the FID is  not in the directory entry (e.g.
3648                  * upgraded 1.8 filesystem without dirdata enabled) then
3649                  * we need to get the FID from the LMA. For a remote directory
3650                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3651                  * case.
3652                  *
3653                  * Delete the entry before the agent inode in order to
3654                  * simplify error handling.  At worst an error after deleting
3655                  * the entry first might leak the agent inode afterward. The
3656                  * reverse would need filesystem abort in case of error deleting
3657                  * the entry after the agent had been removed, or leave a
3658                  * dangling entry pointing at a random inode. */
3659                 if (strcmp((char *)key, dotdot) != 0) {
3660                         LASSERT(de != NULL);
3661                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3662                         /* If Fid is not in dentry, try to get it from LMA */
3663                         if (rc == -ENODATA) {
3664                                 struct osd_inode_id *id;
3665                                 struct inode *inode;
3666
3667                                 /* Before trying to get fid from the inode,
3668                                  * check whether the inode is valid.
3669                                  *
3670                                  * If the inode has been deleted, do not go
3671                                  * ahead to do osd_ea_fid_get, which will set
3672                                  * the inode to bad inode, which might cause
3673                                  * the inode to be deleted uncorrectly */
3674                                 inode = ldiskfs_iget(osd_sb(osd),
3675                                                      le32_to_cpu(de->inode));
3676                                 if (IS_ERR(inode)) {
3677                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3678                                                "error.\n", osd_name(osd),
3679                                                PFID(fid));
3680                                         rc = PTR_ERR(inode);
3681                                 } else {
3682                                         if (likely(inode->i_nlink != 0)) {
3683                                                 id = &osd_oti_get(env)->oti_id;
3684                                                 rc = osd_ea_fid_get(env, obj,
3685                                                         le32_to_cpu(de->inode),
3686                                                                     fid, id);
3687                                         } else {
3688                                                 CDEBUG(D_INFO, "%s: %u "DFID
3689                                                        "deleted.\n",
3690                                                        osd_name(osd),
3691                                                        le32_to_cpu(de->inode),
3692                                                        PFID(fid));
3693                                                 rc = -ESTALE;
3694                                         }
3695                                         iput(inode);
3696                                 }
3697                         }
3698                         if (rc == 0 &&
3699                             unlikely(osd_remote_fid(env, osd, fid)))
3700                                 /* Need to delete agent inode */
3701                                 ino = le32_to_cpu(de->inode);
3702                 }
3703                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3704                 brelse(bh);
3705                 if (rc == 0 && unlikely(ino != 0)) {
3706                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3707                                                           oh);
3708                         if (rc != 0)
3709                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3710                                        osd_name(osd), PFID(fid), rc);
3711                 }
3712         } else {
3713                 rc = -ENOENT;
3714         }
3715         if (hlock != NULL)
3716                 ldiskfs_htree_unlock(hlock);
3717         else
3718                 up_write(&obj->oo_ext_idx_sem);
3719
3720         if (rc != 0)
3721                 GOTO(out, rc);
3722
3723         /* For inode on the remote MDT, .. will point to
3724          * /Agent directory, Check whether it needs to delete
3725          * from agent directory */
3726         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3727                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3728                                                    oh);
3729                 if (rc != 0 && rc != -ENOENT) {
3730                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3731                                osd_name(osd), PFID(fid), rc);
3732                 }
3733
3734                 if (rc == -ENOENT)
3735                         rc = 0;
3736
3737                 GOTO(out, rc);
3738         }
3739 out:
3740
3741         LASSERT(osd_invariant(obj));
3742         RETURN(rc);
3743 }
3744
3745 /**
3746  *      Lookup index for \a key and copy record to \a rec.
3747  *
3748  *      \param  dt      osd index object
3749  *      \param  key     key for index
3750  *      \param  rec     record reference
3751  *
3752  *      \retval  +ve  success : exact mach
3753  *      \retval  0    return record with key not greater than \a key
3754  *      \retval -ve   failure
3755  */
3756 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3757                                 struct dt_rec *rec, const struct dt_key *key,
3758                                 struct lustre_capa *capa)
3759 {
3760         struct osd_object      *obj = osd_dt_obj(dt);
3761         struct iam_path_descr  *ipd;
3762         struct iam_container   *bag = &obj->oo_dir->od_container;
3763         struct osd_thread_info *oti = osd_oti_get(env);
3764         struct iam_iterator    *it = &oti->oti_idx_it;
3765         struct iam_rec         *iam_rec;
3766         int                     rc;
3767
3768         ENTRY;
3769
3770         LASSERT(osd_invariant(obj));
3771         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3772         LASSERT(bag->ic_object == obj->oo_inode);
3773
3774         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3775                 RETURN(-EACCES);
3776
3777         ipd = osd_idx_ipd_get(env, bag);
3778         if (IS_ERR(ipd))
3779                 RETURN(-ENOMEM);
3780
3781         /* got ipd now we can start iterator. */
3782         iam_it_init(it, bag, 0, ipd);
3783
3784         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3785                 /* swab quota uid/gid provided by caller */
3786                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3787                 key = (const struct dt_key *)&oti->oti_quota_id;
3788         }
3789
3790         rc = iam_it_get(it, (struct iam_key *)key);
3791         if (rc >= 0) {
3792                 if (S_ISDIR(obj->oo_inode->i_mode))
3793                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3794                 else
3795                         iam_rec = (struct iam_rec *) rec;
3796
3797                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3798
3799                 if (S_ISDIR(obj->oo_inode->i_mode))
3800                         osd_fid_unpack((struct lu_fid *) rec,
3801                                        (struct osd_fid_pack *)iam_rec);
3802                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3803                         osd_quota_unpack(obj, rec);
3804         }
3805
3806         iam_it_put(it);
3807         iam_it_fini(it);
3808         osd_ipd_put(env, bag, ipd);
3809
3810         LINVRNT(osd_invariant(obj));
3811
3812         RETURN(rc);
3813 }
3814
3815 static int osd_index_declare_iam_insert(const struct lu_env *env,
3816                                         struct dt_object *dt,
3817                                         const struct dt_rec *rec,
3818                                         const struct dt_key *key,
3819                                         struct thandle *handle)
3820 {
3821         struct osd_thandle *oh;
3822
3823         LASSERT(handle != NULL);
3824
3825         oh = container_of0(handle, struct osd_thandle, ot_super);
3826         LASSERT(oh->ot_handle == NULL);
3827
3828         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3829                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3830
3831         return 0;
3832 }
3833
3834 /**
3835  *      Inserts (key, value) pair in \a dt index object.
3836  *
3837  *      \param  dt      osd index object
3838  *      \param  key     key for index
3839  *      \param  rec     record reference
3840  *      \param  th      transaction handler
3841  *
3842  *      \retval  0  success
3843  *      \retval -ve failure
3844  */
3845 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3846                                 const struct dt_rec *rec,
3847                                 const struct dt_key *key, struct thandle *th,
3848                                 struct lustre_capa *capa, int ignore_quota)
3849 {
3850         struct osd_object     *obj = osd_dt_obj(dt);
3851         struct iam_path_descr *ipd;
3852         struct osd_thandle    *oh;
3853         struct iam_container  *bag = &obj->oo_dir->od_container;
3854         struct osd_thread_info *oti = osd_oti_get(env);
3855         struct iam_rec         *iam_rec;
3856         int                     rc;
3857
3858         ENTRY;
3859
3860         LINVRNT(osd_invariant(obj));
3861         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3862         LASSERT(bag->ic_object == obj->oo_inode);
3863         LASSERT(th != NULL);
3864
3865         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3866                 RETURN(-EACCES);
3867
3868         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3869
3870         ipd = osd_idx_ipd_get(env, bag);
3871         if (unlikely(ipd == NULL))
3872                 RETURN(-ENOMEM);
3873
3874         oh = container_of0(th, struct osd_thandle, ot_super);
3875         LASSERT(oh->ot_handle != NULL);
3876         LASSERT(oh->ot_handle->h_transaction != NULL);
3877         if (S_ISDIR(obj->oo_inode->i_mode)) {
3878                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3879                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3880         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3881                 /* pack quota uid/gid */
3882                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3883                 key = (const struct dt_key *)&oti->oti_quota_id;
3884                 /* pack quota record */
3885                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3886                 iam_rec = (struct iam_rec *)rec;
3887         } else {
3888                 iam_rec = (struct iam_rec *)rec;
3889         }
3890
3891         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3892                         iam_rec, ipd);
3893         osd_ipd_put(env, bag, ipd);
3894         LINVRNT(osd_invariant(obj));
3895         RETURN(rc);
3896 }
3897
3898 /**
3899  * Calls ldiskfs_add_entry() to add directory entry
3900  * into the directory. This is required for
3901  * interoperability mode (b11826)
3902  *
3903  * \retval   0, on success
3904  * \retval -ve, on error
3905  */
3906 static int __osd_ea_add_rec(struct osd_thread_info *info,
3907                             struct osd_object *pobj, struct inode  *cinode,
3908                             const char *name, const struct dt_rec *fid,
3909                             struct htree_lock *hlock, struct thandle *th)
3910 {
3911         struct ldiskfs_dentry_param *ldp;
3912         struct dentry               *child;
3913         struct osd_thandle          *oth;
3914         int                          rc;
3915
3916         oth = container_of(th, struct osd_thandle, ot_super);
3917         LASSERT(oth->ot_handle != NULL);
3918         LASSERT(oth->ot_handle->h_transaction != NULL);
3919         LASSERT(pobj->oo_inode);
3920
3921         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3922         if (unlikely(pobj->oo_inode ==
3923                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3924                 ldp->edp_magic = 0;
3925         else
3926                 osd_get_ldiskfs_dirent_param(ldp, fid);
3927         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3928         child->d_fsdata = (void *)ldp;
3929         ll_vfs_dq_init(pobj->oo_inode);
3930         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3931
3932         RETURN(rc);
3933 }
3934
3935 /**
3936  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3937  * into the directory.Also sets flags into osd object to
3938  * indicate dot and dotdot are created. This is required for
3939  * interoperability mode (b11826)
3940  *
3941  * \param dir   directory for dot and dotdot fixup.
3942  * \param obj   child object for linking
3943  *
3944  * \retval   0, on success
3945  * \retval -ve, on error
3946  */
3947 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3948                               struct osd_object *dir,
3949                               struct inode  *parent_dir, const char *name,
3950                               const struct dt_rec *dot_fid,
3951                               const struct dt_rec *dot_dot_fid,
3952                               struct thandle *th)
3953 {
3954         struct inode                *inode = dir->oo_inode;
3955         struct osd_thandle          *oth;
3956         int result = 0;
3957
3958         oth = container_of(th, struct osd_thandle, ot_super);
3959         LASSERT(oth->ot_handle->h_transaction != NULL);
3960         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3961
3962         if (strcmp(name, dot) == 0) {
3963                 if (dir->oo_compat_dot_created) {
3964                         result = -EEXIST;
3965                 } else {
3966                         LASSERT(inode == parent_dir);
3967                         dir->oo_compat_dot_created = 1;
3968                         result = 0;
3969                 }
3970         } else if (strcmp(name, dotdot) == 0) {
3971                 if (!dir->oo_compat_dot_created)
3972                         return -EINVAL;
3973                 /* in case of rename, dotdot is already created */
3974                 if (dir->oo_compat_dotdot_created) {
3975                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3976                                                 dot_dot_fid, NULL, th);
3977                 }
3978
3979                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3980                                                 parent_dir, dot_fid,
3981                                                 dot_dot_fid, oth);
3982                 if (result == 0)
3983                         dir->oo_compat_dotdot_created = 1;
3984         }
3985
3986         return result;
3987 }
3988
3989
3990 /**
3991  * It will call the appropriate osd_add* function and return the
3992  * value, return by respective functions.
3993  */
3994 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3995                           struct inode *cinode, const char *name,
3996                           const struct dt_rec *fid, struct thandle *th)
3997 {
3998         struct osd_thread_info *info   = osd_oti_get(env);
3999         struct htree_lock      *hlock;
4000         int                     rc;
4001
4002         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4003
4004         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4005                                                    name[2] =='\0'))) {
4006                 if (hlock != NULL) {
4007                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4008                                            pobj->oo_inode, 0);
4009                 } else {
4010                         down_write(&pobj->oo_ext_idx_sem);
4011                 }
4012                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4013                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
4014                                         fid, th);
4015         } else {
4016                 if (hlock != NULL) {
4017                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4018                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4019                 } else {
4020                         down_write(&pobj->oo_ext_idx_sem);
4021                 }
4022
4023                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4024                         struct lu_fid *tfid = &info->oti_fid;
4025
4026                         *tfid = *(const struct lu_fid *)fid;
4027                         tfid->f_ver = ~0;
4028                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4029                                               (const struct dt_rec *)tfid,
4030                                               hlock, th);
4031                 } else {
4032                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4033                                               hlock, th);
4034                 }
4035         }
4036         if (hlock != NULL)
4037                 ldiskfs_htree_unlock(hlock);
4038         else
4039                 up_write(&pobj->oo_ext_idx_sem);
4040
4041         return rc;
4042 }
4043
4044 static void
4045 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4046                       struct osd_idmap_cache *oic)
4047 {
4048         struct osd_scrub    *scrub = &dev->od_scrub;
4049         struct lu_fid       *fid   = &oic->oic_fid;
4050         struct osd_inode_id *id    = &oti->oti_id;
4051         int                  once  = 0;
4052         int                  rc;
4053         ENTRY;
4054
4055         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4056                 RETURN_EXIT;
4057
4058 again:
4059         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4060         if (rc != 0 && rc != -ENOENT)
4061                 RETURN_EXIT;
4062
4063         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4064                 RETURN_EXIT;
4065
4066         if (thread_is_running(&scrub->os_thread)) {
4067                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4068                 /* There is race condition between osd_oi_lookup and OI scrub.
4069                  * The OI scrub finished just after osd_oi_lookup() failure.
4070                  * Under such case, it is unnecessary to trigger OI scrub again,
4071                  * but try to call osd_oi_lookup() again. */
4072                 if (unlikely(rc == -EAGAIN))
4073                         goto again;
4074
4075                 RETURN_EXIT;
4076         }
4077
4078         if (!dev->od_noscrub && ++once == 1) {
4079                 rc = osd_scrub_start(dev);
4080                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4081                               ", rc = %d [2]\n",
4082                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4083                               PFID(fid), rc);
4084                 if (rc == 0)
4085                         goto again;
4086         }
4087
4088         EXIT;
4089 }
4090
4091 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4092                                struct osd_device *dev,
4093                                struct osd_idmap_cache *oic,
4094                                struct lu_fid *fid, __u32 ino)
4095 {
4096         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4097         struct inode            *inode;
4098         int                      rc;
4099
4100         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4101         inode = osd_iget(oti, dev, &oic->oic_lid);
4102         if (IS_ERR(inode)) {
4103                 fid_zero(&oic->oic_fid);
4104                 return PTR_ERR(inode);
4105         }
4106
4107         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4108         iput(inode);
4109         if (rc != 0)
4110                 fid_zero(&oic->oic_fid);
4111         else
4112                 *fid = oic->oic_fid = lma->lma_self_fid;
4113         return rc;
4114 }
4115
4116 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4117                      struct osd_inode_id *id, const struct lu_fid *fid)
4118 {
4119         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4120                id->oii_ino, id->oii_gen, info);
4121         info->oti_cache.oic_lid = *id;
4122         info->oti_cache.oic_fid = *fid;
4123         info->oti_cache.oic_dev = osd;
4124
4125         return 0;
4126 }
4127
4128 /**
4129  * Calls ->lookup() to find dentry. From dentry get inode and
4130  * read inode's ea to get fid. This is required for  interoperability
4131  * mode (b11826)
4132  *
4133  * \retval   0, on success
4134  * \retval -ve, on error
4135  */
4136 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4137                              struct dt_rec *rec, const struct dt_key *key)
4138 {
4139         struct inode               *dir    = obj->oo_inode;
4140         struct dentry              *dentry;
4141         struct ldiskfs_dir_entry_2 *de;
4142         struct buffer_head         *bh;
4143         struct lu_fid              *fid = (struct lu_fid *) rec;
4144         struct htree_lock          *hlock = NULL;
4145         int                         ino;
4146         int                         rc;
4147         ENTRY;
4148
4149         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4150
4151         dentry = osd_child_dentry_get(env, obj,
4152                                       (char *)key, strlen((char *)key));
4153
4154         if (obj->oo_hl_head != NULL) {
4155                 hlock = osd_oti_get(env)->oti_hlock;
4156                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4157                                    dir, LDISKFS_HLOCK_LOOKUP);
4158         } else {
4159                 down_read(&obj->oo_ext_idx_sem);
4160         }
4161
4162         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4163         if (bh) {
4164                 struct osd_thread_info *oti = osd_oti_get(env);
4165                 struct osd_inode_id *id = &oti->oti_id;
4166                 struct osd_idmap_cache *oic = &oti->oti_cache;
4167                 struct osd_device *dev = osd_obj2dev(obj);
4168                 struct osd_scrub *scrub = &dev->od_scrub;
4169                 struct scrub_file *sf = &scrub->os_file;
4170
4171                 ino = le32_to_cpu(de->inode);
4172                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4173                         brelse(bh);
4174                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4175                         GOTO(out, rc);
4176                 }
4177
4178                 rc = osd_get_fid_from_dentry(de, rec);
4179
4180                 /* done with de, release bh */
4181                 brelse(bh);
4182                 if (rc != 0)
4183                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
4184                 else
4185                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4186                 if (rc != 0) {
4187                         fid_zero(&oic->oic_fid);
4188                         GOTO(out, rc);
4189                 }
4190
4191                 if (osd_remote_fid(env, dev, fid))
4192                         GOTO(out, rc = 0);
4193
4194                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4195                                       fid);
4196                 if (rc != 0)
4197                         GOTO(out, rc);
4198                 if ((scrub->os_pos_current <= ino) &&
4199                     ((sf->sf_flags & SF_INCONSISTENT) ||
4200                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4201                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4202                                       sf->sf_oi_bitmap)))
4203                         osd_consistency_check(oti, dev, oic);
4204         } else {
4205                 rc = -ENOENT;
4206         }
4207
4208         GOTO(out, rc);
4209
4210 out:
4211         if (hlock != NULL)
4212                 ldiskfs_htree_unlock(hlock);
4213         else
4214                 up_read(&obj->oo_ext_idx_sem);
4215         return rc;
4216 }
4217
4218 /**
4219  * Find the osd object for given fid.
4220  *
4221  * \param fid need to find the osd object having this fid
4222  *
4223  * \retval osd_object on success
4224  * \retval        -ve on error
4225  */
4226 struct osd_object *osd_object_find(const struct lu_env *env,
4227                                    struct dt_object *dt,
4228                                    const struct lu_fid *fid)
4229 {
4230         struct lu_device  *ludev = dt->do_lu.lo_dev;
4231         struct osd_object *child = NULL;
4232         struct lu_object  *luch;
4233         struct lu_object  *lo;
4234
4235         /*
4236          * at this point topdev might not exist yet
4237          * (i.e. MGS is preparing profiles). so we can
4238          * not rely on topdev and instead lookup with
4239          * our device passed as topdev. this can't work
4240          * if the object isn't cached yet (as osd doesn't
4241          * allocate lu_header). IOW, the object must be
4242          * in the cache, otherwise lu_object_alloc() crashes
4243          * -bzzz
4244          */
4245         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4246                                  ludev : ludev->ld_site->ls_top_dev,
4247                                  fid, NULL);
4248         if (!IS_ERR(luch)) {
4249                 if (lu_object_exists(luch)) {
4250                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4251                         if (lo != NULL)
4252                                 child = osd_obj(lo);
4253                         else
4254                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4255                                                 "lu_object can't be located"
4256                                                 DFID"\n", PFID(fid));
4257
4258                         if (child == NULL) {
4259                                 lu_object_put(env, luch);
4260                                 CERROR("Unable to get osd_object\n");
4261                                 child = ERR_PTR(-ENOENT);
4262                         }
4263                 } else {
4264                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4265                                         "lu_object does not exists "DFID"\n",
4266                                         PFID(fid));
4267                         lu_object_put(env, luch);
4268                         child = ERR_PTR(-ENOENT);
4269                 }
4270         } else {
4271                 child = ERR_CAST(luch);
4272         }
4273
4274         return child;
4275 }
4276
4277 /**
4278  * Put the osd object once done with it.
4279  *
4280  * \param obj osd object that needs to be put
4281  */
4282 static inline void osd_object_put(const struct lu_env *env,
4283                                   struct osd_object *obj)
4284 {
4285         lu_object_put(env, &obj->oo_dt.do_lu);
4286 }
4287
4288 static int osd_index_declare_ea_insert(const struct lu_env *env,
4289                                        struct dt_object *dt,
4290                                        const struct dt_rec *rec,
4291                                        const struct dt_key *key,
4292                                        struct thandle *handle)
4293 {
4294         struct osd_thandle      *oh;
4295         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4296         struct lu_fid           *fid = (struct lu_fid *)rec;
4297         int                     rc;
4298         ENTRY;
4299
4300         LASSERT(!dt_object_remote(dt));
4301         LASSERT(handle != NULL);
4302
4303         oh = container_of0(handle, struct osd_thandle, ot_super);
4304         LASSERT(oh->ot_handle == NULL);
4305
4306         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4307                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4308
4309         if (osd_dt_obj(dt)->oo_inode != NULL) {
4310                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4311
4312                 /* We ignore block quota on meta pool (MDTs), so needn't
4313                  * calculate how many blocks will be consumed by this index
4314                  * insert */
4315                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4316                                            i_gid_read(inode), 0,
4317                                            oh, true, true, NULL, false);
4318         }
4319
4320         if (fid == NULL)
4321                 RETURN(0);
4322
4323         rc = osd_remote_fid(env, osd, fid);
4324         if (rc <= 0)
4325                 RETURN(rc);
4326
4327         rc = 0;
4328
4329         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4330                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4331         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4332                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4333         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4334                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4335
4336         RETURN(rc);
4337 }
4338
4339 /**
4340  * Index add function for interoperability mode (b11826).
4341  * It will add the directory entry.This entry is needed to
4342  * maintain name->fid mapping.
4343  *
4344  * \param key it is key i.e. file entry to be inserted
4345  * \param rec it is value of given key i.e. fid
4346  *
4347  * \retval   0, on success
4348  * \retval -ve, on error
4349  */
4350 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4351                                const struct dt_rec *rec,
4352                                const struct dt_key *key, struct thandle *th,
4353                                struct lustre_capa *capa, int ignore_quota)
4354 {
4355         struct osd_object       *obj = osd_dt_obj(dt);
4356         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4357         struct lu_fid           *fid = (struct lu_fid *) rec;
4358         const char              *name = (const char *)key;
4359         struct osd_thread_info  *oti   = osd_oti_get(env);
4360         struct osd_inode_id     *id    = &oti->oti_id;
4361         struct inode            *child_inode = NULL;
4362         struct osd_object       *child = NULL;
4363         int                     rc;
4364         ENTRY;
4365
4366         LASSERT(osd_invariant(obj));
4367         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4368         LASSERT(th != NULL);
4369
4370         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4371
4372         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4373                 RETURN(-EACCES);
4374
4375         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4376
4377         rc = osd_remote_fid(env, osd, fid);
4378         if (rc < 0) {
4379                 CERROR("%s: Can not find object "DFID" rc %d\n",
4380                        osd_name(osd), PFID(fid), rc);
4381                 RETURN(rc);
4382         }
4383
4384         if (rc == 1) {
4385                 /* Insert remote entry */
4386                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4387                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4388                         struct osd_thandle      *oh;
4389
4390                         /* If parent on remote MDT, we need put this object
4391                          * under AGENT */
4392                         oh = container_of(th, typeof(*oh), ot_super);
4393                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4394                         if (rc != 0) {
4395                                 CERROR("%s: add "DFID" error: rc = %d\n",
4396                                        osd_name(osd),
4397                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4398                                 RETURN(rc);
4399                         }
4400
4401                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4402                 } else {
4403                         child_inode = osd_create_local_agent_inode(env, osd,
4404                                                                    obj, fid,
4405                                                                    th);
4406                         if (IS_ERR(child_inode))
4407                                 RETURN(PTR_ERR(child_inode));
4408                 }
4409         } else {
4410                 /* Insert local entry */
4411                 child = osd_object_find(env, dt, fid);
4412                 if (IS_ERR(child)) {
4413                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4414                                osd_name(osd), PFID(fid),
4415                                id->oii_ino, id->oii_gen,
4416                                (int)PTR_ERR(child));
4417                         RETURN(PTR_ERR(child));
4418                 }
4419                 child_inode = igrab(child->oo_inode);
4420         }
4421
4422         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4423
4424         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4425                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4426
4427         iput(child_inode);
4428         if (child != NULL)
4429                 osd_object_put(env, child);
4430         LASSERT(osd_invariant(obj));
4431         RETURN(rc);
4432 }
4433
4434 /**
4435  *  Initialize osd Iterator for given osd index object.
4436  *
4437  *  \param  dt      osd index object
4438  */
4439
4440 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4441                                      struct dt_object *dt,
4442                                      __u32 unused,
4443                                      struct lustre_capa *capa)
4444 {
4445         struct osd_it_iam      *it;
4446         struct osd_thread_info *oti = osd_oti_get(env);
4447         struct osd_object      *obj = osd_dt_obj(dt);
4448         struct lu_object       *lo  = &dt->do_lu;
4449         struct iam_path_descr  *ipd;
4450         struct iam_container   *bag = &obj->oo_dir->od_container;
4451
4452         LASSERT(lu_object_exists(lo));
4453
4454         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4455                 return ERR_PTR(-EACCES);
4456
4457         it = &oti->oti_it;
4458         ipd = osd_it_ipd_get(env, bag);
4459         if (likely(ipd != NULL)) {
4460                 it->oi_obj = obj;
4461                 it->oi_ipd = ipd;
4462                 lu_object_get(lo);
4463                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4464                 return (struct dt_it *)it;
4465         }
4466         return ERR_PTR(-ENOMEM);
4467 }
4468
4469 /**
4470  * free given Iterator.
4471  */
4472
4473 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4474 {
4475         struct osd_it_iam *it = (struct osd_it_iam *)di;
4476         struct osd_object *obj = it->oi_obj;
4477
4478         iam_it_fini(&it->oi_it);
4479         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4480         lu_object_put(env, &obj->oo_dt.do_lu);
4481 }
4482
4483 /**
4484  *  Move Iterator to record specified by \a key
4485  *
4486  *  \param  di      osd iterator
4487  *  \param  key     key for index
4488  *
4489  *  \retval +ve  di points to record with least key not larger than key
4490  *  \retval  0   di points to exact matched key
4491  *  \retval -ve  failure
4492  */
4493
4494 static int osd_it_iam_get(const struct lu_env *env,
4495                           struct dt_it *di, const struct dt_key *key)
4496 {
4497         struct osd_thread_info  *oti = osd_oti_get(env);
4498         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4499
4500         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4501                 /* swab quota uid/gid */
4502                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4503                 key = (struct dt_key *)&oti->oti_quota_id;
4504         }
4505
4506         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4507 }
4508
4509 /**
4510  *  Release Iterator
4511  *
4512  *  \param  di      osd iterator
4513  */
4514 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4515 {
4516         struct osd_it_iam *it = (struct osd_it_iam *)di;
4517
4518         iam_it_put(&it->oi_it);
4519 }
4520
4521 /**
4522  *  Move iterator by one record
4523  *
4524  *  \param  di      osd iterator
4525  *
4526  *  \retval +1   end of container reached
4527  *  \retval  0   success
4528  *  \retval -ve  failure
4529  */
4530
4531 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4532 {
4533         struct osd_it_iam *it = (struct osd_it_iam *)di;
4534
4535         return iam_it_next(&it->oi_it);
4536 }
4537
4538 /**
4539  * Return pointer to the key under iterator.
4540  */
4541
4542 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4543                                  const struct dt_it *di)
4544 {
4545         struct osd_thread_info *oti = osd_oti_get(env);
4546         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4547         struct osd_object      *obj = it->oi_obj;
4548         struct dt_key          *key;
4549
4550         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4551
4552         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4553                 /* swab quota uid/gid */
4554                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4555                 key = (struct dt_key *)&oti->oti_quota_id;
4556         }
4557
4558         return key;
4559 }
4560
4561 /**
4562  * Return size of key under iterator (in bytes)
4563  */
4564
4565 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4566 {
4567         struct osd_it_iam *it = (struct osd_it_iam *)di;
4568
4569         return iam_it_key_size(&it->oi_it);
4570 }
4571
4572 static inline void
4573 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4574 {
4575         /* check if file type is required */
4576         if (ent->lde_attrs & LUDA_TYPE) {
4577                 struct luda_type *lt;
4578                 int align = sizeof(*lt) - 1;
4579
4580                 len = (len + align) & ~align;
4581                 lt = (struct luda_type *)(ent->lde_name + len);
4582                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4583         }
4584
4585         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4586 }
4587
4588 /**
4589  * build lu direct from backend fs dirent.
4590  */
4591
4592 static inline void
4593 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4594                    char *name, __u16 namelen, __u16 type, __u32 attr)
4595 {
4596         ent->lde_attrs = attr | LUDA_FID;
4597         fid_cpu_to_le(&ent->lde_fid, fid);
4598
4599         ent->lde_hash = cpu_to_le64(offset);
4600         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4601
4602         strncpy(ent->lde_name, name, namelen);
4603         ent->lde_name[namelen] = '\0';
4604         ent->lde_namelen = cpu_to_le16(namelen);
4605
4606         /* append lustre attributes */
4607         osd_it_append_attrs(ent, namelen, type);
4608 }
4609
4610 /**
4611  * Return pointer to the record under iterator.
4612  */
4613 static int osd_it_iam_rec(const struct lu_env *env,
4614                           const struct dt_it *di,
4615                           struct dt_rec *dtrec, __u32 attr)
4616 {
4617         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4618         struct osd_thread_info *info = osd_oti_get(env);
4619         ENTRY;
4620
4621         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4622                 const struct osd_fid_pack *rec;
4623                 struct lu_fid             *fid = &info->oti_fid;
4624                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4625                 char                      *name;
4626                 int                        namelen;
4627                 __u64                      hash;
4628                 int                        rc;
4629
4630                 name = (char *)iam_it_key_get(&it->oi_it);
4631                 if (IS_ERR(name))
4632                         RETURN(PTR_ERR(name));
4633
4634                 namelen = iam_it_key_size(&it->oi_it);
4635
4636                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4637                 if (IS_ERR(rec))
4638                         RETURN(PTR_ERR(rec));
4639
4640                 rc = osd_fid_unpack(fid, rec);
4641                 if (rc)
4642                         RETURN(rc);
4643
4644                 hash = iam_it_store(&it->oi_it);
4645
4646                 /* IAM does not store object type in IAM index (dir) */
4647                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4648                                    0, LUDA_FID);
4649         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4650                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4651                            (struct iam_rec *)dtrec);
4652                 osd_quota_unpack(it->oi_obj, dtrec);
4653         } else {
4654                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4655                            (struct iam_rec *)dtrec);
4656         }
4657
4658         RETURN(0);
4659 }
4660
4661 /**
4662  * Returns cookie for current Iterator position.
4663  */
4664 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4665 {
4666         struct osd_it_iam *it = (struct osd_it_iam *)di;
4667
4668         return iam_it_store(&it->oi_it);
4669 }
4670
4671 /**
4672  * Restore iterator from cookie.
4673  *
4674  * \param  di      osd iterator
4675  * \param  hash    Iterator location cookie
4676  *
4677  * \retval +ve  di points to record with least key not larger than key.
4678  * \retval  0   di points to exact matched key
4679  * \retval -ve  failure
4680  */
4681
4682 static int osd_it_iam_load(const struct lu_env *env,
4683                            const struct dt_it *di, __u64 hash)
4684 {
4685         struct osd_it_iam *it = (struct osd_it_iam *)di;
4686
4687         return iam_it_load(&it->oi_it, hash);
4688 }
4689
4690 static const struct dt_index_operations osd_index_iam_ops = {
4691         .dio_lookup         = osd_index_iam_lookup,
4692         .dio_declare_insert = osd_index_declare_iam_insert,
4693         .dio_insert         = osd_index_iam_insert,
4694         .dio_declare_delete = osd_index_declare_iam_delete,
4695         .dio_delete         = osd_index_iam_delete,
4696         .dio_it     = {
4697                 .init     = osd_it_iam_init,
4698                 .fini     = osd_it_iam_fini,
4699                 .get      = osd_it_iam_get,
4700                 .put      = osd_it_iam_put,
4701                 .next     = osd_it_iam_next,
4702                 .key      = osd_it_iam_key,
4703                 .key_size = osd_it_iam_key_size,
4704                 .rec      = osd_it_iam_rec,
4705                 .store    = osd_it_iam_store,
4706                 .load     = osd_it_iam_load
4707         }
4708 };
4709
4710
4711 /**
4712  * Creates or initializes iterator context.
4713  *
4714  * \retval struct osd_it_ea, iterator structure on success
4715  *
4716  */
4717 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4718                                     struct dt_object *dt,
4719                                     __u32 attr,
4720                                     struct lustre_capa *capa)
4721 {
4722         struct osd_object       *obj  = osd_dt_obj(dt);
4723         struct osd_thread_info  *info = osd_oti_get(env);
4724         struct osd_it_ea        *it   = &info->oti_it_ea;
4725         struct file             *file = &it->oie_file;
4726         struct lu_object        *lo   = &dt->do_lu;
4727         struct dentry           *obj_dentry = &info->oti_it_dentry;
4728         ENTRY;
4729         LASSERT(lu_object_exists(lo));
4730
4731         obj_dentry->d_inode = obj->oo_inode;
4732         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4733         obj_dentry->d_name.hash = 0;
4734
4735         it->oie_rd_dirent       = 0;
4736         it->oie_it_dirent       = 0;
4737         it->oie_dirent          = NULL;
4738         it->oie_buf             = info->oti_it_ea_buf;
4739         it->oie_obj             = obj;
4740
4741         /* Reset the "file" totally to avoid to reuse any old value from
4742          * former readdir handling, the "file->f_pos" should be zero. */
4743         memset(file, 0, sizeof(*file));
4744         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4745         if (attr & LUDA_64BITHASH)
4746                 file->f_mode    = FMODE_64BITHASH;
4747         else
4748                 file->f_mode    = FMODE_32BITHASH;
4749         file->f_dentry          = obj_dentry;
4750         file->f_mapping         = obj->oo_inode->i_mapping;
4751         file->f_op              = obj->oo_inode->i_fop;
4752         set_file_inode(file, obj->oo_inode);
4753
4754         lu_object_get(lo);
4755         RETURN((struct dt_it *) it);
4756 }
4757
4758 /**
4759  * Destroy or finishes iterator context.
4760  *
4761  * \param di iterator structure to be destroyed
4762  */
4763 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4764 {
4765         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4766         struct osd_object    *obj  = it->oie_obj;
4767         struct inode       *inode  = obj->oo_inode;
4768
4769         ENTRY;
4770         it->oie_file.f_op->release(inode, &it->oie_file);
4771         lu_object_put(env, &obj->oo_dt.do_lu);
4772         EXIT;
4773 }
4774
4775 /**
4776  * It position the iterator at given key, so that next lookup continues from
4777  * that key Or it is similar to dio_it->load() but based on a key,
4778  * rather than file position.
4779  *
4780  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4781  * to the beginning.
4782  *
4783  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4784  */
4785 static int osd_it_ea_get(const struct lu_env *env,
4786                          struct dt_it *di, const struct dt_key *key)
4787 {
4788         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4789
4790         ENTRY;
4791         LASSERT(((const char *)key)[0] == '\0');
4792         it->oie_file.f_pos      = 0;
4793         it->oie_rd_dirent       = 0;
4794         it->oie_it_dirent       = 0;
4795         it->oie_dirent          = NULL;
4796
4797         RETURN(+1);
4798 }
4799
4800 /**
4801  * Does nothing
4802  */
4803 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4804 {
4805 }
4806
4807 struct osd_filldir_cbs {
4808 #ifdef HAVE_DIR_CONTEXT
4809         struct dir_context ctx;
4810 #endif
4811         struct osd_it_ea  *it;
4812 };
4813 /**
4814  * It is called internally by ->readdir(). It fills the
4815  * iterator's in-memory data structure with required
4816  * information i.e. name, namelen, rec_size etc.
4817  *
4818  * \param buf in which information to be filled in.
4819  * \param name name of the file in given dir
4820  *
4821  * \retval 0 on success
4822  * \retval 1 on buffer full
4823  */
4824 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4825                                loff_t offset, __u64 ino,
4826                                unsigned d_type)
4827 {
4828         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4829         struct osd_object       *obj  = it->oie_obj;
4830         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4831         struct lu_fid           *fid  = &ent->oied_fid;
4832         struct osd_fid_pack     *rec;
4833         ENTRY;
4834
4835         /* this should never happen */
4836         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4837                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4838                 RETURN(-EIO);
4839         }
4840
4841         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4842             OSD_IT_EA_BUFSIZE)
4843                 RETURN(1);
4844
4845         /* "." is just the object itself. */
4846         if (namelen == 1 && name[0] == '.') {
4847                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4848         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4849                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4850                 if (osd_fid_unpack(fid, rec) != 0)
4851                         fid_zero(fid);
4852         } else {
4853                 fid_zero(fid);
4854         }
4855         d_type &= ~LDISKFS_DIRENT_LUFID;
4856
4857         /* NOT export local root. */
4858         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4859                 ino = obj->oo_inode->i_ino;
4860                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4861         }
4862
4863         ent->oied_ino     = ino;
4864         ent->oied_off     = offset;
4865         ent->oied_namelen = namelen;
4866         ent->oied_type    = d_type;
4867
4868         memcpy(ent->oied_name, name, namelen);
4869
4870         it->oie_rd_dirent++;
4871         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4872         RETURN(0);
4873 }
4874
4875 /**
4876  * Calls ->readdir() to load a directory entry at a time
4877  * and stored it in iterator's in-memory data structure.
4878  *
4879  * \param di iterator's in memory structure
4880  *
4881  * \retval   0 on success
4882  * \retval -ve on error
4883  * \retval +1 reach the end of entry
4884  */
4885 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4886                                const struct dt_it *di)
4887 {
4888         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4889         struct osd_object  *obj   = it->oie_obj;
4890         struct inode       *inode = obj->oo_inode;
4891         struct htree_lock  *hlock = NULL;
4892         struct file        *filp  = &it->oie_file;
4893         int                 rc = 0;
4894         struct osd_filldir_cbs buf = {
4895 #ifdef HAVE_DIR_CONTEXT
4896                 .ctx.actor = osd_ldiskfs_filldir,
4897 #endif
4898                 .it = it
4899         };
4900
4901         ENTRY;
4902         it->oie_dirent = it->oie_buf;
4903         it->oie_rd_dirent = 0;
4904
4905         if (obj->oo_hl_head != NULL) {
4906                 hlock = osd_oti_get(env)->oti_hlock;
4907                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4908                                    inode, LDISKFS_HLOCK_READDIR);
4909         } else {
4910                 down_read(&obj->oo_ext_idx_sem);
4911         }
4912
4913 #ifdef HAVE_DIR_CONTEXT
4914         buf.ctx.pos = filp->f_pos;
4915         rc = inode->i_fop->iterate(filp, &buf.ctx);
4916         filp->f_pos = buf.ctx.pos;
4917 #else
4918         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4919 #endif
4920
4921         if (hlock != NULL)
4922                 ldiskfs_htree_unlock(hlock);
4923         else
4924                 up_read(&obj->oo_ext_idx_sem);
4925
4926         if (it->oie_rd_dirent == 0) {
4927                 /*If it does not get any dirent, it means it has been reached
4928                  *to the end of the dir */
4929                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4930                 if (rc == 0)
4931                         rc = 1;
4932         } else {
4933                 it->oie_dirent = it->oie_buf;
4934                 it->oie_it_dirent = 1;
4935         }
4936
4937         RETURN(rc);
4938 }
4939
4940 /**
4941  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4942  * to load a directory entry at a time and stored it in
4943  * iterator's in-memory data structure.
4944  *
4945  * \param di iterator's in memory structure
4946  *
4947  * \retval +ve iterator reached to end
4948  * \retval   0 iterator not reached to end
4949  * \retval -ve on error
4950  */
4951 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4952 {
4953         struct osd_it_ea *it = (struct osd_it_ea *)di;
4954         int rc;
4955
4956         ENTRY;
4957
4958         if (it->oie_it_dirent < it->oie_rd_dirent) {
4959                 it->oie_dirent =
4960                         (void *) it->oie_dirent +
4961                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4962                                        it->oie_dirent->oied_namelen);
4963                 it->oie_it_dirent++;
4964                 RETURN(0);
4965         } else {
4966                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4967                         rc = +1;
4968                 else
4969                         rc = osd_ldiskfs_it_fill(env, di);
4970         }
4971
4972         RETURN(rc);
4973 }
4974
4975 /**
4976  * Returns the key at current position from iterator's in memory structure.
4977  *
4978  * \param di iterator's in memory structure
4979  *
4980  * \retval key i.e. struct dt_key on success
4981  */
4982 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4983                                     const struct dt_it *di)
4984 {
4985         struct osd_it_ea *it = (struct osd_it_ea *)di;
4986
4987         return (struct dt_key *)it->oie_dirent->oied_name;
4988 }
4989
4990 /**
4991  * Returns the key's size at current position from iterator's in memory structure.
4992  *
4993  * \param di iterator's in memory structure
4994  *
4995  * \retval key_size i.e. struct dt_key on success
4996  */
4997 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4998 {
4999         struct osd_it_ea *it = (struct osd_it_ea *)di;
5000
5001         return it->oie_dirent->oied_namelen;
5002 }
5003
5004 static int
5005 osd_dirent_update(handle_t *jh, struct super_block *sb,
5006                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5007                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5008 {
5009         struct osd_fid_pack *rec;
5010         int                  rc;
5011         ENTRY;
5012
5013         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5014         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5015
5016         rc = ldiskfs_journal_get_write_access(jh, bh);
5017         if (rc != 0)
5018                 RETURN(rc);
5019
5020         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5021         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5022         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5023
5024         RETURN(rc);
5025 }
5026
5027 static inline int
5028 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5029 {
5030         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5031             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5032                 return 1;
5033         else
5034                 return 0;
5035 }
5036
5037 static inline int
5038 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5039 {
5040         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5041                  "dot_dotdot = %d\n", dot_dotdot);
5042
5043         if (LDISKFS_DIR_REC_LEN(de) >=
5044             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5045                 return 1;
5046         else
5047                 return 0;
5048 }
5049
5050 static int
5051 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5052                     struct inode *dir, struct inode *inode,
5053                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5054                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5055                     struct htree_lock *hlock)
5056 {
5057         struct dentry               *dentry;
5058         struct osd_fid_pack         *rec;
5059         struct ldiskfs_dentry_param *ldp;
5060         int                          rc;
5061         ENTRY;
5062
5063         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5064                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5065                 RETURN(0);
5066
5067         /* There is enough space to hold the FID-in-dirent. */
5068         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5069                                  dir->i_sb->s_blocksize)) {
5070                 rc = ldiskfs_journal_get_write_access(jh, bh);
5071                 if (rc != 0)
5072                         RETURN(rc);
5073
5074                 de->name[de->name_len] = 0;
5075                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5076                 rec->fp_len = sizeof(struct lu_fid) + 1;
5077                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5078                 de->file_type |= LDISKFS_DIRENT_LUFID;
5079
5080                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5081
5082                 RETURN(rc);
5083         }
5084
5085         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5086         if (rc != 0)
5087                 RETURN(rc);
5088
5089         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5090                                            ent->oied_namelen);
5091         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5092         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5093         dentry->d_fsdata = (void *)ldp;
5094         ll_vfs_dq_init(dir);
5095         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5096         /* It is too bad, we cannot reinsert the name entry back.
5097          * That means we lose it! */
5098         if (rc != 0)
5099                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5100                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5101                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5102                        dir->i_ino, dir->i_generation,
5103                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5104
5105         RETURN(rc);
5106 }
5107
5108 static int
5109 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5110                         struct osd_it_ea *it, struct lu_fid *fid,
5111                         struct osd_inode_id *id, __u32 *attr)
5112 {
5113         struct osd_thread_info     *info        = osd_oti_get(env);
5114         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5115         struct osd_device          *dev         = osd_obj2dev(obj);
5116         struct super_block         *sb          = osd_sb(dev);
5117         const char                 *devname     =
5118                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5119         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5120         struct inode               *dir         = obj->oo_inode;
5121         struct htree_lock          *hlock       = NULL;
5122         struct buffer_head         *bh          = NULL;
5123         handle_t                   *jh          = NULL;
5124         struct ldiskfs_dir_entry_2 *de;
5125         struct dentry              *dentry;
5126         struct inode               *inode;
5127         int                         credits;
5128         int                         rc;
5129         int                         dot_dotdot  = 0;
5130         bool                        dirty       = false;
5131         ENTRY;
5132
5133         if (ent->oied_name[0] == '.') {
5134                 if (ent->oied_namelen == 1)
5135                         dot_dotdot = 1;
5136                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5137                         dot_dotdot = 2;
5138         }
5139
5140         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5141                                       ent->oied_namelen);
5142
5143         /* We need to ensure that the name entry is still valid.
5144          * Because it may be removed or renamed by other already.
5145          *
5146          * The unlink or rename operation will start journal before PDO lock,
5147          * so to avoid deadlock, here we need to start journal handle before
5148          * related PDO lock also. But because we do not know whether there
5149          * will be something to be repaired before PDO lock, we just start
5150          * journal without conditions.
5151          *
5152          * We may need to remove the name entry firstly, then insert back.
5153          * One credit is for user quota file update.
5154          * One credit is for group quota file update.
5155          * Two credits are for dirty inode. */
5156         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5157                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5158
5159 again:
5160         if (dev->od_dirent_journal) {
5161                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5162                 if (IS_ERR(jh)) {
5163                         rc = PTR_ERR(jh);
5164                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5165                                "check_repair, dir = %lu/%u, credits = %d, "
5166                                "name = %.*s: rc = %d\n",
5167                                devname, dir->i_ino, dir->i_generation, credits,
5168                                ent->oied_namelen, ent->oied_name, rc);
5169                         RETURN(rc);
5170                 }
5171
5172                 if (obj->oo_hl_head != NULL) {
5173                         hlock = osd_oti_get(env)->oti_hlock;
5174                         /* "0" means exclusive lock for the whole directory.
5175                          * We need to prevent others access such name entry
5176                          * during the delete + insert. Neither HLOCK_ADD nor
5177                          * HLOCK_DEL cannot guarantee the atomicity. */
5178                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5179                 } else {
5180                         down_write(&obj->oo_ext_idx_sem);
5181                 }
5182         } else {
5183                 if (obj->oo_hl_head != NULL) {
5184                         hlock = osd_oti_get(env)->oti_hlock;
5185                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5186                                            LDISKFS_HLOCK_LOOKUP);
5187                 } else {
5188                         down_read(&obj->oo_ext_idx_sem);
5189                 }
5190         }
5191
5192         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5193         /* For dot/dotdot entry, if there is not enough space to hold the
5194          * FID-in-dirent, just keep them there. It only happens when the
5195          * device upgraded from 1.8 or restored from MDT file-level backup.
5196          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5197          * and needs to get FID from LMA when readdir, it will not affect the
5198          * performance much. */
5199         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5200             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5201                 *attr |= LUDA_IGNORE;
5202                 GOTO(out_journal, rc = 0);
5203         }
5204
5205         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5206         inode = osd_iget(info, dev, id);
5207         if (IS_ERR(inode)) {
5208                 rc = PTR_ERR(inode);
5209                 if (rc == -ENOENT || rc == -ESTALE) {
5210                         *attr |= LUDA_IGNORE;
5211                         rc = 0;
5212                 } else {
5213                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5214                                "check_repair, dir = %lu/%u, name = %.*s: "
5215                                "rc = %d\n",
5216                                devname, dir->i_ino, dir->i_generation,
5217                                ent->oied_namelen, ent->oied_name, rc);
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                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5261                                        "in the dirent, dir = %lu/%u, "
5262                                        "name = %.*s, "DFID": rc = %d\n",
5263                                        devname, dir->i_ino, dir->i_generation,
5264                                        ent->oied_namelen, ent->oied_name,
5265                                        PFID(fid), rc);
5266                 } else {
5267                         /* Do not repair under dryrun mode. */
5268                         if (*attr & LUDA_VERIFY_DRYRUN) {
5269                                 *fid = lma->lma_self_fid;
5270                                 *attr |= LUDA_REPAIR;
5271                                 GOTO(out_inode, rc = 0);
5272                         }
5273
5274                         if (!dev->od_dirent_journal) {
5275                                 iput(inode);
5276                                 brelse(bh);
5277                                 if (hlock != NULL)
5278                                         ldiskfs_htree_unlock(hlock);
5279                                 else
5280                                         up_read(&obj->oo_ext_idx_sem);
5281                                 dev->od_dirent_journal = 1;
5282                                 goto again;
5283                         }
5284
5285                         *fid = lma->lma_self_fid;
5286                         dirty = true;
5287                         /* Append the FID-in-dirent. */
5288                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5289                                                  fid, bh, de, hlock);
5290                         if (rc == 0)
5291                                 *attr |= LUDA_REPAIR;
5292                         else
5293                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5294                                        "after the dirent, dir = %lu/%u, "
5295                                        "name = %.*s, "DFID": rc = %d\n",
5296                                        devname, dir->i_ino, dir->i_generation,
5297                                        ent->oied_namelen, ent->oied_name,
5298                                        PFID(fid), rc);
5299                 }
5300         } else if (rc == -ENODATA) {
5301                 /* Do not repair under dryrun mode. */
5302                 if (*attr & LUDA_VERIFY_DRYRUN) {
5303                         if (fid_is_sane(fid)) {
5304                                 *attr |= LUDA_REPAIR;
5305                         } else {
5306                                 lu_igif_build(fid, inode->i_ino,
5307                                               inode->i_generation);
5308                                 *attr |= LUDA_UPGRADE;
5309                         }
5310                         GOTO(out_inode, rc = 0);
5311                 }
5312
5313                 if (!dev->od_dirent_journal) {
5314                         iput(inode);
5315                         brelse(bh);
5316                         if (hlock != NULL)
5317                                 ldiskfs_htree_unlock(hlock);
5318                         else
5319                                 up_read(&obj->oo_ext_idx_sem);
5320                         dev->od_dirent_journal = 1;
5321                         goto again;
5322                 }
5323
5324                 dirty = true;
5325                 if (unlikely(fid_is_sane(fid))) {
5326                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5327                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5328                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5329                         if (rc == 0)
5330                                 *attr |= LUDA_REPAIR;
5331                         else
5332                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5333                                        "update dirent, dir = %lu/%u, "
5334                                        "name = %.*s, "DFID": rc = %d\n",
5335                                        devname, dir->i_ino, dir->i_generation,
5336                                        ent->oied_namelen, ent->oied_name,
5337                                        PFID(fid), rc);
5338                 } else {
5339                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5340                         /* It is probably IGIF object. Only aappend the
5341                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5342                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5343                                                  fid, bh, de, hlock);
5344                         if (rc == 0)
5345                                 *attr |= LUDA_UPGRADE;
5346                         else
5347                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5348                                        "after the dirent, dir = %lu/%u, "
5349                                        "name = %.*s, "DFID": rc = %d\n",
5350                                        devname, dir->i_ino, dir->i_generation,
5351                                        ent->oied_namelen, ent->oied_name,
5352                                        PFID(fid), rc);
5353                 }
5354         }
5355
5356         GOTO(out_inode, rc);
5357
5358 out_inode:
5359         iput(inode);
5360
5361 out_journal:
5362         brelse(bh);
5363         if (hlock != NULL) {
5364                 ldiskfs_htree_unlock(hlock);
5365         } else {
5366                 if (dev->od_dirent_journal)
5367                         up_write(&obj->oo_ext_idx_sem);
5368                 else
5369                         up_read(&obj->oo_ext_idx_sem);
5370         }
5371         if (jh != NULL)
5372                 ldiskfs_journal_stop(jh);
5373         if (rc >= 0 && !dirty)
5374                 dev->od_dirent_journal = 0;
5375         return rc;
5376 }
5377
5378 /**
5379  * Returns the value at current position from iterator's in memory structure.
5380  *
5381  * \param di struct osd_it_ea, iterator's in memory structure
5382  * \param attr attr requested for dirent.
5383  * \param lde lustre dirent
5384  *
5385  * \retval   0 no error and \param lde has correct lustre dirent.
5386  * \retval -ve on error
5387  */
5388 static inline int osd_it_ea_rec(const struct lu_env *env,
5389                                 const struct dt_it *di,
5390                                 struct dt_rec *dtrec, __u32 attr)
5391 {
5392         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5393         struct osd_object      *obj   = it->oie_obj;
5394         struct osd_device      *dev   = osd_obj2dev(obj);
5395         struct osd_scrub       *scrub = &dev->od_scrub;
5396         struct scrub_file      *sf    = &scrub->os_file;
5397         struct osd_thread_info *oti   = osd_oti_get(env);
5398         struct osd_inode_id    *id    = &oti->oti_id;
5399         struct osd_idmap_cache *oic   = &oti->oti_cache;
5400         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5401         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5402         __u32                   ino   = it->oie_dirent->oied_ino;
5403         int                     rc    = 0;
5404         ENTRY;
5405
5406         if (attr & LUDA_VERIFY) {
5407                 attr |= LUDA_TYPE;
5408                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5409                         attr |= LUDA_IGNORE;
5410                         rc = 0;
5411                 } else {
5412                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5413                                                      &attr);
5414                 }
5415         } else {
5416                 attr &= ~LU_DIRENT_ATTRS_MASK;
5417                 if (!fid_is_sane(fid)) {
5418                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5419                             likely(it->oie_dirent->oied_namelen != 2 ||
5420                                    it->oie_dirent->oied_name[0] != '.' ||
5421                                    it->oie_dirent->oied_name[1] != '.'))
5422                                 RETURN(-ENOENT);
5423
5424                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
5425                 } else {
5426                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5427                 }
5428         }
5429
5430         /* Pack the entry anyway, at least the offset is right. */
5431         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5432                            it->oie_dirent->oied_name,
5433                            it->oie_dirent->oied_namelen,
5434                            it->oie_dirent->oied_type, attr);
5435
5436         if (rc < 0)
5437                 RETURN(rc);
5438
5439         if (osd_remote_fid(env, dev, fid))
5440                 RETURN(0);
5441
5442         if (likely(!(attr & LUDA_IGNORE)))
5443                 rc = osd_add_oi_cache(oti, dev, id, fid);
5444
5445         if (!(attr & LUDA_VERIFY) &&
5446             (scrub->os_pos_current <= ino) &&
5447             ((sf->sf_flags & SF_INCONSISTENT) ||
5448              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5449              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5450                 osd_consistency_check(oti, dev, oic);
5451
5452         RETURN(rc);
5453 }
5454
5455 /**
5456  * Returns the record size size at current position.
5457  *
5458  * This function will return record(lu_dirent) size in bytes.
5459  *
5460  * \param[in] env       execution environment
5461  * \param[in] di        iterator's in memory structure
5462  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5463  *                      calculate the lu_dirent size.
5464  *
5465  * \retval      record size(in bytes & in memory) of the current lu_dirent
5466  *              entry.
5467  */
5468 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5469                               __u32 attr)
5470 {
5471         struct osd_it_ea *it = (struct osd_it_ea *)di;
5472
5473         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5474 }
5475
5476 /**
5477  * Returns a cookie for current position of the iterator head, so that
5478  * user can use this cookie to load/start the iterator next time.
5479  *
5480  * \param di iterator's in memory structure
5481  *
5482  * \retval cookie for current position, on success
5483  */
5484 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5485 {
5486         struct osd_it_ea *it = (struct osd_it_ea *)di;
5487
5488         return it->oie_dirent->oied_off;
5489 }
5490
5491 /**
5492  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5493  * to load a directory entry at a time and stored it i inn,
5494  * in iterator's in-memory data structure.
5495  *
5496  * \param di struct osd_it_ea, iterator's in memory structure
5497  *
5498  * \retval +ve on success
5499  * \retval -ve on error
5500  */
5501 static int osd_it_ea_load(const struct lu_env *env,
5502                           const struct dt_it *di, __u64 hash)
5503 {
5504         struct osd_it_ea *it = (struct osd_it_ea *)di;
5505         int rc;
5506
5507         ENTRY;
5508         it->oie_file.f_pos = hash;
5509
5510         rc =  osd_ldiskfs_it_fill(env, di);
5511         if (rc > 0)
5512                 rc = -ENODATA;
5513
5514         if (rc == 0)
5515                 rc = +1;
5516
5517         RETURN(rc);
5518 }
5519
5520 /**
5521  * Index lookup function for interoperability mode (b11826).
5522  *
5523  * \param key,  key i.e. file name to be searched
5524  *
5525  * \retval +ve, on success
5526  * \retval -ve, on error
5527  */
5528 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5529                                struct dt_rec *rec, const struct dt_key *key,
5530                                struct lustre_capa *capa)
5531 {
5532         struct osd_object *obj = osd_dt_obj(dt);
5533         int rc = 0;
5534
5535         ENTRY;
5536
5537         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5538         LINVRNT(osd_invariant(obj));
5539
5540         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5541                 return -EACCES;
5542
5543         rc = osd_ea_lookup_rec(env, obj, rec, key);
5544         if (rc == 0)
5545                 rc = +1;
5546         RETURN(rc);
5547 }
5548
5549 /**
5550  * Index and Iterator operations for interoperability
5551  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5552  */
5553 static const struct dt_index_operations osd_index_ea_ops = {
5554         .dio_lookup         = osd_index_ea_lookup,
5555         .dio_declare_insert = osd_index_declare_ea_insert,
5556         .dio_insert         = osd_index_ea_insert,
5557         .dio_declare_delete = osd_index_declare_ea_delete,
5558         .dio_delete         = osd_index_ea_delete,
5559         .dio_it     = {
5560                 .init     = osd_it_ea_init,
5561                 .fini     = osd_it_ea_fini,
5562                 .get      = osd_it_ea_get,
5563                 .put      = osd_it_ea_put,
5564                 .next     = osd_it_ea_next,
5565                 .key      = osd_it_ea_key,
5566                 .key_size = osd_it_ea_key_size,
5567                 .rec      = osd_it_ea_rec,
5568                 .rec_size = osd_it_ea_rec_size,
5569                 .store    = osd_it_ea_store,
5570                 .load     = osd_it_ea_load
5571         }
5572 };
5573
5574 static void *osd_key_init(const struct lu_context *ctx,
5575                           struct lu_context_key *key)
5576 {
5577         struct osd_thread_info *info;
5578
5579         OBD_ALLOC_PTR(info);
5580         if (info == NULL)
5581                 return ERR_PTR(-ENOMEM);
5582
5583         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5584         if (info->oti_it_ea_buf == NULL)
5585                 goto out_free_info;
5586
5587         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5588
5589         info->oti_hlock = ldiskfs_htree_lock_alloc();
5590         if (info->oti_hlock == NULL)
5591                 goto out_free_ea;
5592
5593         return info;
5594
5595  out_free_ea:
5596         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5597  out_free_info:
5598         OBD_FREE_PTR(info);
5599         return ERR_PTR(-ENOMEM);
5600 }
5601
5602 static void osd_key_fini(const struct lu_context *ctx,
5603                          struct lu_context_key *key, void* data)
5604 {
5605         struct osd_thread_info *info = data;
5606
5607         if (info->oti_hlock != NULL)
5608                 ldiskfs_htree_lock_free(info->oti_hlock);
5609         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5610         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5611         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5612         OBD_FREE_PTR(info);
5613 }
5614
5615 static void osd_key_exit(const struct lu_context *ctx,
5616                          struct lu_context_key *key, void *data)
5617 {
5618         struct osd_thread_info *info = data;
5619
5620         LASSERT(info->oti_r_locks == 0);
5621         LASSERT(info->oti_w_locks == 0);
5622         LASSERT(info->oti_txns    == 0);
5623 }
5624
5625 /* type constructor/destructor: osd_type_init, osd_type_fini */
5626 LU_TYPE_INIT_FINI(osd, &osd_key);
5627
5628 struct lu_context_key osd_key = {
5629         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5630         .lct_init = osd_key_init,
5631         .lct_fini = osd_key_fini,
5632         .lct_exit = osd_key_exit
5633 };
5634
5635
5636 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5637                            const char *name, struct lu_device *next)
5638 {
5639         struct osd_device *osd = osd_dev(d);
5640
5641         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5642             >= sizeof(osd->od_svname))
5643                 return -E2BIG;
5644         return osd_procfs_init(osd, name);
5645 }
5646
5647 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5648 {
5649         struct seq_server_site  *ss = osd_seq_site(osd);
5650         int                     rc;
5651         ENTRY;
5652
5653         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5654                 RETURN(0);
5655
5656         if (unlikely(ss == NULL))
5657                 RETURN(-ENODEV);
5658
5659         OBD_ALLOC_PTR(osd->od_cl_seq);
5660         if (osd->od_cl_seq == NULL)
5661                 RETURN(-ENOMEM);
5662
5663         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5664                              osd->od_svname, ss->ss_server_seq);
5665
5666         if (rc != 0) {
5667                 OBD_FREE_PTR(osd->od_cl_seq);
5668                 osd->od_cl_seq = NULL;
5669         }
5670
5671         RETURN(rc);
5672 }
5673
5674 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5675 {
5676         if (osd->od_cl_seq == NULL)
5677                 return;
5678
5679         seq_client_fini(osd->od_cl_seq);
5680         OBD_FREE_PTR(osd->od_cl_seq);
5681         osd->od_cl_seq = NULL;
5682 }
5683
5684 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5685 {
5686         ENTRY;
5687
5688         /* shutdown quota slave instance associated with the device */
5689         if (o->od_quota_slave != NULL) {
5690                 qsd_fini(env, o->od_quota_slave);
5691                 o->od_quota_slave = NULL;
5692         }
5693
5694         osd_fid_fini(env, o);
5695
5696         RETURN(0);
5697 }
5698
5699 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5700 {
5701         ENTRY;
5702
5703         if (o->od_mnt != NULL) {
5704                 shrink_dcache_sb(osd_sb(o));
5705                 osd_sync(env, &o->od_dt_dev);
5706
5707                 mntput(o->od_mnt);
5708                 o->od_mnt = NULL;
5709         }
5710
5711         EXIT;
5712 }
5713
5714 static int osd_mount(const struct lu_env *env,
5715                      struct osd_device *o, struct lustre_cfg *cfg)
5716 {
5717         const char              *name  = lustre_cfg_string(cfg, 0);
5718         const char              *dev  = lustre_cfg_string(cfg, 1);
5719         const char              *opts;
5720         unsigned long            page, s_flags, lmd_flags = 0;
5721         struct page             *__page;
5722         struct file_system_type *type;
5723         char                    *options = NULL;
5724         char                    *str;
5725         struct osd_thread_info  *info = osd_oti_get(env);
5726         struct lu_fid           *fid = &info->oti_fid;
5727         struct inode            *inode;
5728         int                      rc = 0;
5729         ENTRY;
5730
5731         if (o->od_mnt != NULL)
5732                 RETURN(0);
5733
5734         if (strlen(dev) >= sizeof(o->od_mntdev))
5735                 RETURN(-E2BIG);
5736         strcpy(o->od_mntdev, dev);
5737
5738         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5739         if (__page == NULL)
5740                 GOTO(out, rc = -ENOMEM);
5741
5742         str = lustre_cfg_string(cfg, 2);
5743         s_flags = simple_strtoul(str, NULL, 0);
5744         str = strstr(str, ":");
5745         if (str)
5746                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5747         opts = lustre_cfg_string(cfg, 3);
5748         page = (unsigned long)page_address(__page);
5749         options = (char *)page;
5750         *options = '\0';
5751         if (opts == NULL)
5752                 strcat(options, "user_xattr,acl");
5753         else
5754                 strcat(options, opts);
5755
5756         /* Glom up mount options */
5757         if (*options != '\0')
5758                 strcat(options, ",");
5759         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5760
5761         type = get_fs_type("ldiskfs");
5762         if (!type) {
5763                 CERROR("%s: cannot find ldiskfs module\n", name);
5764                 GOTO(out, rc = -ENODEV);
5765         }
5766
5767         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5768         module_put(type->owner);
5769
5770         if (IS_ERR(o->od_mnt)) {
5771                 rc = PTR_ERR(o->od_mnt);
5772                 o->od_mnt = NULL;
5773                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5774                 GOTO(out, rc);
5775         }
5776
5777 #ifdef HAVE_DEV_SET_RDONLY
5778         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5779                 CERROR("%s: underlying device %s is marked as read-only. "
5780                        "Setup failed\n", name, dev);
5781                 GOTO(out_mnt, rc = -EROFS);
5782         }
5783 #endif
5784
5785         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5786                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5787                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5788                 GOTO(out_mnt, rc = -EINVAL);
5789         }
5790
5791 #ifdef LDISKFS_MOUNT_DIRDATA
5792         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5793                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5794                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5795         else if (!o->od_is_ost)
5796                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5797                       "enabling the dirdata feature. If you do not want to "
5798                       "downgrade to Lustre-1.x again, you can enable it via "
5799                       "'tune2fs -O dirdata device'\n", name, dev);
5800 #endif
5801         inode = osd_sb(o)->s_root->d_inode;
5802         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5803         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5804         if (rc != 0) {
5805                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5806                 GOTO(out_mnt, rc);
5807         }
5808
5809         if (lmd_flags & LMD_FLG_NOSCRUB)
5810                 o->od_noscrub = 1;
5811
5812         GOTO(out, rc = 0);
5813
5814 out_mnt:
5815         mntput(o->od_mnt);
5816         o->od_mnt = NULL;
5817
5818 out:
5819         if (__page)
5820                 OBD_PAGE_FREE(__page);
5821
5822         return rc;
5823 }
5824
5825 static struct lu_device *osd_device_fini(const struct lu_env *env,
5826                                          struct lu_device *d)
5827 {
5828         struct osd_device *o = osd_dev(d);
5829         ENTRY;
5830
5831         osd_shutdown(env, o);
5832         osd_procfs_fini(o);
5833         osd_scrub_cleanup(env, o);
5834         osd_obj_map_fini(o);
5835         osd_umount(env, o);
5836
5837         RETURN(NULL);
5838 }
5839
5840 static int osd_device_init0(const struct lu_env *env,
5841                             struct osd_device *o,
5842                             struct lustre_cfg *cfg)
5843 {
5844         struct lu_device        *l = osd2lu_dev(o);
5845         struct osd_thread_info *info;
5846         int                     rc;
5847         int                     cplen = 0;
5848
5849         /* if the module was re-loaded, env can loose its keys */
5850         rc = lu_env_refill((struct lu_env *) env);
5851         if (rc)
5852                 GOTO(out, rc);
5853         info = osd_oti_get(env);
5854         LASSERT(info);
5855
5856         l->ld_ops = &osd_lu_ops;
5857         o->od_dt_dev.dd_ops = &osd_dt_ops;
5858
5859         spin_lock_init(&o->od_osfs_lock);
5860         mutex_init(&o->od_otable_mutex);
5861
5862         o->od_capa_hash = init_capa_hash();
5863         if (o->od_capa_hash == NULL)
5864                 GOTO(out, rc = -ENOMEM);
5865
5866         o->od_read_cache = 1;
5867         o->od_writethrough_cache = 1;
5868         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5869
5870         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5871                         sizeof(o->od_svname));
5872         if (cplen >= sizeof(o->od_svname)) {
5873                 rc = -E2BIG;
5874                 GOTO(out_capa, rc);
5875         }
5876
5877         if (server_name_is_ost(o->od_svname))
5878                 o->od_is_ost = 1;
5879
5880         rc = osd_mount(env, o, cfg);
5881         if (rc != 0)
5882                 GOTO(out_capa, rc);
5883
5884         rc = osd_obj_map_init(env, o);
5885         if (rc != 0)
5886                 GOTO(out_mnt, rc);
5887
5888         rc = lu_site_init(&o->od_site, l);
5889         if (rc != 0)
5890                 GOTO(out_compat, rc);
5891         o->od_site.ls_bottom_dev = l;
5892
5893         rc = lu_site_init_finish(&o->od_site);
5894         if (rc != 0)
5895                 GOTO(out_site, rc);
5896
5897         /* self-repair LMA by default */
5898         o->od_lma_self_repair = 1;
5899
5900         INIT_LIST_HEAD(&o->od_ios_list);
5901         /* setup scrub, including OI files initialization */
5902         rc = osd_scrub_setup(env, o);
5903         if (rc < 0)
5904                 GOTO(out_site, rc);
5905
5906         rc = osd_procfs_init(o, o->od_svname);
5907         if (rc != 0) {
5908                 CERROR("%s: can't initialize procfs: rc = %d\n",
5909                        o->od_svname, rc);
5910                 GOTO(out_scrub, rc);
5911         }
5912
5913         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5914
5915         /* initialize quota slave instance */
5916         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5917                                      o->od_proc_entry);
5918         if (IS_ERR(o->od_quota_slave)) {
5919                 rc = PTR_ERR(o->od_quota_slave);
5920                 o->od_quota_slave = NULL;
5921                 GOTO(out_procfs, rc);
5922         }
5923
5924         RETURN(0);
5925
5926 out_procfs:
5927         osd_procfs_fini(o);
5928 out_scrub:
5929         osd_scrub_cleanup(env, o);
5930 out_site:
5931         lu_site_fini(&o->od_site);
5932 out_compat:
5933         osd_obj_map_fini(o);
5934 out_mnt:
5935         osd_umount(env, o);
5936 out_capa:
5937         cleanup_capa_hash(o->od_capa_hash);
5938 out:
5939         return rc;
5940 }
5941
5942 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5943                                           struct lu_device_type *t,
5944                                           struct lustre_cfg *cfg)
5945 {
5946         struct osd_device *o;
5947         int                rc;
5948
5949         OBD_ALLOC_PTR(o);
5950         if (o == NULL)
5951                 return ERR_PTR(-ENOMEM);
5952
5953         rc = dt_device_init(&o->od_dt_dev, t);
5954         if (rc == 0) {
5955                 /* Because the ctx might be revived in dt_device_init,
5956                  * refill the env here */
5957                 lu_env_refill((struct lu_env *)env);
5958                 rc = osd_device_init0(env, o, cfg);
5959                 if (rc)
5960                         dt_device_fini(&o->od_dt_dev);
5961         }
5962
5963         if (unlikely(rc != 0))
5964                 OBD_FREE_PTR(o);
5965
5966         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5967 }
5968
5969 static struct lu_device *osd_device_free(const struct lu_env *env,
5970                                          struct lu_device *d)
5971 {
5972         struct osd_device *o = osd_dev(d);
5973         ENTRY;
5974
5975         cleanup_capa_hash(o->od_capa_hash);
5976         /* XXX: make osd top device in order to release reference */
5977         d->ld_site->ls_top_dev = d;
5978         lu_site_purge(env, d->ld_site, -1);
5979         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5980                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5981                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5982         }
5983         lu_site_fini(&o->od_site);
5984         dt_device_fini(&o->od_dt_dev);
5985         OBD_FREE_PTR(o);
5986         RETURN(NULL);
5987 }
5988
5989 static int osd_process_config(const struct lu_env *env,
5990                               struct lu_device *d, struct lustre_cfg *cfg)
5991 {
5992         struct osd_device               *o = osd_dev(d);
5993         int                             rc;
5994         ENTRY;
5995
5996         switch (cfg->lcfg_command) {
5997         case LCFG_SETUP:
5998                 rc = osd_mount(env, o, cfg);
5999                 break;
6000         case LCFG_CLEANUP:
6001                 lu_dev_del_linkage(d->ld_site, d);
6002                 rc = osd_shutdown(env, o);
6003                 break;
6004         case LCFG_PARAM:
6005                 LASSERT(&o->od_dt_dev);
6006                 rc = class_process_proc_seq_param(PARAM_OSD,
6007                                                   lprocfs_osd_obd_vars,
6008                                                   cfg, &o->od_dt_dev);
6009                 if (rc > 0 || rc == -ENOSYS)
6010                         rc = class_process_proc_seq_param(PARAM_OST,
6011                                                           lprocfs_osd_obd_vars,
6012                                                           cfg, &o->od_dt_dev);
6013                 break;
6014         default:
6015                 rc = -ENOSYS;
6016         }
6017
6018         RETURN(rc);
6019 }
6020
6021 static int osd_recovery_complete(const struct lu_env *env,
6022                                  struct lu_device *d)
6023 {
6024         struct osd_device       *osd = osd_dev(d);
6025         int                      rc = 0;
6026         ENTRY;
6027
6028         if (osd->od_quota_slave == NULL)
6029                 RETURN(0);
6030
6031         /* start qsd instance on recovery completion, this notifies the quota
6032          * slave code that we are about to process new requests now */
6033         rc = qsd_start(env, osd->od_quota_slave);
6034         RETURN(rc);
6035 }
6036
6037 /*
6038  * we use exports to track all osd users
6039  */
6040 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6041                            struct obd_device *obd, struct obd_uuid *cluuid,
6042                            struct obd_connect_data *data, void *localdata)
6043 {
6044         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6045         struct lustre_handle  conn;
6046         int                   rc;
6047         ENTRY;
6048
6049         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6050
6051         rc = class_connect(&conn, obd, cluuid);
6052         if (rc)
6053                 RETURN(rc);
6054
6055         *exp = class_conn2export(&conn);
6056
6057         spin_lock(&osd->od_osfs_lock);
6058         osd->od_connects++;
6059         spin_unlock(&osd->od_osfs_lock);
6060
6061         RETURN(0);
6062 }
6063
6064 /*
6065  * once last export (we don't count self-export) disappeared
6066  * osd can be released
6067  */
6068 static int osd_obd_disconnect(struct obd_export *exp)
6069 {
6070         struct obd_device *obd = exp->exp_obd;
6071         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6072         int                rc, release = 0;
6073         ENTRY;
6074
6075         /* Only disconnect the underlying layers on the final disconnect. */
6076         spin_lock(&osd->od_osfs_lock);
6077         osd->od_connects--;
6078         if (osd->od_connects == 0)
6079                 release = 1;
6080         spin_unlock(&osd->od_osfs_lock);
6081
6082         rc = class_disconnect(exp); /* bz 9811 */
6083
6084         if (rc == 0 && release)
6085                 class_manual_cleanup(obd);
6086         RETURN(rc);
6087 }
6088
6089 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6090                        struct lu_device *dev)
6091 {
6092         struct osd_device *osd = osd_dev(dev);
6093         int                result = 0;
6094         ENTRY;
6095
6096         if (osd->od_quota_slave != NULL) {
6097                 /* set up quota slave objects */
6098                 result = qsd_prepare(env, osd->od_quota_slave);
6099                 if (result != 0)
6100                         RETURN(result);
6101         }
6102
6103         result = osd_fid_init(env, osd);
6104
6105         RETURN(result);
6106 }
6107
6108 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6109                   struct lu_fid *fid, struct md_op_data *op_data)
6110 {
6111         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6112
6113         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6114 }
6115
6116 static const struct lu_object_operations osd_lu_obj_ops = {
6117         .loo_object_init      = osd_object_init,
6118         .loo_object_delete    = osd_object_delete,
6119         .loo_object_release   = osd_object_release,
6120         .loo_object_free      = osd_object_free,
6121         .loo_object_print     = osd_object_print,
6122         .loo_object_invariant = osd_object_invariant
6123 };
6124
6125 const struct lu_device_operations osd_lu_ops = {
6126         .ldo_object_alloc      = osd_object_alloc,
6127         .ldo_process_config    = osd_process_config,
6128         .ldo_recovery_complete = osd_recovery_complete,
6129         .ldo_prepare           = osd_prepare,
6130 };
6131
6132 static const struct lu_device_type_operations osd_device_type_ops = {
6133         .ldto_init = osd_type_init,
6134         .ldto_fini = osd_type_fini,
6135
6136         .ldto_start = osd_type_start,
6137         .ldto_stop  = osd_type_stop,
6138
6139         .ldto_device_alloc = osd_device_alloc,
6140         .ldto_device_free  = osd_device_free,
6141
6142         .ldto_device_init    = osd_device_init,
6143         .ldto_device_fini    = osd_device_fini
6144 };
6145
6146 struct lu_device_type osd_device_type = {
6147         .ldt_tags     = LU_DEVICE_DT,
6148         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6149         .ldt_ops      = &osd_device_type_ops,
6150         .ldt_ctx_tags = LCT_LOCAL,
6151 };
6152
6153 /*
6154  * lprocfs legacy support.
6155  */
6156 static struct obd_ops osd_obd_device_ops = {
6157         .o_owner = THIS_MODULE,
6158         .o_connect      = osd_obd_connect,
6159         .o_disconnect   = osd_obd_disconnect,
6160         .o_fid_alloc    = osd_fid_alloc,
6161 };
6162
6163 static int __init osd_mod_init(void)
6164 {
6165         int rc;
6166
6167         osd_oi_mod_init();
6168
6169         rc = lu_kmem_init(ldiskfs_caches);
6170         if (rc)
6171                 return rc;
6172
6173         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6174                                  lprocfs_osd_module_vars,
6175 #ifndef HAVE_ONLY_PROCFS_SEQ
6176                                  NULL,
6177 #endif
6178                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6179         if (rc)
6180                 lu_kmem_fini(ldiskfs_caches);
6181         return rc;
6182 }
6183
6184 static void __exit osd_mod_exit(void)
6185 {
6186         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6187         lu_kmem_fini(ldiskfs_caches);
6188 }
6189
6190 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6191 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6192 MODULE_LICENSE("GPL");
6193
6194 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);