lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76 #include <lustre_linkea.h>
  77
  78 int ldiskfs_pdo = 1;
  79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  80                 "ldiskfs with parallel directory operations");
  81
  82 int ldiskfs_track_declares_assert;
  83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  84                 "LBUG during tracking of declares");
  85
  86 /* Slab to allocate dynlocks */
  87 struct kmem_cache *dynlock_cachep;
  88
  89 static struct lu_kmem_descr ldiskfs_caches[] = {
  90         {
  91                 .ckd_cache = &dynlock_cachep,
  92                 .ckd_name  = "dynlock_cache",
  93                 .ckd_size  = sizeof(struct dynlock_handle)
  94         },
  95         {
  96                 .ckd_cache = NULL
  97         }
  98 };
  99
 100 static const char dot[] = ".";
 101 static const char dotdot[] = "..";
 102 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 103
 104 static const struct lu_object_operations      osd_lu_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ops;
 106 static const struct dt_object_operations      osd_obj_ea_ops;
 107 static const struct dt_object_operations      osd_obj_otable_it_ops;
 108 static const struct dt_index_operations       osd_index_iam_ops;
 109 static const struct dt_index_operations       osd_index_ea_ops;
 110
 111 int osd_trans_declare_op2rb[] = {
 112         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 113         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 114         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 115         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 116         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 117         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 118         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 119         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 120         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 121         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 122         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 123         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 124 };
 125
 126 static int osd_has_index(const struct osd_object *obj)
 127 {
 128         return obj->oo_dt.do_index_ops != NULL;
 129 }
 130
 131 static int osd_object_invariant(const struct lu_object *l)
 132 {
 133         return osd_invariant(osd_obj(l));
 134 }
 135
 136 /*
 137  * Concurrency: doesn't matter
 138  */
 139
 140 /*
 141  * Concurrency: doesn't matter
 142  */
 143 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 144 {
 145         struct osd_thread_info *oti = osd_oti_get(env);
 146         return oti->oti_w_locks > 0 && o->oo_owner == env;
 147 }
 148
 149 /*
 150  * Concurrency: doesn't access mutable data
 151  */
 152 static int osd_root_get(const struct lu_env *env,
 153                         struct dt_device *dev, struct lu_fid *f)
 154 {
 155         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 156         return 0;
 157 }
 158
 159 /*
 160  * OSD object methods.
 161  */
 162
 163 /*
 164  * Concurrency: no concurrent access is possible that early in object
 165  * life-cycle.
 166  */
 167 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 168                                           const struct lu_object_header *hdr,
 169                                           struct lu_device *d)
 170 {
 171         struct osd_object *mo;
 172
 173         OBD_ALLOC_PTR(mo);
 174         if (mo != NULL) {
 175                 struct lu_object *l;
 176
 177                 l = &mo->oo_dt.do_lu;
 178                 dt_object_init(&mo->oo_dt, NULL, d);
 179                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 180                 l->lo_ops = &osd_lu_obj_ops;
 181                 init_rwsem(&mo->oo_sem);
 182                 init_rwsem(&mo->oo_ext_idx_sem);
 183                 spin_lock_init(&mo->oo_guard);
 184                 return l;
 185         } else {
 186                 return NULL;
 187         }
 188 }
 189
 190 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 191                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 192 {
 193         int rc;
 194
 195         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 196         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 197                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 198         if (rc > 0) {
 199                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 200                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 201                 rc = 0;
 202                 lustre_lma_swab(lma);
 203                 /* Check LMA compatibility */
 204                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 205                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 206                               "for fid = "DFID", ino = %lu\n",
 207                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 208                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 209                               PFID(&lma->lma_self_fid), inode->i_ino);
 210                         rc = -EOPNOTSUPP;
 211                 }
 212         } else if (rc == 0) {
 213                 rc = -ENODATA;
 214         }
 215
 216         return rc;
 217 }
 218
 219 /*
 220  * retrieve object from backend ext fs.
 221  **/
 222 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 223                        struct osd_inode_id *id)
 224 {
 225         struct inode *inode = NULL;
 226
 227         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 228         if (IS_ERR(inode)) {
 229                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 230                        id->oii_ino, PTR_ERR(inode));
 231         } else if (id->oii_gen != OSD_OII_NOGEN &&
 232                    inode->i_generation != id->oii_gen) {
 233                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 234                        "i_generation = %u\n",
 235                        id->oii_ino, id->oii_gen, inode->i_generation);
 236                 iput(inode);
 237                 inode = ERR_PTR(-ESTALE);
 238         } else if (inode->i_nlink == 0) {
 239                 /* due to parallel readdir and unlink,
 240                 * we can have dead inode here. */
 241                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 242                 iput(inode);
 243                 inode = ERR_PTR(-ESTALE);
 244         } else if (is_bad_inode(inode)) {
 245                 CWARN("%.16s: bad inode: ino = %u\n",
 246                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 247                 iput(inode);
 248                 inode = ERR_PTR(-ENOENT);
 249         } else {
 250                 if (id->oii_gen == OSD_OII_NOGEN)
 251                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 252
 253                 /* Do not update file c/mtime in ldiskfs.
 254                  * NB: we don't have any lock to protect this because we don't
 255                  * have reference on osd_object now, but contention with
 256                  * another lookup + attr_set can't happen in the tiny window
 257                  * between if (...) and set S_NOCMTIME. */
 258                 if (!(inode->i_flags & S_NOCMTIME))
 259                         inode->i_flags |= S_NOCMTIME;
 260         }
 261         return inode;
 262 }
 263
 264 static struct inode *
 265 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 266              struct osd_inode_id *id, struct lu_fid *fid)
 267 {
 268         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 269         struct inode            *inode;
 270         int                      rc;
 271
 272         inode = osd_iget(info, dev, id);
 273         if (IS_ERR(inode))
 274                 return inode;
 275
 276         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 277         if (rc == 0) {
 278                 *fid = lma->lma_self_fid;
 279         } else if (rc == -ENODATA) {
 280                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 281                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 282                 else
 283                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 284         } else {
 285                 iput(inode);
 286                 inode = ERR_PTR(rc);
 287         }
 288         return inode;
 289 }
 290
 291 static struct inode *osd_iget_check(struct osd_thread_info *info,
 292                                     struct osd_device *dev,
 293                                     const struct lu_fid *fid,
 294                                     struct osd_inode_id *id,
 295                                     bool in_oi)
 296 {
 297         struct inode    *inode;
 298         int              rc     = 0;
 299         ENTRY;
 300
 301         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 302         if (IS_ERR(inode)) {
 303                 rc = PTR_ERR(inode);
 304                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 305                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 306                                id->oii_ino, rc);
 307
 308                         GOTO(put, rc);
 309                 }
 310
 311                 goto check_oi;
 312         }
 313
 314         if (is_bad_inode(inode)) {
 315                 rc = -ENOENT;
 316                 if (!in_oi) {
 317                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 318
 319                         GOTO(put, rc);
 320                 }
 321
 322                 goto check_oi;
 323         }
 324
 325         if (id->oii_gen != OSD_OII_NOGEN &&
 326             inode->i_generation != id->oii_gen) {
 327                 rc = -ESTALE;
 328                 if (!in_oi) {
 329                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 330                                "oii_gen = %u, i_generation = %u\n",
 331                                id->oii_ino, id->oii_gen, inode->i_generation);
 332
 333                         GOTO(put, rc);
 334                 }
 335
 336                 goto check_oi;
 337         }
 338
 339         if (inode->i_nlink == 0) {
 340                 rc = -ENOENT;
 341                 if (!in_oi) {
 342                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 343
 344                         GOTO(put, rc);
 345                 }
 346
 347                 goto check_oi;
 348         }
 349
 350 check_oi:
 351         if (rc != 0) {
 352                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 353
 354                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 355                 /* XXX: There are some possible cases:
 356                  *      1. rc = 0.
 357                  *         Backup/restore caused the OI invalid.
 358                  *      2. rc = 0.
 359                  *         Someone unlinked the object but NOT removed
 360                  *         the OI mapping, such as mount target device
 361                  *         as ldiskfs, and modify something directly.
 362                  *      3. rc = -ENOENT.
 363                  *         Someone just removed the object between the
 364                  *         former oi_lookup and the iget. It is normal.
 365                  *      4. Other failure cases.
 366                  *
 367                  *      Generally, when the device is mounted, it will
 368                  *      auto check whether the system is restored from
 369                  *      file-level backup or not. We trust such detect
 370                  *      to distinguish the 1st case from the 2nd case. */
 371                 if (rc == 0) {
 372                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 373                             inode->i_generation == id->oii_gen)
 374                                 rc = -ENOENT;
 375                         else
 376                                 rc = -EREMCHG;
 377                 }
 378         } else {
 379                 if (id->oii_gen == OSD_OII_NOGEN)
 380                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 381
 382                 /* Do not update file c/mtime in ldiskfs.
 383                  * NB: we don't have any lock to protect this because we don't
 384                  * have reference on osd_object now, but contention with
 385                  * another lookup + attr_set can't happen in the tiny window
 386                  * between if (...) and set S_NOCMTIME. */
 387                 if (!(inode->i_flags & S_NOCMTIME))
 388                         inode->i_flags |= S_NOCMTIME;
 389         }
 390
 391         GOTO(put, rc);
 392
 393 put:
 394         if (rc != 0) {
 395                 if (!IS_ERR(inode))
 396                         iput(inode);
 397
 398                 inode = ERR_PTR(rc);
 399         }
 400
 401         return inode;
 402 }
 403
 404 /**
 405  * \retval +v: new filter_fid, does not contain self-fid
 406  * \retval 0:  filter_fid_old, contains self-fid
 407  * \retval -v: other failure cases
 408  */
 409 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 410                  struct dentry *dentry, struct lu_fid *fid)
 411 {
 412         struct filter_fid_old   *ff     = &info->oti_ff;
 413         struct ost_id           *ostid  = &info->oti_ostid;
 414         int                      rc;
 415
 416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 417         if (rc == sizeof(*ff)) {
 418                 rc = 0;
 419                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 420                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 421                 /* XXX: should use real OST index in the future. LU-3569 */
 422                 ostid_to_fid(fid, ostid, 0);
 423         } else if (rc == sizeof(struct filter_fid)) {
 424                 rc = 1;
 425         } else if (rc >= 0) {
 426                 rc = -EINVAL;
 427         }
 428
 429         return rc;
 430 }
 431
 432 static int osd_lma_self_repair(struct osd_thread_info *info,
 433                                struct osd_device *osd, struct inode *inode,
 434                                const struct lu_fid *fid, __u32 compat)
 435 {
 436         handle_t *jh;
 437         int       rc;
 438
 439         LASSERT(current->journal_info == NULL);
 440
 441         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 442                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 443         if (IS_ERR(jh)) {
 444                 rc = PTR_ERR(jh);
 445                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 446                       osd_name(osd), rc);
 447                 return rc;
 448         }
 449
 450         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 451         if (rc != 0)
 452                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 453                       osd_name(osd), rc);
 454         ldiskfs_journal_stop(jh);
 455         return rc;
 456 }
 457
 458 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 459 {
 460         struct osd_thread_info  *info   = osd_oti_get(env);
 461         struct osd_device       *osd    = osd_obj2dev(obj);
 462         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 463         struct inode            *inode  = obj->oo_inode;
 464         struct dentry           *dentry = &info->oti_obj_dentry;
 465         struct lu_fid           *fid    = NULL;
 466         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 467         int                      rc;
 468         ENTRY;
 469
 470         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 471         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 472                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 473         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 474                 fid = &lma->lma_self_fid;
 475                 rc = osd_get_idif(info, inode, dentry, fid);
 476                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 477                         /* For the given OST-object, if it has neither LMA nor
 478                          * FID in XATTR_NAME_FID, then the given FID (which is
 479                          * contained in the @obj, from client RPC for locating
 480                          * the OST-object) is trusted. We use it to generate
 481                          * the LMA. */
 482                         osd_lma_self_repair(info, osd, inode, rfid,
 483                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 484                                 LMAC_FID_ON_OST : 0);
 485                         RETURN(0);
 486                 }
 487         }
 488
 489         if (unlikely(rc == -ENODATA))
 490                 RETURN(0);
 491
 492         if (rc < 0)
 493                 RETURN(rc);
 494
 495         if (rc > 0) {
 496                 rc = 0;
 497                 lustre_lma_swab(lma);
 498                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 499                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 500                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 501                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 502                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 503                               PFID(rfid), inode->i_ino);
 504                         rc = -EOPNOTSUPP;
 505                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 506                         fid = &lma->lma_self_fid;
 507                 }
 508         }
 509
 510         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 511                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 512                         struct ost_id   *oi   = &info->oti_ostid;
 513                         struct lu_fid   *fid1 = &info->oti_fid3;
 514                         __u32            idx  = fid_idif_ost_idx(rfid);
 515
 516                         /* For old IDIF, the OST index is not part of the IDIF,
 517                          * Means that different OSTs may have the same IDIFs.
 518                          * Under such case, we need to make some compatible
 519                          * check to make sure to trigger OI scrub properly. */
 520                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 521                                 /* Given @rfid is new, LMA is old. */
 522                                 fid_to_ostid(fid, oi);
 523                                 ostid_to_fid(fid1, oi, idx);
 524                                 if (lu_fid_eq(fid1, rfid)) {
 525                                         if (osd->od_lma_self_repair)
 526                                                 osd_lma_self_repair(info, osd,
 527                                                         inode, rfid,
 528                                                         LMAC_FID_ON_OST);
 529                                         RETURN(0);
 530                                 }
 531                         }
 532                 }
 533
 534                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 535                        osd_name(osd), PFID(rfid), PFID(fid));
 536                 rc = -EREMCHG;
 537         }
 538
 539         RETURN(rc);
 540 }
 541
 542 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 543                           const struct lu_fid *fid,
 544                           const struct lu_object_conf *conf)
 545 {
 546         struct osd_thread_info *info;
 547         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 548         struct osd_device      *dev;
 549         struct osd_idmap_cache *oic;
 550         struct osd_inode_id    *id;
 551         struct inode           *inode;
 552         struct osd_scrub       *scrub;
 553         struct scrub_file      *sf;
 554         int                     result;
 555         int                     saved  = 0;
 556         bool                    in_oi  = false;
 557         bool                    triggered = false;
 558         ENTRY;
 559
 560         LINVRNT(osd_invariant(obj));
 561         LASSERT(obj->oo_inode == NULL);
 562         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 563
 564         dev = osd_dev(ldev);
 565         scrub = &dev->od_scrub;
 566         sf = &scrub->os_file;
 567         info = osd_oti_get(env);
 568         LASSERT(info);
 569         oic = &info->oti_cache;
 570
 571         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 572                 RETURN(-ENOENT);
 573
 574         /* For the object is created as locking anchor, or for the object to
 575          * be created on disk. No need to osd_oi_lookup() at here because FID
 576          * shouldn't never be re-used, if it's really a duplicate FID from
 577          * unexpected reason, we should be able to detect it later by calling
 578          * do_create->osd_oi_insert(). */
 579         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 580                 GOTO(out, result = 0);
 581
 582         /* Search order: 1. per-thread cache. */
 583         if (lu_fid_eq(fid, &oic->oic_fid) &&
 584             likely(oic->oic_dev == dev)) {
 585                 id = &oic->oic_lid;
 586                 goto iget;
 587         }
 588
 589         id = &info->oti_id;
 590         if (!list_empty(&scrub->os_inconsistent_items)) {
 591                 /* Search order: 2. OI scrub pending list. */
 592                 result = osd_oii_lookup(dev, fid, id);
 593                 if (result == 0)
 594                         goto iget;
 595         }
 596
 597         /* Search order: 3. OI files. */
 598         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 599         if (result == -ENOENT) {
 600                 if (!fid_is_norm(fid) ||
 601                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 602                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 603                                       sf->sf_oi_bitmap))
 604                         GOTO(out, result = 0);
 605
 606                 goto trigger;
 607         }
 608
 609         if (result != 0)
 610                 GOTO(out, result);
 611
 612         in_oi = true;
 613
 614 iget:
 615         inode = osd_iget_check(info, dev, fid, id, in_oi);
 616         if (IS_ERR(inode)) {
 617                 result = PTR_ERR(inode);
 618                 if (result == -ENOENT || result == -ESTALE) {
 619                         if (!in_oi)
 620                                 fid_zero(&oic->oic_fid);
 621
 622                         GOTO(out, result = -ENOENT);
 623                 } else if (result == -EREMCHG) {
 624
 625 trigger:
 626                         if (!in_oi)
 627                                 fid_zero(&oic->oic_fid);
 628
 629                         if (unlikely(triggered))
 630                                 GOTO(out, result = saved);
 631
 632                         triggered = true;
 633                         if (thread_is_running(&scrub->os_thread)) {
 634                                 result = -EINPROGRESS;
 635                         } else if (!dev->od_noscrub) {
 636                                 result = osd_scrub_start(dev);
 637                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 638                                               "for "DFID", rc = %d [1]\n",
 639                                               osd_name(dev), PFID(fid),result);
 640                                 if (result == 0 || result == -EALREADY)
 641                                         result = -EINPROGRESS;
 642                                 else
 643                                         result = -EREMCHG;
 644                         }
 645
 646                         /* We still have chance to get the valid inode: for the
 647                          * object which is referenced by remote name entry, the
 648                          * object on the local MDT will be linked under the dir
 649                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 650                          *
 651                          * We do not know whether the object for the given FID
 652                          * is referenced by some remote name entry or not, and
 653                          * especially for DNE II, a multiple-linked object may
 654                          * have many name entries reside on many MDTs.
 655                          *
 656                          * To simplify the operation, OSD will not distinguish
 657                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 658                          * only happened for the RPC from other MDT during the
 659                          * OI scrub, or for the client side RPC with FID only,
 660                          * such as FID to path, or from old connected client. */
 661                         saved = result;
 662                         result = osd_lookup_in_remote_parent(info, dev,
 663                                                              fid, id);
 664                         if (result == 0) {
 665                                 in_oi = false;
 666                                 goto iget;
 667                         }
 668
 669                         result = saved;
 670                 }
 671
 672                 GOTO(out, result);
 673         }
 674
 675         obj->oo_inode = inode;
 676         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 677
 678         result = osd_check_lma(env, obj);
 679         if (result != 0) {
 680                 iput(inode);
 681                 obj->oo_inode = NULL;
 682                 if (result == -EREMCHG) {
 683                         if (!in_oi) {
 684                                 result = osd_oi_lookup(info, dev, fid, id,
 685                                                        OI_CHECK_FLD);
 686                                 if (result != 0) {
 687                                         fid_zero(&oic->oic_fid);
 688                                         GOTO(out, result);
 689                                 }
 690                         }
 691
 692                         goto trigger;
 693                 }
 694
 695                 GOTO(out, result);
 696         }
 697
 698         obj->oo_compat_dot_created = 1;
 699         obj->oo_compat_dotdot_created = 1;
 700
 701         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 702                 GOTO(out, result = 0);
 703
 704         LASSERT(obj->oo_hl_head == NULL);
 705         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 706         if (obj->oo_hl_head == NULL) {
 707                 obj->oo_inode = NULL;
 708                 iput(inode);
 709                 GOTO(out, result = -ENOMEM);
 710         }
 711         GOTO(out, result = 0);
 712
 713 out:
 714         LINVRNT(osd_invariant(obj));
 715         return result;
 716 }
 717
 718 /*
 719  * Concurrency: shouldn't matter.
 720  */
 721 static void osd_object_init0(struct osd_object *obj)
 722 {
 723         LASSERT(obj->oo_inode != NULL);
 724         obj->oo_dt.do_body_ops = &osd_body_ops;
 725         obj->oo_dt.do_lu.lo_header->loh_attr |=
 726                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 727 }
 728
 729 /*
 730  * Concurrency: no concurrent access is possible that early in object
 731  * life-cycle.
 732  */
 733 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 734                            const struct lu_object_conf *conf)
 735 {
 736         struct osd_object *obj = osd_obj(l);
 737         int result;
 738
 739         LINVRNT(osd_invariant(obj));
 740
 741         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 742                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 743                 l->lo_header->loh_attr |= LOHA_EXISTS;
 744                 return 0;
 745         }
 746
 747         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 748         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 749         if (result == 0 && obj->oo_inode != NULL)
 750                 osd_object_init0(obj);
 751
 752         LINVRNT(osd_invariant(obj));
 753         return result;
 754 }
 755
 756 /*
 757  * Concurrency: no concurrent access is possible that late in object
 758  * life-cycle.
 759  */
 760 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 761 {
 762         struct osd_object *obj = osd_obj(l);
 763
 764         LINVRNT(osd_invariant(obj));
 765
 766         dt_object_fini(&obj->oo_dt);
 767         if (obj->oo_hl_head != NULL)
 768                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 769         OBD_FREE_PTR(obj);
 770 }
 771
 772 /*
 773  * Concurrency: no concurrent access is possible that late in object
 774  * life-cycle.
 775  */
 776 static void osd_index_fini(struct osd_object *o)
 777 {
 778         struct iam_container *bag;
 779
 780         if (o->oo_dir != NULL) {
 781                 bag = &o->oo_dir->od_container;
 782                 if (o->oo_inode != NULL) {
 783                         if (bag->ic_object == o->oo_inode)
 784                                 iam_container_fini(bag);
 785                 }
 786                 OBD_FREE_PTR(o->oo_dir);
 787                 o->oo_dir = NULL;
 788         }
 789 }
 790
 791 /*
 792  * Concurrency: no concurrent access is possible that late in object
 793  * life-cycle (for all existing callers, that is. New callers have to provide
 794  * their own locking.)
 795  */
 796 static int osd_inode_unlinked(const struct inode *inode)
 797 {
 798         return inode->i_nlink == 0;
 799 }
 800
 801 enum {
 802         OSD_TXN_OI_DELETE_CREDITS    = 20,
 803         OSD_TXN_INODE_DELETE_CREDITS = 20
 804 };
 805
 806 /*
 807  * Journal
 808  */
 809
 810 #if OSD_THANDLE_STATS
 811 /**
 812  * Set time when the handle is allocated
 813  */
 814 static void osd_th_alloced(struct osd_thandle *oth)
 815 {
 816         oth->oth_alloced = cfs_time_current();
 817 }
 818
 819 /**
 820  * Set time when the handle started
 821  */
 822 static void osd_th_started(struct osd_thandle *oth)
 823 {
 824         oth->oth_started = cfs_time_current();
 825 }
 826
 827 /**
 828  * Helper function to convert time interval to microseconds packed in
 829  * long int.
 830  */
 831 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 832 {
 833         struct timeval val;
 834
 835         cfs_duration_usec(cfs_time_sub(end, start), &val);
 836         return val.tv_sec * 1000000 + val.tv_usec;
 837 }
 838
 839 /**
 840  * Check whether the we deal with this handle for too long.
 841  */
 842 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 843                                 cfs_time_t alloced, cfs_time_t started,
 844                                 cfs_time_t closed)
 845 {
 846         cfs_time_t now = cfs_time_current();
 847
 848         LASSERT(dev != NULL);
 849
 850         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 851                             interval_to_usec(alloced, started));
 852         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 853                             interval_to_usec(started, closed));
 854         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 855                             interval_to_usec(closed, now));
 856
 857         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 858                 CWARN("transaction handle %p was open for too long: "
 859                       "now "CFS_TIME_T" ,"
 860                       "alloced "CFS_TIME_T" ,"
 861                       "started "CFS_TIME_T" ,"
 862                       "closed "CFS_TIME_T"\n",
 863                       oth, now, alloced, started, closed);
 864                 libcfs_debug_dumpstack(NULL);
 865         }
 866 }
 867
 868 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 869 {                                                                       \
 870         cfs_time_t __closed = cfs_time_current();                       \
 871         cfs_time_t __alloced = oth->oth_alloced;                        \
 872         cfs_time_t __started = oth->oth_started;                        \
 873                                                                         \
 874         expr;                                                           \
 875         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 876 }
 877
 878 #else /* OSD_THANDLE_STATS */
 879
 880 #define osd_th_alloced(h)                  do {} while(0)
 881 #define osd_th_started(h)                  do {} while(0)
 882 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 883
 884 #endif /* OSD_THANDLE_STATS */
 885
 886 /*
 887  * Concurrency: doesn't access mutable data.
 888  */
 889 static int osd_param_is_not_sane(const struct osd_device *dev,
 890                                  const struct thandle *th)
 891 {
 892         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 893
 894         return oh->ot_credits > osd_transaction_size(dev);
 895 }
 896
 897 /*
 898  * Concurrency: shouldn't matter.
 899  */
 900 static void osd_trans_commit_cb(struct super_block *sb,
 901                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 902 {
 903         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 904         struct thandle     *th  = &oh->ot_super;
 905         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 906         struct dt_txn_commit_cb *dcb, *tmp;
 907
 908         LASSERT(oh->ot_handle == NULL);
 909
 910         if (error)
 911                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 912
 913         dt_txn_hook_commit(th);
 914
 915         /* call per-transaction callbacks if any */
 916         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 917                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 918                          "commit callback entry: magic=%x name='%s'\n",
 919                          dcb->dcb_magic, dcb->dcb_name);
 920                 list_del_init(&dcb->dcb_linkage);
 921                 dcb->dcb_func(NULL, th, dcb, error);
 922         }
 923
 924         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 925         lu_device_put(lud);
 926         th->th_dev = NULL;
 927
 928         lu_context_exit(&th->th_ctx);
 929         lu_context_fini(&th->th_ctx);
 930         thandle_put(th);
 931 }
 932
 933 static struct thandle *osd_trans_create(const struct lu_env *env,
 934                                         struct dt_device *d)
 935 {
 936         struct osd_thread_info  *oti = osd_oti_get(env);
 937         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 938         struct osd_thandle      *oh;
 939         struct thandle          *th;
 940         ENTRY;
 941
 942         /* on pending IO in this thread should left from prev. request */
 943         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 944
 945         th = ERR_PTR(-ENOMEM);
 946         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 947         if (oh != NULL) {
 948                 oh->ot_quota_trans = &oti->oti_quota_trans;
 949                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 950                 th = &oh->ot_super;
 951                 th->th_dev = d;
 952                 th->th_result = 0;
 953                 th->th_tags = LCT_TX_HANDLE;
 954                 oh->ot_credits = 0;
 955                 atomic_set(&th->th_refc, 1);
 956                 th->th_alloc_size = sizeof(*oh);
 957                 oti->oti_dev = osd_dt_dev(d);
 958                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 959                 osd_th_alloced(oh);
 960
 961                 memset(oti->oti_declare_ops, 0,
 962                                         sizeof(oti->oti_declare_ops));
 963                 memset(oti->oti_declare_ops_rb, 0,
 964                                         sizeof(oti->oti_declare_ops_rb));
 965                 memset(oti->oti_declare_ops_cred, 0,
 966                                         sizeof(oti->oti_declare_ops_cred));
 967                 oti->oti_rollback = false;
 968         }
 969         RETURN(th);
 970 }
 971
 972 /*
 973  * Concurrency: shouldn't matter.
 974  */
 975 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 976                     struct thandle *th)
 977 {
 978         struct osd_thread_info *oti = osd_oti_get(env);
 979         struct osd_device  *dev = osd_dt_dev(d);
 980         handle_t           *jh;
 981         struct osd_thandle *oh;
 982         int rc;
 983
 984         ENTRY;
 985
 986         LASSERT(current->journal_info == NULL);
 987
 988         oh = container_of0(th, struct osd_thandle, ot_super);
 989         LASSERT(oh != NULL);
 990         LASSERT(oh->ot_handle == NULL);
 991
 992         rc = dt_txn_hook_start(env, d, th);
 993         if (rc != 0)
 994                 GOTO(out, rc);
 995
 996         if (unlikely(osd_param_is_not_sane(dev, th))) {
 997                 static unsigned long last_printed;
 998                 static int last_credits;
 999
1000                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1001                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1002                       oh->ot_credits,
1003                       osd_journal(dev)->j_max_transaction_buffers);
1004                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1005                       oti->oti_declare_ops[OSD_OT_CREATE],
1006                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1007                       oti->oti_declare_ops[OSD_OT_DESTROY],
1008                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1009                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1010                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1012                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1013                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1014                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1015                       oti->oti_declare_ops[OSD_OT_WRITE],
1016                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1017                       oti->oti_declare_ops[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1019                       oti->oti_declare_ops[OSD_OT_QUOTA],
1020                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1021                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1022                       oti->oti_declare_ops[OSD_OT_INSERT],
1023                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1024                       oti->oti_declare_ops[OSD_OT_DELETE],
1025                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1026                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1027                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1029                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1030                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1031
1032                 if (last_credits != oh->ot_credits &&
1033                     time_after(jiffies, last_printed +
1034                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1035                         libcfs_debug_dumpstack(NULL);
1036                         last_credits = oh->ot_credits;
1037                         last_printed = jiffies;
1038                 }
1039                 /* XXX Limit the credits to 'max_transaction_buffers', and
1040                  *     let the underlying filesystem to catch the error if
1041                  *     we really need so many credits.
1042                  *
1043                  *     This should be removed when we can calculate the
1044                  *     credits precisely. */
1045                 oh->ot_credits = osd_transaction_size(dev);
1046         }
1047
1048         /*
1049          * XXX temporary stuff. Some abstraction layer should
1050          * be used.
1051          */
1052         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1053         osd_th_started(oh);
1054         if (!IS_ERR(jh)) {
1055                 oh->ot_handle = jh;
1056                 LASSERT(oti->oti_txns == 0);
1057                 lu_context_init(&th->th_ctx, th->th_tags);
1058                 lu_context_enter(&th->th_ctx);
1059
1060                 lu_device_get(&d->dd_lu_dev);
1061                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1062                               "osd-tx", th);
1063                 oti->oti_txns++;
1064                 rc = 0;
1065         } else {
1066                 rc = PTR_ERR(jh);
1067         }
1068 out:
1069         RETURN(rc);
1070 }
1071
1072 static int osd_seq_exists(const struct lu_env *env,
1073                               struct osd_device *osd, obd_seq seq)
1074 {
1075         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1076         struct seq_server_site  *ss = osd_seq_site(osd);
1077         int                     rc;
1078         ENTRY;
1079
1080         LASSERT(ss != NULL);
1081         LASSERT(ss->ss_server_fld != NULL);
1082
1083         rc = osd_fld_lookup(env, osd, seq, range);
1084         if (rc != 0) {
1085                 if (rc != -ENOENT)
1086                         CERROR("%s: can't lookup FLD sequence "LPX64
1087                                ": rc = %d\n", osd_name(osd), seq, rc);
1088                 RETURN(0);
1089         }
1090
1091         RETURN(ss->ss_node_id == range->lsr_index);
1092 }
1093
1094 /*
1095  * Concurrency: shouldn't matter.
1096  */
1097 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1098                           struct thandle *th)
1099 {
1100         int                     rc = 0;
1101         struct osd_thandle     *oh;
1102         struct osd_thread_info *oti = osd_oti_get(env);
1103         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1104         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1105         struct lquota_trans    *qtrans;
1106         ENTRY;
1107
1108         oh = container_of0(th, struct osd_thandle, ot_super);
1109
1110         qtrans = oh->ot_quota_trans;
1111         oh->ot_quota_trans = NULL;
1112
1113         if (oh->ot_handle != NULL) {
1114                 handle_t *hdl = oh->ot_handle;
1115
1116                 /*
1117                  * add commit callback
1118                  * notice we don't do this in osd_trans_start()
1119                  * as underlying transaction can change during truncate
1120                  */
1121                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1122                                          &oh->ot_jcb);
1123
1124                 LASSERT(oti->oti_txns == 1);
1125                 oti->oti_txns--;
1126                 rc = dt_txn_hook_stop(env, th);
1127                 if (rc != 0)
1128                         CERROR("Failure in transaction hook: %d\n", rc);
1129
1130                 /* hook functions might modify th_sync */
1131                 hdl->h_sync = th->th_sync;
1132
1133                 oh->ot_handle = NULL;
1134                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1135                                   rc = ldiskfs_journal_stop(hdl));
1136                 if (rc != 0)
1137                         CERROR("Failure to stop transaction: %d\n", rc);
1138         } else {
1139                 thandle_put(&oh->ot_super);
1140         }
1141
1142         /* inform the quota slave device that the transaction is stopping */
1143         qsd_op_end(env, qsd, qtrans);
1144
1145         /* as we want IO to journal and data IO be concurrent, we don't block
1146          * awaiting data IO completion in osd_do_bio(), instead we wait here
1147          * once transaction is submitted to the journal. all reqular requests
1148          * don't do direct IO (except read/write), thus this wait_event becomes
1149          * no-op for them.
1150          *
1151          * IMPORTANT: we have to wait till any IO submited by the thread is
1152          * completed otherwise iobuf may be corrupted by different request
1153          */
1154         wait_event(iobuf->dr_wait,
1155                        atomic_read(&iobuf->dr_numreqs) == 0);
1156         osd_fini_iobuf(oti->oti_dev, iobuf);
1157         if (!rc)
1158                 rc = iobuf->dr_error;
1159
1160         RETURN(rc);
1161 }
1162
1163 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1164 {
1165         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1166                                                ot_super);
1167
1168         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1169         LASSERT(&dcb->dcb_func != NULL);
1170         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1171
1172         return 0;
1173 }
1174
1175 /*
1176  * Called just before object is freed. Releases all resources except for
1177  * object itself (that is released by osd_object_free()).
1178  *
1179  * Concurrency: no concurrent access is possible that late in object
1180  * life-cycle.
1181  */
1182 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1183 {
1184         struct osd_object *obj   = osd_obj(l);
1185         struct inode      *inode = obj->oo_inode;
1186
1187         LINVRNT(osd_invariant(obj));
1188
1189         /*
1190          * If object is unlinked remove fid->ino mapping from object index.
1191          */
1192
1193         osd_index_fini(obj);
1194         if (inode != NULL) {
1195                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1196                 qid_t                    uid = i_uid_read(inode);
1197                 qid_t                    gid = i_gid_read(inode);
1198
1199                 iput(inode);
1200                 obj->oo_inode = NULL;
1201
1202                 if (qsd != NULL) {
1203                         struct osd_thread_info  *info = osd_oti_get(env);
1204                         struct lquota_id_info   *qi = &info->oti_qi;
1205
1206                         /* Release granted quota to master if necessary */
1207                         qi->lqi_id.qid_uid = uid;
1208                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1209
1210                         qi->lqi_id.qid_uid = gid;
1211                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1212                 }
1213         }
1214 }
1215
1216 /*
1217  * Concurrency: ->loo_object_release() is called under site spin-lock.
1218  */
1219 static void osd_object_release(const struct lu_env *env,
1220                                struct lu_object *l)
1221 {
1222 }
1223
1224 /*
1225  * Concurrency: shouldn't matter.
1226  */
1227 static int osd_object_print(const struct lu_env *env, void *cookie,
1228                             lu_printer_t p, const struct lu_object *l)
1229 {
1230         struct osd_object *o = osd_obj(l);
1231         struct iam_descr  *d;
1232
1233         if (o->oo_dir != NULL)
1234                 d = o->oo_dir->od_container.ic_descr;
1235         else
1236                 d = NULL;
1237         return (*p)(env, cookie,
1238                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1239                     o, o->oo_inode,
1240                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1241                     o->oo_inode ? o->oo_inode->i_generation : 0,
1242                     d ? d->id_ops->id_name : "plain");
1243 }
1244
1245 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1246
1247 /*
1248  * Concurrency: shouldn't matter.
1249  */
1250 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1251                struct obd_statfs *sfs)
1252 {
1253         struct osd_device  *osd = osd_dt_dev(d);
1254         struct super_block *sb = osd_sb(osd);
1255         struct kstatfs     *ksfs;
1256         int result = 0;
1257
1258         if (unlikely(osd->od_mnt == NULL))
1259                 return -EINPROGRESS;
1260
1261         /* osd_lproc.c call this without env, allocate ksfs for that case */
1262         if (unlikely(env == NULL)) {
1263                 OBD_ALLOC_PTR(ksfs);
1264                 if (ksfs == NULL)
1265                         return -ENOMEM;
1266         } else {
1267                 ksfs = &osd_oti_get(env)->oti_ksfs;
1268         }
1269
1270         spin_lock(&osd->od_osfs_lock);
1271         result = sb->s_op->statfs(sb->s_root, ksfs);
1272         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1273                 statfs_pack(sfs, ksfs);
1274                 if (sb->s_flags & MS_RDONLY)
1275                         sfs->os_state = OS_STATE_READONLY;
1276         }
1277
1278         spin_unlock(&osd->od_osfs_lock);
1279
1280         if (unlikely(env == NULL))
1281                 OBD_FREE_PTR(ksfs);
1282
1283         /* Reserve a small amount of space for local objects like last_rcvd,
1284          * llog, quota files, ... */
1285         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1286                 sfs->os_bavail = 0;
1287         } else {
1288                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1289                 /** Take out metadata overhead for indirect blocks */
1290                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1291         }
1292
1293         return result;
1294 }
1295
1296 /**
1297  * Estimate space needed for file creations. We assume the largest filename
1298  * which is 2^64 - 1, hence a filename of 20 chars.
1299  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1300  */
1301 #ifdef __LDISKFS_DIR_REC_LEN
1302 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1303 #else
1304 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1305 #endif
1306
1307 /*
1308  * Concurrency: doesn't access mutable data.
1309  */
1310 static void osd_conf_get(const struct lu_env *env,
1311                          const struct dt_device *dev,
1312                          struct dt_device_param *param)
1313 {
1314         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1315         int                ea_overhead;
1316
1317         /*
1318          * XXX should be taken from not-yet-existing fs abstraction layer.
1319          */
1320         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1321         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1322         param->ddp_block_shift  = sb->s_blocksize_bits;
1323         param->ddp_mount_type     = LDD_MT_LDISKFS;
1324         param->ddp_maxbytes       = sb->s_maxbytes;
1325         /* Overhead estimate should be fairly accurate, so we really take a tiny
1326          * error margin which also avoids fragmenting the filesystem too much */
1327         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1328         /* inode are statically allocated, so per-inode space consumption
1329          * is the space consumed by the directory entry */
1330         param->ddp_inodespace     = PER_OBJ_USAGE;
1331         /* per-fragment overhead to be used by the client code */
1332         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1333         param->ddp_mntopts      = 0;
1334         if (test_opt(sb, XATTR_USER))
1335                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1336         if (test_opt(sb, POSIX_ACL))
1337                 param->ddp_mntopts |= MNTOPT_ACL;
1338
1339         /* LOD might calculate the max stripe count based on max_ea_size,
1340          * so we need take account in the overhead as well,
1341          * xattr_header + magic + xattr_entry_head */
1342         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1343                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1344
1345 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1346         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1347                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1348                                                                 ea_overhead;
1349         else
1350 #endif
1351                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1352 }
1353
1354 /*
1355  * Concurrency: shouldn't matter.
1356  */
1357 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1358 {
1359         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1360         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1361 }
1362
1363 /**
1364  * Start commit for OSD device.
1365  *
1366  * An implementation of dt_commit_async method for OSD device.
1367  * Asychronously starts underlayng fs sync and thereby a transaction
1368  * commit.
1369  *
1370  * \param env environment
1371  * \param d dt device
1372  *
1373  * \see dt_device_operations
1374  */
1375 static int osd_commit_async(const struct lu_env *env,
1376                             struct dt_device *d)
1377 {
1378         struct super_block *s = osd_sb(osd_dt_dev(d));
1379         ENTRY;
1380
1381         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1382         RETURN(s->s_op->sync_fs(s, 0));
1383 }
1384
1385 /*
1386  * Concurrency: shouldn't matter.
1387  */
1388
1389 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1390 {
1391         struct super_block *sb = osd_sb(osd_dt_dev(d));
1392         struct block_device *dev = sb->s_bdev;
1393 #ifdef HAVE_DEV_SET_RDONLY
1394         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1395         int rc = 0;
1396 #else
1397         int rc = -EOPNOTSUPP;
1398 #endif
1399         ENTRY;
1400
1401 #ifdef HAVE_DEV_SET_RDONLY
1402         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1403
1404         if (jdev && (jdev != dev)) {
1405                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1406                        (long)jdev);
1407                 dev_set_rdonly(jdev);
1408         }
1409         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1410         dev_set_rdonly(dev);
1411 #else
1412         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1413                osd_dt_dev(d)->od_svname, (long)dev, rc);
1414 #endif
1415         RETURN(rc);
1416 }
1417
1418 /*
1419  * Concurrency: serialization provided by callers.
1420  */
1421 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1422                               int mode, unsigned long timeout, __u32 alg,
1423                               struct lustre_capa_key *keys)
1424 {
1425         struct osd_device *dev = osd_dt_dev(d);
1426         ENTRY;
1427
1428         dev->od_fl_capa = mode;
1429         dev->od_capa_timeout = timeout;
1430         dev->od_capa_alg = alg;
1431         dev->od_capa_keys = keys;
1432         RETURN(0);
1433 }
1434
1435 /**
1436  * Note: we do not count into QUOTA here.
1437  * If we mount with --data_journal we may need more.
1438  */
1439 const int osd_dto_credits_noquota[DTO_NR] = {
1440         /**
1441          * Insert.
1442          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1443          * SINGLEDATA_TRANS_BLOCKS(8)
1444          * XXX Note: maybe iam need more, since iam have more level than
1445          *           EXT3 htree.
1446          */
1447         [DTO_INDEX_INSERT]  = 16,
1448         /**
1449          * Delete
1450          * just modify a single entry, probably merge few within a block
1451          */
1452         [DTO_INDEX_DELETE]  = 1,
1453         /**
1454          * Used for OI scrub
1455          */
1456         [DTO_INDEX_UPDATE]  = 16,
1457         /**
1458          * 4(inode, inode bits, groups, GDT)
1459          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1460          */
1461         [DTO_OBJECT_CREATE] = 4,
1462         /**
1463          * 4(inode, inode bits, groups, GDT)
1464          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1465          */
1466         [DTO_OBJECT_DELETE] = 4,
1467         /**
1468          * Attr set credits (inode)
1469          */
1470         [DTO_ATTR_SET_BASE] = 1,
1471         /**
1472          * Xattr set. The same as xattr of EXT3.
1473          * DATA_TRANS_BLOCKS(14)
1474          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1475          * are also counted in. Do not know why?
1476          */
1477         [DTO_XATTR_SET]     = 14,
1478         /**
1479          * credits for inode change during write.
1480          */
1481         [DTO_WRITE_BASE]    = 3,
1482         /**
1483          * credits for single block write.
1484          */
1485         [DTO_WRITE_BLOCK]   = 14,
1486         /**
1487          * Attr set credits for chown.
1488          * This is extra credits for setattr, and it is null without quota
1489          */
1490         [DTO_ATTR_SET_CHOWN] = 0
1491 };
1492
1493 static const struct dt_device_operations osd_dt_ops = {
1494         .dt_root_get       = osd_root_get,
1495         .dt_statfs         = osd_statfs,
1496         .dt_trans_create   = osd_trans_create,
1497         .dt_trans_start    = osd_trans_start,
1498         .dt_trans_stop     = osd_trans_stop,
1499         .dt_trans_cb_add   = osd_trans_cb_add,
1500         .dt_conf_get       = osd_conf_get,
1501         .dt_sync           = osd_sync,
1502         .dt_ro             = osd_ro,
1503         .dt_commit_async   = osd_commit_async,
1504         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1505 };
1506
1507 static void osd_object_read_lock(const struct lu_env *env,
1508                                  struct dt_object *dt, unsigned role)
1509 {
1510         struct osd_object *obj = osd_dt_obj(dt);
1511         struct osd_thread_info *oti = osd_oti_get(env);
1512
1513         LINVRNT(osd_invariant(obj));
1514
1515         LASSERT(obj->oo_owner != env);
1516         down_read_nested(&obj->oo_sem, role);
1517
1518         LASSERT(obj->oo_owner == NULL);
1519         oti->oti_r_locks++;
1520 }
1521
1522 static void osd_object_write_lock(const struct lu_env *env,
1523                                   struct dt_object *dt, unsigned role)
1524 {
1525         struct osd_object *obj = osd_dt_obj(dt);
1526         struct osd_thread_info *oti = osd_oti_get(env);
1527
1528         LINVRNT(osd_invariant(obj));
1529
1530         LASSERT(obj->oo_owner != env);
1531         down_write_nested(&obj->oo_sem, role);
1532
1533         LASSERT(obj->oo_owner == NULL);
1534         obj->oo_owner = env;
1535         oti->oti_w_locks++;
1536 }
1537
1538 static void osd_object_read_unlock(const struct lu_env *env,
1539                                    struct dt_object *dt)
1540 {
1541         struct osd_object *obj = osd_dt_obj(dt);
1542         struct osd_thread_info *oti = osd_oti_get(env);
1543
1544         LINVRNT(osd_invariant(obj));
1545
1546         LASSERT(oti->oti_r_locks > 0);
1547         oti->oti_r_locks--;
1548         up_read(&obj->oo_sem);
1549 }
1550
1551 static void osd_object_write_unlock(const struct lu_env *env,
1552                                     struct dt_object *dt)
1553 {
1554         struct osd_object *obj = osd_dt_obj(dt);
1555         struct osd_thread_info *oti = osd_oti_get(env);
1556
1557         LINVRNT(osd_invariant(obj));
1558
1559         LASSERT(obj->oo_owner == env);
1560         LASSERT(oti->oti_w_locks > 0);
1561         oti->oti_w_locks--;
1562         obj->oo_owner = NULL;
1563         up_write(&obj->oo_sem);
1564 }
1565
1566 static int osd_object_write_locked(const struct lu_env *env,
1567                                    struct dt_object *dt)
1568 {
1569         struct osd_object *obj = osd_dt_obj(dt);
1570
1571         LINVRNT(osd_invariant(obj));
1572
1573         return obj->oo_owner == env;
1574 }
1575
1576 static int capa_is_sane(const struct lu_env *env,
1577                         struct osd_device *dev,
1578                         struct lustre_capa *capa,
1579                         struct lustre_capa_key *keys)
1580 {
1581         struct osd_thread_info *oti = osd_oti_get(env);
1582         struct lustre_capa *tcapa = &oti->oti_capa;
1583         struct obd_capa *oc;
1584         int i, rc = 0;
1585         ENTRY;
1586
1587         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1588         if (oc) {
1589                 if (capa_is_expired(oc)) {
1590                         DEBUG_CAPA(D_ERROR, capa, "expired");
1591                         rc = -ESTALE;
1592                 }
1593                 capa_put(oc);
1594                 RETURN(rc);
1595         }
1596
1597         if (capa_is_expired_sec(capa)) {
1598                 DEBUG_CAPA(D_ERROR, capa, "expired");
1599                 RETURN(-ESTALE);
1600         }
1601
1602         spin_lock(&capa_lock);
1603         for (i = 0; i < 2; i++) {
1604                 if (keys[i].lk_keyid == capa->lc_keyid) {
1605                         oti->oti_capa_key = keys[i];
1606                         break;
1607                 }
1608         }
1609         spin_unlock(&capa_lock);
1610
1611         if (i == 2) {
1612                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1613                 RETURN(-ESTALE);
1614         }
1615
1616         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1617         if (rc)
1618                 RETURN(rc);
1619
1620         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1621                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1622                 RETURN(-EACCES);
1623         }
1624
1625         oc = capa_add(dev->od_capa_hash, capa);
1626         capa_put(oc);
1627
1628         RETURN(0);
1629 }
1630
1631 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1632                     struct lustre_capa *capa, __u64 opc)
1633 {
1634         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1635         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1636         struct lu_capainfo *lci;
1637         int rc;
1638
1639         if (!osd->od_fl_capa)
1640                 return 0;
1641
1642         if (capa == BYPASS_CAPA)
1643                 return 0;
1644
1645         lci = lu_capainfo_get(env);
1646         if (unlikely(lci == NULL))
1647                 return 0;
1648
1649         if (lci->lci_auth == LC_ID_NONE)
1650                 return 0;
1651
1652         if (capa == NULL) {
1653                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1654                        osd_name(osd), PFID(fid), -EACCES);
1655                 return -EACCES;
1656         }
1657
1658         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1659                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1660                            PFID(fid));
1661                 return -EACCES;
1662         }
1663
1664         if (!capa_opc_supported(capa, opc)) {
1665                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1666                 return -EACCES;
1667         }
1668
1669         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1670         if (rc != 0) {
1671                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1672                 return -EACCES;
1673         }
1674
1675         return 0;
1676 }
1677
1678 static struct timespec *osd_inode_time(const struct lu_env *env,
1679                                        struct inode *inode, __u64 seconds)
1680 {
1681         struct osd_thread_info  *oti = osd_oti_get(env);
1682         struct timespec         *t   = &oti->oti_time;
1683
1684         t->tv_sec = seconds;
1685         t->tv_nsec = 0;
1686         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1687         return t;
1688 }
1689
1690
1691 static void osd_inode_getattr(const struct lu_env *env,
1692                               struct inode *inode, struct lu_attr *attr)
1693 {
1694         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1695                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1696                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1697                            LA_TYPE;
1698
1699         attr->la_atime   = LTIME_S(inode->i_atime);
1700         attr->la_mtime   = LTIME_S(inode->i_mtime);
1701         attr->la_ctime   = LTIME_S(inode->i_ctime);
1702         attr->la_mode    = inode->i_mode;
1703         attr->la_size    = i_size_read(inode);
1704         attr->la_blocks  = inode->i_blocks;
1705         attr->la_uid     = i_uid_read(inode);
1706         attr->la_gid     = i_gid_read(inode);
1707         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1708         attr->la_nlink   = inode->i_nlink;
1709         attr->la_rdev    = inode->i_rdev;
1710         attr->la_blksize = 1 << inode->i_blkbits;
1711         attr->la_blkbits = inode->i_blkbits;
1712 }
1713
1714 static int osd_attr_get(const struct lu_env *env,
1715                         struct dt_object *dt,
1716                         struct lu_attr *attr,
1717                         struct lustre_capa *capa)
1718 {
1719         struct osd_object *obj = osd_dt_obj(dt);
1720
1721         LASSERT(dt_object_exists(dt));
1722         LASSERT(!dt_object_remote(dt));
1723         LINVRNT(osd_invariant(obj));
1724
1725         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1726                 return -EACCES;
1727
1728         spin_lock(&obj->oo_guard);
1729         osd_inode_getattr(env, obj->oo_inode, attr);
1730         spin_unlock(&obj->oo_guard);
1731         return 0;
1732 }
1733
1734 static int osd_declare_attr_set(const struct lu_env *env,
1735                                 struct dt_object *dt,
1736                                 const struct lu_attr *attr,
1737                                 struct thandle *handle)
1738 {
1739         struct osd_thandle     *oh;
1740         struct osd_object      *obj;
1741         struct osd_thread_info *info = osd_oti_get(env);
1742         struct lquota_id_info  *qi = &info->oti_qi;
1743         qid_t                   uid;
1744         qid_t                   gid;
1745         long long               bspace;
1746         int                     rc = 0;
1747         bool                    enforce;
1748         ENTRY;
1749
1750         LASSERT(dt != NULL);
1751         LASSERT(handle != NULL);
1752
1753         obj = osd_dt_obj(dt);
1754         LASSERT(osd_invariant(obj));
1755
1756         oh = container_of0(handle, struct osd_thandle, ot_super);
1757         LASSERT(oh->ot_handle == NULL);
1758
1759         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1760                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1761
1762         if (attr == NULL || obj->oo_inode == NULL)
1763                 RETURN(rc);
1764
1765         bspace   = obj->oo_inode->i_blocks;
1766         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1767         bspace   = toqb(bspace);
1768
1769         /* Changing ownership is always preformed by super user, it should not
1770          * fail with EDQUOT.
1771          *
1772          * We still need to call the osd_declare_qid() to calculate the journal
1773          * credits for updating quota accounting files and to trigger quota
1774          * space adjustment once the operation is completed.*/
1775         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1776                 /* USERQUOTA */
1777                 uid = i_uid_read(obj->oo_inode);
1778                 qi->lqi_type = USRQUOTA;
1779                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1780                 /* inode accounting */
1781                 qi->lqi_is_blk = false;
1782
1783                 /* one more inode for the new uid ... */
1784                 qi->lqi_id.qid_uid = attr->la_uid;
1785                 qi->lqi_space      = 1;
1786                 /* Reserve credits for the new uid */
1787                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1788                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1789                         rc = 0;
1790                 if (rc)
1791                         RETURN(rc);
1792
1793                 /* and one less inode for the current uid */
1794                 qi->lqi_id.qid_uid = uid;
1795                 qi->lqi_space      = -1;
1796                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1797                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1798                         rc = 0;
1799                 if (rc)
1800                         RETURN(rc);
1801
1802                 /* block accounting */
1803                 qi->lqi_is_blk = true;
1804
1805                 /* more blocks for the new uid ... */
1806                 qi->lqi_id.qid_uid = attr->la_uid;
1807                 qi->lqi_space      = bspace;
1808                 /*
1809                  * Credits for the new uid has been reserved, re-use "obj"
1810                  * to save credit reservation.
1811                  */
1812                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1813                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1814                         rc = 0;
1815                 if (rc)
1816                         RETURN(rc);
1817
1818                 /* and finally less blocks for the current uid */
1819                 qi->lqi_id.qid_uid = uid;
1820                 qi->lqi_space      = -bspace;
1821                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1822                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1823                         rc = 0;
1824                 if (rc)
1825                         RETURN(rc);
1826
1827                 /* GROUP QUOTA */
1828                 gid = i_gid_read(obj->oo_inode);
1829                 qi->lqi_type = GRPQUOTA;
1830                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1831
1832                 /* inode accounting */
1833                 qi->lqi_is_blk = false;
1834
1835                 /* one more inode for the new gid ... */
1836                 qi->lqi_id.qid_gid = attr->la_gid;
1837                 qi->lqi_space      = 1;
1838                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1839                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1840                         rc = 0;
1841                 if (rc)
1842                         RETURN(rc);
1843
1844                 /* and one less inode for the current gid */
1845                 qi->lqi_id.qid_gid = gid;
1846                 qi->lqi_space      = -1;
1847                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1848                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1849                         rc = 0;
1850                 if (rc)
1851                         RETURN(rc);
1852
1853                 /* block accounting */
1854                 qi->lqi_is_blk = true;
1855
1856                 /* more blocks for the new gid ... */
1857                 qi->lqi_id.qid_gid = attr->la_gid;
1858                 qi->lqi_space      = bspace;
1859                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1860                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1861                         rc = 0;
1862                 if (rc)
1863                         RETURN(rc);
1864
1865                 /* and finally less blocks for the current gid */
1866                 qi->lqi_id.qid_gid = gid;
1867                 qi->lqi_space      = -bspace;
1868                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1869                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1870                         rc = 0;
1871                 if (rc)
1872                         RETURN(rc);
1873         }
1874
1875         RETURN(rc);
1876 }
1877
1878 static int osd_inode_setattr(const struct lu_env *env,
1879                              struct inode *inode, const struct lu_attr *attr)
1880 {
1881         __u64 bits = attr->la_valid;
1882
1883         /* Only allow set size for regular file */
1884         if (!S_ISREG(inode->i_mode))
1885                 bits &= ~(LA_SIZE | LA_BLOCKS);
1886
1887         if (bits == 0)
1888                 return 0;
1889
1890         if (bits & LA_ATIME)
1891                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1892         if (bits & LA_CTIME)
1893                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1894         if (bits & LA_MTIME)
1895                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1896         if (bits & LA_SIZE) {
1897                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1898                 i_size_write(inode, attr->la_size);
1899         }
1900
1901 #if 0
1902         /* OSD should not change "i_blocks" which is used by quota.
1903          * "i_blocks" should be changed by ldiskfs only. */
1904         if (bits & LA_BLOCKS)
1905                 inode->i_blocks = attr->la_blocks;
1906 #endif
1907         if (bits & LA_MODE)
1908                 inode->i_mode = (inode->i_mode & S_IFMT) |
1909                                 (attr->la_mode & ~S_IFMT);
1910         if (bits & LA_UID)
1911                 i_uid_write(inode, attr->la_uid);
1912         if (bits & LA_GID)
1913                 i_gid_write(inode, attr->la_gid);
1914         if (bits & LA_NLINK)
1915                 set_nlink(inode, attr->la_nlink);
1916         if (bits & LA_RDEV)
1917                 inode->i_rdev = attr->la_rdev;
1918
1919         if (bits & LA_FLAGS) {
1920                 /* always keep S_NOCMTIME */
1921                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1922                                  S_NOCMTIME;
1923         }
1924         return 0;
1925 }
1926
1927 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1928 {
1929         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1930             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1931                 struct iattr    iattr;
1932                 int             rc;
1933
1934                 ll_vfs_dq_init(inode);
1935                 iattr.ia_valid = 0;
1936                 if (attr->la_valid & LA_UID)
1937                         iattr.ia_valid |= ATTR_UID;
1938                 if (attr->la_valid & LA_GID)
1939                         iattr.ia_valid |= ATTR_GID;
1940                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1941                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1942
1943                 rc = ll_vfs_dq_transfer(inode, &iattr);
1944                 if (rc) {
1945                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1946                                "enforcement enabled on the ldiskfs "
1947                                "filesystem?\n", inode->i_sb->s_id, rc);
1948                         return rc;
1949                 }
1950         }
1951         return 0;
1952 }
1953
1954 static int osd_attr_set(const struct lu_env *env,
1955                         struct dt_object *dt,
1956                         const struct lu_attr *attr,
1957                         struct thandle *handle,
1958                         struct lustre_capa *capa)
1959 {
1960         struct osd_object *obj = osd_dt_obj(dt);
1961         struct inode      *inode;
1962         int rc;
1963
1964         LASSERT(handle != NULL);
1965         LASSERT(dt_object_exists(dt));
1966         LASSERT(!dt_object_remote(dt));
1967         LASSERT(osd_invariant(obj));
1968
1969         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1970                 return -EACCES;
1971
1972         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1973
1974         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1975                 struct osd_thread_info  *oti  = osd_oti_get(env);
1976                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1977                 struct lu_fid           *fid1 = &oti->oti_fid;
1978                 struct osd_inode_id     *id   = &oti->oti_id;
1979                 struct iam_path_descr   *ipd;
1980                 struct iam_container    *bag;
1981                 struct osd_thandle      *oh;
1982                 int                      rc;
1983
1984                 fid_cpu_to_be(fid1, fid0);
1985                 memset(id, 1, sizeof(*id));
1986                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1987                                   fid0)->oi_dir.od_container;
1988                 ipd = osd_idx_ipd_get(env, bag);
1989                 if (unlikely(ipd == NULL))
1990                         RETURN(-ENOMEM);
1991
1992                 oh = container_of0(handle, struct osd_thandle, ot_super);
1993                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1994                                 (const struct iam_rec *)id, ipd);
1995                 osd_ipd_put(env, bag, ipd);
1996                 return(rc > 0 ? 0 : rc);
1997         }
1998
1999         inode = obj->oo_inode;
2000
2001         rc = osd_quota_transfer(inode, attr);
2002         if (rc)
2003                 return rc;
2004
2005         spin_lock(&obj->oo_guard);
2006         rc = osd_inode_setattr(env, inode, attr);
2007         spin_unlock(&obj->oo_guard);
2008
2009         if (!rc)
2010                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2011         return rc;
2012 }
2013
2014 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2015                                     struct osd_object *obj,
2016                                     const char *name, const int namelen)
2017 {
2018         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2019 }
2020
2021 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2022                       umode_t mode, struct dt_allocation_hint *hint,
2023                       struct thandle *th)
2024 {
2025         int result;
2026         struct osd_device  *osd = osd_obj2dev(obj);
2027         struct osd_thandle *oth;
2028         struct dt_object   *parent = NULL;
2029         struct inode       *inode;
2030
2031         LINVRNT(osd_invariant(obj));
2032         LASSERT(obj->oo_inode == NULL);
2033         LASSERT(obj->oo_hl_head == NULL);
2034
2035         if (S_ISDIR(mode) && ldiskfs_pdo) {
2036                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2037                 if (obj->oo_hl_head == NULL)
2038                         return -ENOMEM;
2039         }
2040
2041         oth = container_of(th, struct osd_thandle, ot_super);
2042         LASSERT(oth->ot_handle->h_transaction != NULL);
2043
2044         if (hint && hint->dah_parent)
2045                 parent = hint->dah_parent;
2046
2047         inode = ldiskfs_create_inode(oth->ot_handle,
2048                                      parent ? osd_dt_obj(parent)->oo_inode :
2049                                               osd_sb(osd)->s_root->d_inode,
2050                                      mode);
2051         if (!IS_ERR(inode)) {
2052                 /* Do not update file c/mtime in ldiskfs.
2053                  * NB: don't need any lock because no contention at this
2054                  * early stage */
2055                 inode->i_flags |= S_NOCMTIME;
2056
2057                 /* For new created object, it must be consistent,
2058                  * and it is unnecessary to scrub against it. */
2059                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2060                 obj->oo_inode = inode;
2061                 result = 0;
2062         } else {
2063                 if (obj->oo_hl_head != NULL) {
2064                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2065                         obj->oo_hl_head = NULL;
2066                 }
2067                 result = PTR_ERR(inode);
2068         }
2069         LINVRNT(osd_invariant(obj));
2070         return result;
2071 }
2072
2073 enum {
2074         OSD_NAME_LEN = 255
2075 };
2076
2077 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2078                      struct lu_attr *attr,
2079                      struct dt_allocation_hint *hint,
2080                      struct dt_object_format *dof,
2081                      struct thandle *th)
2082 {
2083         int result;
2084         struct osd_thandle *oth;
2085         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2086
2087         LASSERT(S_ISDIR(attr->la_mode));
2088
2089         oth = container_of(th, struct osd_thandle, ot_super);
2090         LASSERT(oth->ot_handle->h_transaction != NULL);
2091         result = osd_mkfile(info, obj, mode, hint, th);
2092
2093         return result;
2094 }
2095
2096 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2097                         struct lu_attr *attr,
2098                         struct dt_allocation_hint *hint,
2099                         struct dt_object_format *dof,
2100                         struct thandle *th)
2101 {
2102         int result;
2103         struct osd_thandle *oth;
2104         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2105
2106         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2107
2108         LASSERT(S_ISREG(attr->la_mode));
2109
2110         oth = container_of(th, struct osd_thandle, ot_super);
2111         LASSERT(oth->ot_handle->h_transaction != NULL);
2112
2113         result = osd_mkfile(info, obj, mode, hint, th);
2114         if (result == 0) {
2115                 LASSERT(obj->oo_inode != NULL);
2116                 if (feat->dif_flags & DT_IND_VARKEY)
2117                         result = iam_lvar_create(obj->oo_inode,
2118                                                  feat->dif_keysize_max,
2119                                                  feat->dif_ptrsize,
2120                                                  feat->dif_recsize_max,
2121                                                  oth->ot_handle);
2122                 else
2123                         result = iam_lfix_create(obj->oo_inode,
2124                                                  feat->dif_keysize_max,
2125                                                  feat->dif_ptrsize,
2126                                                  feat->dif_recsize_max,
2127                                                  oth->ot_handle);
2128
2129         }
2130         return result;
2131 }
2132
2133 static int osd_mkreg(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_ISREG(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_mksym(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         LASSERT(S_ISLNK(attr->la_mode));
2151         return osd_mkfile(info, obj, (attr->la_mode &
2152                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2153 }
2154
2155 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2156                      struct lu_attr *attr,
2157                      struct dt_allocation_hint *hint,
2158                      struct dt_object_format *dof,
2159                      struct thandle *th)
2160 {
2161         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2162         int result;
2163
2164         LINVRNT(osd_invariant(obj));
2165         LASSERT(obj->oo_inode == NULL);
2166         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2167                 S_ISFIFO(mode) || S_ISSOCK(mode));
2168
2169         result = osd_mkfile(info, obj, mode, hint, th);
2170         if (result == 0) {
2171                 LASSERT(obj->oo_inode != NULL);
2172                 /*
2173                  * This inode should be marked dirty for i_rdev.  Currently
2174                  * that is done in the osd_attr_init().
2175                  */
2176                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2177                                    attr->la_rdev);
2178         }
2179         LINVRNT(osd_invariant(obj));
2180         return result;
2181 }
2182
2183 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2184                               struct lu_attr *,
2185                               struct dt_allocation_hint *hint,
2186                               struct dt_object_format *dof,
2187                               struct thandle *);
2188
2189 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2190 {
2191         osd_obj_type_f result;
2192
2193         switch (type) {
2194         case DFT_DIR:
2195                 result = osd_mkdir;
2196                 break;
2197         case DFT_REGULAR:
2198                 result = osd_mkreg;
2199                 break;
2200         case DFT_SYM:
2201                 result = osd_mksym;
2202                 break;
2203         case DFT_NODE:
2204                 result = osd_mknod;
2205                 break;
2206         case DFT_INDEX:
2207                 result = osd_mk_index;
2208                 break;
2209
2210         default:
2211                 LBUG();
2212                 break;
2213         }
2214         return result;
2215 }
2216
2217
2218 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2219                         struct dt_object *parent, struct dt_object *child,
2220                         umode_t child_mode)
2221 {
2222         LASSERT(ah);
2223
2224         ah->dah_parent = parent;
2225         ah->dah_mode = child_mode;
2226 }
2227
2228 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2229                           struct lu_attr *attr, struct dt_object_format *dof)
2230 {
2231         struct inode   *inode = obj->oo_inode;
2232         __u64           valid = attr->la_valid;
2233         int             result;
2234
2235         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2236
2237         if (dof->dof_type != DFT_NODE)
2238                 attr->la_valid &= ~LA_RDEV;
2239         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2240                 attr->la_valid &= ~LA_ATIME;
2241         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2242                 attr->la_valid &= ~LA_CTIME;
2243         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2244                 attr->la_valid &= ~LA_MTIME;
2245
2246         result = osd_quota_transfer(inode, attr);
2247         if (result)
2248                 return;
2249
2250         if (attr->la_valid != 0) {
2251                 result = osd_inode_setattr(info->oti_env, inode, attr);
2252                 /*
2253                  * The osd_inode_setattr() should always succeed here.  The
2254                  * only error that could be returned is EDQUOT when we are
2255                  * trying to change the UID or GID of the inode. However, this
2256                  * should not happen since quota enforcement is no longer
2257                  * enabled on ldiskfs (lquota takes care of it).
2258                  */
2259                 LASSERTF(result == 0, "%d\n", result);
2260                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2261         }
2262
2263         attr->la_valid = valid;
2264 }
2265
2266 /**
2267  * Helper function for osd_object_create()
2268  *
2269  * \retval 0, on success
2270  */
2271 static int __osd_object_create(struct osd_thread_info *info,
2272                                struct osd_object *obj, struct lu_attr *attr,
2273                                struct dt_allocation_hint *hint,
2274                                struct dt_object_format *dof,
2275                                struct thandle *th)
2276 {
2277         int     result;
2278         __u32   umask;
2279
2280         /* we drop umask so that permissions we pass are not affected */
2281         umask = current->fs->umask;
2282         current->fs->umask = 0;
2283
2284         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2285                                                   th);
2286         if (result == 0) {
2287                 osd_attr_init(info, obj, attr, dof);
2288                 osd_object_init0(obj);
2289                 /* bz 24037 */
2290                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2291                         unlock_new_inode(obj->oo_inode);
2292         }
2293
2294         /* restore previous umask value */
2295         current->fs->umask = umask;
2296
2297         return result;
2298 }
2299
2300 /**
2301  * Helper function for osd_object_create()
2302  *
2303  * \retval 0, on success
2304  */
2305 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2306                            const struct lu_fid *fid, struct thandle *th)
2307 {
2308         struct osd_thread_info *info = osd_oti_get(env);
2309         struct osd_inode_id    *id   = &info->oti_id;
2310         struct osd_device      *osd  = osd_obj2dev(obj);
2311         struct osd_thandle     *oh;
2312
2313         LASSERT(obj->oo_inode != NULL);
2314
2315         oh = container_of0(th, struct osd_thandle, ot_super);
2316         LASSERT(oh->ot_handle);
2317
2318         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2319         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2320 }
2321
2322 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2323                    obd_seq seq, struct lu_seq_range *range)
2324 {
2325         struct seq_server_site  *ss = osd_seq_site(osd);
2326
2327         if (fid_seq_is_idif(seq)) {
2328                 fld_range_set_ost(range);
2329                 range->lsr_index = idif_ost_idx(seq);
2330                 return 0;
2331         }
2332
2333         if (!fid_seq_in_fldb(seq)) {
2334                 fld_range_set_mdt(range);
2335                 if (ss != NULL)
2336                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2337                          * at all */
2338                         range->lsr_index = ss->ss_node_id;
2339                 return 0;
2340         }
2341
2342         LASSERT(ss != NULL);
2343         fld_range_set_any(range);
2344         /* OSD will only do local fld lookup */
2345         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2346 }
2347
2348 /*
2349  * Concurrency: no external locking is necessary.
2350  */
2351 static int osd_declare_object_create(const struct lu_env *env,
2352                                      struct dt_object *dt,
2353                                      struct lu_attr *attr,
2354                                      struct dt_allocation_hint *hint,
2355                                      struct dt_object_format *dof,
2356                                      struct thandle *handle)
2357 {
2358         struct osd_thandle      *oh;
2359         int                      rc;
2360         ENTRY;
2361
2362         LASSERT(handle != NULL);
2363
2364         oh = container_of0(handle, struct osd_thandle, ot_super);
2365         LASSERT(oh->ot_handle == NULL);
2366
2367         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2368                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2369         /* Reuse idle OI block may cause additional one OI block
2370          * to be changed. */
2371         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2372                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2373
2374         /* If this is directory, then we expect . and .. to be inserted as
2375          * well. The one directory block always needs to be created for the
2376          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2377          * block), there is no danger of needing a tree for the first block.
2378          */
2379         if (attr && S_ISDIR(attr->la_mode)) {
2380                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2381                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2382                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2383         }
2384
2385         if (!attr)
2386                 RETURN(0);
2387
2388         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2389                                    osd_dt_obj(dt), false, NULL, false);
2390         if (rc != 0)
2391                 RETURN(rc);
2392
2393         RETURN(rc);
2394 }
2395
2396 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2397                              struct lu_attr *attr,
2398                              struct dt_allocation_hint *hint,
2399                              struct dt_object_format *dof, struct thandle *th)
2400 {
2401         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2402         struct osd_object       *obj    = osd_dt_obj(dt);
2403         struct osd_thread_info  *info   = osd_oti_get(env);
2404         int result;
2405         ENTRY;
2406
2407         LINVRNT(osd_invariant(obj));
2408         LASSERT(!dt_object_exists(dt));
2409         LASSERT(!dt_object_remote(dt));
2410         LASSERT(osd_write_locked(env, obj));
2411         LASSERT(th != NULL);
2412
2413         if (unlikely(fid_is_acct(fid)))
2414                 /* Quota files can't be created from the kernel any more,
2415                  * 'tune2fs -O quota' will take care of creating them */
2416                 RETURN(-EPERM);
2417
2418         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2419         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2420
2421         result = __osd_object_create(info, obj, attr, hint, dof, th);
2422         if (result == 0)
2423                 result = __osd_oi_insert(env, obj, fid, th);
2424
2425         LASSERT(ergo(result == 0,
2426                      dt_object_exists(dt) && !dt_object_remote(dt)));
2427
2428         LASSERT(osd_invariant(obj));
2429         RETURN(result);
2430 }
2431
2432 /**
2433  * Called to destroy on-disk representation of the object
2434  *
2435  * Concurrency: must be locked
2436  */
2437 static int osd_declare_object_destroy(const struct lu_env *env,
2438                                       struct dt_object *dt,
2439                                       struct thandle *th)
2440 {
2441         struct osd_object  *obj = osd_dt_obj(dt);
2442         struct inode       *inode = obj->oo_inode;
2443         struct osd_thandle *oh;
2444         int                 rc;
2445         ENTRY;
2446
2447         oh = container_of0(th, struct osd_thandle, ot_super);
2448         LASSERT(oh->ot_handle == NULL);
2449         LASSERT(inode);
2450
2451         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2452                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2453         /* Recycle idle OI leaf may cause additional three OI blocks
2454          * to be changed. */
2455         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2456                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2457         /* one less inode */
2458         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2459                                    -1, oh, obj, false, NULL, false);
2460         if (rc)
2461                 RETURN(rc);
2462         /* data to be truncated */
2463         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2464                                    0, oh, obj, true, NULL, false);
2465         RETURN(rc);
2466 }
2467
2468 static int osd_object_destroy(const struct lu_env *env,
2469                               struct dt_object *dt,
2470                               struct thandle *th)
2471 {
2472         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2473         struct osd_object      *obj = osd_dt_obj(dt);
2474         struct inode           *inode = obj->oo_inode;
2475         struct osd_device      *osd = osd_obj2dev(obj);
2476         struct osd_thandle     *oh;
2477         int                     result;
2478         ENTRY;
2479
2480         oh = container_of0(th, struct osd_thandle, ot_super);
2481         LASSERT(oh->ot_handle);
2482         LASSERT(inode);
2483         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2484
2485         if (unlikely(fid_is_acct(fid)))
2486                 RETURN(-EPERM);
2487
2488         if (S_ISDIR(inode->i_mode)) {
2489                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2490                         inode->i_nlink == 2);
2491                 /* it will check/delete the inode from remote parent,
2492                  * how to optimize it? unlink performance impaction XXX */
2493                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2494                 if (result != 0 && result != -ENOENT) {
2495                         CERROR("%s: delete inode "DFID": rc = %d\n",
2496                                osd_name(osd), PFID(fid), result);
2497                 }
2498                 spin_lock(&obj->oo_guard);
2499                 clear_nlink(inode);
2500                 spin_unlock(&obj->oo_guard);
2501                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2502         }
2503
2504         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2505
2506         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2507                                OI_CHECK_FLD);
2508
2509         /* XXX: add to ext3 orphan list */
2510         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2511
2512         /* not needed in the cache anymore */
2513         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2514
2515         RETURN(0);
2516 }
2517
2518 /**
2519  * Put the fid into lustre_mdt_attrs, and then place the structure
2520  * inode's ea. This fid should not be altered during the life time
2521  * of the inode.
2522  *
2523  * \retval +ve, on success
2524  * \retval -ve, on error
2525  *
2526  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2527  */
2528 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2529                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2530 {
2531         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2532         int                      rc;
2533         ENTRY;
2534
2535         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2536                 RETURN(0);
2537
2538         lustre_lma_init(lma, fid, compat, incompat);
2539         lustre_lma_swab(lma);
2540
2541         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2542                              XATTR_CREATE);
2543         /* LMA may already exist, but we need to check that all the
2544          * desired compat/incompat flags have been added. */
2545         if (unlikely(rc == -EEXIST)) {
2546                 if (compat == 0 && incompat == 0)
2547                         RETURN(0);
2548
2549                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2550                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2551                                      LMA_OLD_SIZE);
2552                 if (rc <= 0)
2553                         RETURN(-EINVAL);
2554
2555                 lustre_lma_swab(lma);
2556                 if (!(~lma->lma_compat & compat) &&
2557                     !(~lma->lma_incompat & incompat))
2558                         RETURN(0);
2559
2560                 lma->lma_compat |= compat;
2561                 lma->lma_incompat |= incompat;
2562                 lustre_lma_swab(lma);
2563                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2564                                      sizeof(*lma), XATTR_REPLACE);
2565         }
2566
2567         RETURN(rc);
2568 }
2569
2570 /**
2571  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2572  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2573  * To have compatilibility with 1.8 ldiskfs driver we need to have
2574  * magic number at start of fid data.
2575  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2576  * its inmemory API.
2577  */
2578 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2579                                   const struct lu_fid *fid)
2580 {
2581         if (!fid_is_namespace_visible(fid) ||
2582             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2583                 param->edp_magic = 0;
2584                 return;
2585         }
2586
2587         param->edp_magic = LDISKFS_LUFID_MAGIC;
2588         param->edp_len =  sizeof(struct lu_fid) + 1;
2589         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2590 }
2591
2592 /**
2593  * Try to read the fid from inode ea into dt_rec.
2594  *
2595  * \param fid object fid.
2596  *
2597  * \retval 0 on success
2598  */
2599 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2600                           __u32 ino, struct lu_fid *fid,
2601                           struct osd_inode_id *id)
2602 {
2603         struct osd_thread_info *info  = osd_oti_get(env);
2604         struct inode           *inode;
2605         ENTRY;
2606
2607         osd_id_gen(id, ino, OSD_OII_NOGEN);
2608         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2609         if (IS_ERR(inode))
2610                 RETURN(PTR_ERR(inode));
2611
2612         iput(inode);
2613         RETURN(0);
2614 }
2615
2616 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2617                                         struct inode *dir,
2618                                         struct inode *parent_dir,
2619                                         const struct lu_fid *dot_fid,
2620                                         const struct lu_fid *dot_dot_fid,
2621                                         struct osd_thandle *oth)
2622 {
2623         struct ldiskfs_dentry_param *dot_ldp;
2624         struct ldiskfs_dentry_param *dot_dot_ldp;
2625
2626         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2627         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2628
2629         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2630         dot_ldp->edp_magic = 0;
2631         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2632                                         dir, dot_ldp, dot_dot_ldp);
2633 }
2634
2635 /**
2636  * Create an local agent inode for remote entry
2637  */
2638 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2639                                                   struct osd_device *osd,
2640                                                   struct osd_object *pobj,
2641                                                   const struct lu_fid *fid,
2642                                                   __u32 type,
2643                                                   struct thandle *th)
2644 {
2645         struct osd_thread_info  *info = osd_oti_get(env);
2646         struct inode            *local;
2647         struct osd_thandle      *oh;
2648         int                     rc;
2649         ENTRY;
2650
2651         LASSERT(th);
2652         oh = container_of(th, struct osd_thandle, ot_super);
2653         LASSERT(oh->ot_handle->h_transaction != NULL);
2654
2655         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2656         if (IS_ERR(local)) {
2657                 CERROR("%s: create local error %d\n", osd_name(osd),
2658                        (int)PTR_ERR(local));
2659                 RETURN(local);
2660         }
2661
2662         /* Set special LMA flag for local agent inode */
2663         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2664         if (rc != 0) {
2665                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2666                        osd_name(osd), PFID(fid), rc);
2667                 RETURN(ERR_PTR(rc));
2668         }
2669
2670         if (!S_ISDIR(type))
2671                 RETURN(local);
2672
2673         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2674                                          lu_object_fid(&pobj->oo_dt.do_lu),
2675                                          fid, oh);
2676         if (rc != 0) {
2677                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2678                         osd_name(osd), PFID(fid), rc);
2679                 RETURN(ERR_PTR(rc));
2680         }
2681
2682         RETURN(local);
2683 }
2684
2685 /**
2686  * Delete local agent inode for remote entry
2687  */
2688 static int osd_delete_local_agent_inode(const struct lu_env *env,
2689                                         struct osd_device *osd,
2690                                         const struct lu_fid *fid,
2691                                         __u32 ino, struct osd_thandle *oh)
2692 {
2693         struct osd_thread_info  *oti = osd_oti_get(env);
2694         struct osd_inode_id     *id = &oti->oti_id;
2695         struct inode            *inode;
2696         ENTRY;
2697
2698         id->oii_ino = le32_to_cpu(ino);
2699         id->oii_gen = OSD_OII_NOGEN;
2700         inode = osd_iget(oti, osd, id);
2701         if (IS_ERR(inode)) {
2702                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2703                        PFID(fid), id->oii_ino, id->oii_gen);
2704                 RETURN(PTR_ERR(inode));
2705         }
2706
2707         clear_nlink(inode);
2708         mark_inode_dirty(inode);
2709         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2710                 osd_name(osd), PFID(fid), inode->i_ino);
2711         iput(inode);
2712         RETURN(0);
2713 }
2714
2715 /**
2716  * OSD layer object create function for interoperability mode (b11826).
2717  * This is mostly similar to osd_object_create(). Only difference being, fid is
2718  * inserted into inode ea here.
2719  *
2720  * \retval   0, on success
2721  * \retval -ve, on error
2722  */
2723 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2724                                 struct lu_attr *attr,
2725                                 struct dt_allocation_hint *hint,
2726                                 struct dt_object_format *dof,
2727                                 struct thandle *th)
2728 {
2729         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2730         struct osd_object       *obj    = osd_dt_obj(dt);
2731         struct osd_thread_info  *info   = osd_oti_get(env);
2732         int                      result;
2733
2734         ENTRY;
2735
2736         LASSERT(osd_invariant(obj));
2737         LASSERT(!dt_object_exists(dt));
2738         LASSERT(!dt_object_remote(dt));
2739         LASSERT(osd_write_locked(env, obj));
2740         LASSERT(th != NULL);
2741
2742         if (unlikely(fid_is_acct(fid)))
2743                 /* Quota files can't be created from the kernel any more,
2744                  * 'tune2fs -O quota' will take care of creating them */
2745                 RETURN(-EPERM);
2746
2747         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2748         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2749
2750         result = __osd_object_create(info, obj, attr, hint, dof, th);
2751         if (result == 0)
2752                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2753                                 fid_is_on_ost(info, osd_obj2dev(obj),
2754                                               fid, OI_CHECK_FLD) ?
2755                                 LMAC_FID_ON_OST : 0, 0);
2756
2757         if (result == 0)
2758                 result = __osd_oi_insert(env, obj, fid, th);
2759
2760         LASSERT(ergo(result == 0,
2761                      dt_object_exists(dt) && !dt_object_remote(dt)));
2762         LINVRNT(osd_invariant(obj));
2763         RETURN(result);
2764 }
2765
2766 static int osd_declare_object_ref_add(const struct lu_env *env,
2767                                       struct dt_object *dt,
2768                                       struct thandle *handle)
2769 {
2770         struct osd_thandle       *oh;
2771
2772         /* it's possible that object doesn't exist yet */
2773         LASSERT(handle != NULL);
2774
2775         oh = container_of0(handle, struct osd_thandle, ot_super);
2776         LASSERT(oh->ot_handle == NULL);
2777
2778         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2779                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2780
2781         return 0;
2782 }
2783
2784 /*
2785  * Concurrency: @dt is write locked.
2786  */
2787 static int osd_object_ref_add(const struct lu_env *env,
2788                               struct dt_object *dt, struct thandle *th)
2789 {
2790         struct osd_object  *obj = osd_dt_obj(dt);
2791         struct inode       *inode = obj->oo_inode;
2792         struct osd_thandle *oh;
2793         int                 rc = 0;
2794
2795         LINVRNT(osd_invariant(obj));
2796         LASSERT(dt_object_exists(dt));
2797         LASSERT(!dt_object_remote(dt));
2798         LASSERT(osd_write_locked(env, obj));
2799         LASSERT(th != NULL);
2800
2801         oh = container_of0(th, struct osd_thandle, ot_super);
2802         LASSERT(oh->ot_handle != NULL);
2803
2804         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2805
2806         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2807                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2808         /*
2809          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2810          * (65000) subdirectories by storing "1" in i_nlink if the link count
2811          * would otherwise overflow. Directory tranversal tools understand
2812          * that (st_nlink == 1) indicates that the filesystem dose not track
2813          * hard links count on the directory, and will not abort subdirectory
2814          * scanning early once (st_nlink - 2) subdirs have been found.
2815          *
2816          * This also has to properly handle the case of inodes with nlink == 0
2817          * in case they are being linked into the PENDING directory
2818          */
2819         spin_lock(&obj->oo_guard);
2820         if (unlikely(inode->i_nlink == 0))
2821                 /* inc_nlink from 0 may cause WARN_ON */
2822                 set_nlink(inode, 1);
2823         else {
2824                 ldiskfs_inc_count(oh->ot_handle, inode);
2825                 if (!S_ISDIR(inode->i_mode))
2826                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2827         }
2828         spin_unlock(&obj->oo_guard);
2829
2830         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2831         LINVRNT(osd_invariant(obj));
2832
2833         return rc;
2834 }
2835
2836 static int osd_declare_object_ref_del(const struct lu_env *env,
2837                                       struct dt_object *dt,
2838                                       struct thandle *handle)
2839 {
2840         struct osd_thandle *oh;
2841
2842         LASSERT(dt_object_exists(dt));
2843         LASSERT(!dt_object_remote(dt));
2844         LASSERT(handle != NULL);
2845
2846         oh = container_of0(handle, struct osd_thandle, ot_super);
2847         LASSERT(oh->ot_handle == NULL);
2848
2849         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2850                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2851
2852         return 0;
2853 }
2854
2855 /*
2856  * Concurrency: @dt is write locked.
2857  */
2858 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2859                               struct thandle *th)
2860 {
2861         struct osd_object       *obj = osd_dt_obj(dt);
2862         struct inode            *inode = obj->oo_inode;
2863         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2864         struct osd_thandle      *oh;
2865
2866         LINVRNT(osd_invariant(obj));
2867         LASSERT(dt_object_exists(dt));
2868         LASSERT(!dt_object_remote(dt));
2869         LASSERT(osd_write_locked(env, obj));
2870         LASSERT(th != NULL);
2871
2872         oh = container_of0(th, struct osd_thandle, ot_super);
2873         LASSERT(oh->ot_handle != NULL);
2874
2875         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2876
2877         spin_lock(&obj->oo_guard);
2878         /* That can be result of upgrade from old Lustre version and
2879          * applied only to local files.  Just skip this ref_del call.
2880          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2881         if (inode->i_nlink == 0) {
2882                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2883                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2884                              ", maybe an upgraded file? (LU-3915)\n",
2885                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2886                 spin_unlock(&obj->oo_guard);
2887                 return 0;
2888         }
2889
2890         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2891                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2892
2893         ldiskfs_dec_count(oh->ot_handle, inode);
2894         spin_unlock(&obj->oo_guard);
2895
2896         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2897         LINVRNT(osd_invariant(obj));
2898
2899         return 0;
2900 }
2901
2902 /*
2903  * Get the 64-bit version for an inode.
2904  */
2905 static int osd_object_version_get(const struct lu_env *env,
2906                                   struct dt_object *dt, dt_obj_version_t *ver)
2907 {
2908         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2909
2910         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2911                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2912         *ver = LDISKFS_I(inode)->i_fs_version;
2913         return 0;
2914 }
2915
2916 /*
2917  * Concurrency: @dt is read locked.
2918  */
2919 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2920                          struct lu_buf *buf, const char *name,
2921                          struct lustre_capa *capa)
2922 {
2923         struct osd_object      *obj    = osd_dt_obj(dt);
2924         struct inode           *inode  = obj->oo_inode;
2925         struct osd_thread_info *info   = osd_oti_get(env);
2926         struct dentry          *dentry = &info->oti_obj_dentry;
2927
2928         /* version get is not real XATTR but uses xattr API */
2929         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2930                 /* for version we are just using xattr API but change inode
2931                  * field instead */
2932                 if (buf->lb_len == 0)
2933                         return sizeof(dt_obj_version_t);
2934
2935                 if (buf->lb_len < sizeof(dt_obj_version_t))
2936                         return -ERANGE;
2937
2938                 osd_object_version_get(env, dt, buf->lb_buf);
2939
2940                 return sizeof(dt_obj_version_t);
2941         }
2942
2943         LASSERT(dt_object_exists(dt));
2944         LASSERT(!dt_object_remote(dt));
2945         LASSERT(inode->i_op != NULL);
2946         LASSERT(inode->i_op->getxattr != NULL);
2947
2948         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2949                 return -EACCES;
2950
2951         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2952 }
2953
2954
2955 static int osd_declare_xattr_set(const struct lu_env *env,
2956                                  struct dt_object *dt,
2957                                  const struct lu_buf *buf, const char *name,
2958                                  int fl, struct thandle *handle)
2959 {
2960         struct osd_thandle *oh;
2961         int credits;
2962         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2963
2964         LASSERT(handle != NULL);
2965
2966         oh = container_of0(handle, struct osd_thandle, ot_super);
2967         LASSERT(oh->ot_handle == NULL);
2968
2969         /* optimistic optimization: LMA is set first and usually fit inode */
2970         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2971                 if (dt_object_exists(dt))
2972                         credits = 0;
2973                 else
2974                         credits = 1;
2975         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2976                 credits = 1;
2977         } else {
2978                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2979                 if (buf && buf->lb_len > sb->s_blocksize) {
2980                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2981                                         sb->s_blocksize_bits;
2982                 }
2983                 /*
2984                  * xattr set may involve inode quota change, reserve credits for
2985                  * dquot_initialize()
2986                  */
2987                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2988         }
2989
2990         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2991
2992         return 0;
2993 }
2994
2995 /*
2996  * Set the 64-bit version for object
2997  */
2998 static void osd_object_version_set(const struct lu_env *env,
2999                                    struct dt_object *dt,
3000                                    dt_obj_version_t *new_version)
3001 {
3002         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3003
3004         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3005                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3006
3007         LDISKFS_I(inode)->i_fs_version = *new_version;
3008         /** Version is set after all inode operations are finished,
3009          *  so we should mark it dirty here */
3010         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3011 }
3012
3013 /*
3014  * Concurrency: @dt is write locked.
3015  */
3016 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3017                          const struct lu_buf *buf, const char *name, int fl,
3018                          struct thandle *handle, struct lustre_capa *capa)
3019 {
3020         struct osd_object      *obj      = osd_dt_obj(dt);
3021         struct inode           *inode    = obj->oo_inode;
3022         struct osd_thread_info *info     = osd_oti_get(env);
3023         int                     fs_flags = 0;
3024         ENTRY;
3025
3026         LASSERT(handle != NULL);
3027
3028         /* version set is not real XATTR */
3029         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3030                 /* for version we are just using xattr API but change inode
3031                  * field instead */
3032                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3033                 osd_object_version_set(env, dt, buf->lb_buf);
3034                 return sizeof(dt_obj_version_t);
3035         }
3036
3037         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3038                 return -EACCES;
3039
3040         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3041                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3042
3043         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3044         if (fl & LU_XATTR_REPLACE)
3045                 fs_flags |= XATTR_REPLACE;
3046
3047         if (fl & LU_XATTR_CREATE)
3048                 fs_flags |= XATTR_CREATE;
3049
3050         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3051                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3052                 int                      rc;
3053
3054                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3055                 if (rc != 0)
3056                         RETURN(rc);
3057
3058                 lma->lma_incompat |= LMAI_STRIPED;
3059                 lustre_lma_swab(lma);
3060                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3061                                      sizeof(*lma), XATTR_REPLACE);
3062                 if (rc != 0)
3063                         RETURN(rc);
3064         }
3065
3066         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3067             strcmp(name, XATTR_NAME_LINK) == 0)
3068                 return -ENOSPC;
3069
3070         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3071                                fs_flags);
3072 }
3073
3074 /*
3075  * Concurrency: @dt is read locked.
3076  */
3077 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3078                           struct lu_buf *buf, struct lustre_capa *capa)
3079 {
3080         struct osd_object      *obj    = osd_dt_obj(dt);
3081         struct inode           *inode  = obj->oo_inode;
3082         struct osd_thread_info *info   = osd_oti_get(env);
3083         struct dentry          *dentry = &info->oti_obj_dentry;
3084
3085         LASSERT(dt_object_exists(dt));
3086         LASSERT(!dt_object_remote(dt));
3087         LASSERT(inode->i_op != NULL);
3088         LASSERT(inode->i_op->listxattr != NULL);
3089
3090         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3091                 return -EACCES;
3092
3093         dentry->d_inode = inode;
3094         dentry->d_sb = inode->i_sb;
3095         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3096 }
3097
3098 static int osd_declare_xattr_del(const struct lu_env *env,
3099                                  struct dt_object *dt, const char *name,
3100                                  struct thandle *handle)
3101 {
3102         struct osd_thandle *oh;
3103         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3104
3105         LASSERT(dt_object_exists(dt));
3106         LASSERT(!dt_object_remote(dt));
3107         LASSERT(handle != NULL);
3108
3109         oh = container_of0(handle, struct osd_thandle, ot_super);
3110         LASSERT(oh->ot_handle == NULL);
3111
3112         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3113                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3114         /*
3115          * xattr del may involve inode quota change, reserve credits for
3116          * dquot_initialize()
3117          */
3118         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3119
3120         return 0;
3121 }
3122
3123 /*
3124  * Concurrency: @dt is write locked.
3125  */
3126 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3127                          const char *name, struct thandle *handle,
3128                          struct lustre_capa *capa)
3129 {
3130         struct osd_object      *obj    = osd_dt_obj(dt);
3131         struct inode           *inode  = obj->oo_inode;
3132         struct osd_thread_info *info   = osd_oti_get(env);
3133         struct dentry          *dentry = &info->oti_obj_dentry;
3134         int                     rc;
3135
3136         LASSERT(dt_object_exists(dt));
3137         LASSERT(!dt_object_remote(dt));
3138         LASSERT(inode->i_op != NULL);
3139         LASSERT(inode->i_op->removexattr != NULL);
3140         LASSERT(handle != NULL);
3141
3142         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3143                 return -EACCES;
3144
3145         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3146
3147         ll_vfs_dq_init(inode);
3148         dentry->d_inode = inode;
3149         dentry->d_sb = inode->i_sb;
3150         rc = inode->i_op->removexattr(dentry, name);
3151         return rc;
3152 }
3153
3154 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3155                                      struct dt_object *dt,
3156                                      struct lustre_capa *old, __u64 opc)
3157 {
3158         struct osd_thread_info *info = osd_oti_get(env);
3159         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3160         struct osd_object *obj = osd_dt_obj(dt);
3161         struct osd_device *osd = osd_obj2dev(obj);
3162         struct lustre_capa_key *key = &info->oti_capa_key;
3163         struct lustre_capa *capa = &info->oti_capa;
3164         struct obd_capa *oc;
3165         struct lu_capainfo *lci;
3166         int rc;
3167         ENTRY;
3168
3169         if (!osd->od_fl_capa)
3170                 RETURN(ERR_PTR(-ENOENT));
3171
3172         LASSERT(dt_object_exists(dt));
3173         LASSERT(!dt_object_remote(dt));
3174         LINVRNT(osd_invariant(obj));
3175
3176         /* renewal sanity check */
3177         if (old && osd_object_auth(env, dt, old, opc))
3178                 RETURN(ERR_PTR(-EACCES));
3179
3180         lci = lu_capainfo_get(env);
3181         if (unlikely(lci == NULL))
3182                 RETURN(ERR_PTR(-ENOENT));
3183
3184         switch (lci->lci_auth) {
3185         case LC_ID_NONE:
3186                 RETURN(NULL);
3187         case LC_ID_PLAIN:
3188                 capa->lc_uid = i_uid_read(obj->oo_inode);
3189                 capa->lc_gid = i_gid_read(obj->oo_inode);
3190                 capa->lc_flags = LC_ID_PLAIN;
3191                 break;
3192         case LC_ID_CONVERT: {
3193                 __u32 d[4], s[4];
3194
3195                 s[0] = i_uid_read(obj->oo_inode);
3196                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3197                 s[2] = i_uid_read(obj->oo_inode);
3198                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3199                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3200                 if (unlikely(rc))
3201                         RETURN(ERR_PTR(rc));
3202
3203                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3204                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3205                 capa->lc_flags = LC_ID_CONVERT;
3206                 break;
3207         }
3208         default:
3209                 RETURN(ERR_PTR(-EINVAL));
3210         }
3211
3212         capa->lc_fid = *fid;
3213         capa->lc_opc = opc;
3214         capa->lc_flags |= osd->od_capa_alg << 24;
3215         capa->lc_timeout = osd->od_capa_timeout;
3216         capa->lc_expiry = 0;
3217
3218         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3219         if (oc) {
3220                 LASSERT(!capa_is_expired(oc));
3221                 RETURN(oc);
3222         }
3223
3224         spin_lock(&capa_lock);
3225         *key = osd->od_capa_keys[1];
3226         spin_unlock(&capa_lock);
3227
3228         capa->lc_keyid = key->lk_keyid;
3229         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3230
3231         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3232         if (rc) {
3233                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3234                 RETURN(ERR_PTR(rc));
3235         }
3236
3237         oc = capa_add(osd->od_capa_hash, capa);
3238         RETURN(oc);
3239 }
3240
3241 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3242                            __u64 start, __u64 end)
3243 {
3244         struct osd_object       *obj    = osd_dt_obj(dt);
3245         struct inode            *inode  = obj->oo_inode;
3246         struct osd_thread_info  *info   = osd_oti_get(env);
3247         struct dentry           *dentry = &info->oti_obj_dentry;
3248         struct file             *file   = &info->oti_file;
3249         int                     rc;
3250
3251         ENTRY;
3252
3253         dentry->d_inode = inode;
3254         dentry->d_sb = inode->i_sb;
3255         file->f_dentry = dentry;
3256         file->f_mapping = inode->i_mapping;
3257         file->f_op = inode->i_fop;
3258         set_file_inode(file, inode);
3259
3260 #ifdef HAVE_FILE_FSYNC_4ARGS
3261         rc = file->f_op->fsync(file, start, end, 0);
3262 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3263         mutex_lock(&inode->i_mutex);
3264         rc = file->f_op->fsync(file, 0);
3265         mutex_unlock(&inode->i_mutex);
3266 #else
3267         mutex_lock(&inode->i_mutex);
3268         rc = file->f_op->fsync(file, dentry, 0);
3269         mutex_unlock(&inode->i_mutex);
3270 #endif
3271
3272         RETURN(rc);
3273 }
3274
3275 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3276                         void **data)
3277 {
3278         struct osd_object *obj = osd_dt_obj(dt);
3279         ENTRY;
3280
3281         *data = (void *)obj->oo_inode;
3282         RETURN(0);
3283 }
3284
3285 /*
3286  * Index operations.
3287  */
3288
3289 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3290                            const struct dt_index_features *feat)
3291 {
3292         struct iam_descr *descr;
3293
3294         if (osd_object_is_root(o))
3295                 return feat == &dt_directory_features;
3296
3297         LASSERT(o->oo_dir != NULL);
3298
3299         descr = o->oo_dir->od_container.ic_descr;
3300         if (feat == &dt_directory_features) {
3301                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3302                         return 1;
3303                 else
3304                         return 0;
3305         } else {
3306                 return
3307                         feat->dif_keysize_min <= descr->id_key_size &&
3308                         descr->id_key_size <= feat->dif_keysize_max &&
3309                         feat->dif_recsize_min <= descr->id_rec_size &&
3310                         descr->id_rec_size <= feat->dif_recsize_max &&
3311                         !(feat->dif_flags & (DT_IND_VARKEY |
3312                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3313                         ergo(feat->dif_flags & DT_IND_UPDATE,
3314                              1 /* XXX check that object (and file system) is
3315                                 * writable */);
3316         }
3317 }
3318
3319 static int osd_iam_container_init(const struct lu_env *env,
3320                                   struct osd_object *obj,
3321                                   struct osd_directory *dir)
3322 {
3323         struct iam_container *bag = &dir->od_container;
3324         int result;
3325
3326         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3327         if (result != 0)
3328                 return result;
3329
3330         result = iam_container_setup(bag);
3331         if (result == 0)
3332                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3333         else
3334                 iam_container_fini(bag);
3335
3336         return result;
3337 }
3338
3339
3340 /*
3341  * Concurrency: no external locking is necessary.
3342  */
3343 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3344                          const struct dt_index_features *feat)
3345 {
3346         int                      result;
3347         int                      skip_iam = 0;
3348         struct osd_object       *obj = osd_dt_obj(dt);
3349
3350         LINVRNT(osd_invariant(obj));
3351
3352         if (osd_object_is_root(obj)) {
3353                 dt->do_index_ops = &osd_index_ea_ops;
3354                 result = 0;
3355         } else if (feat == &dt_directory_features) {
3356                 dt->do_index_ops = &osd_index_ea_ops;
3357                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3358                         result = 0;
3359                 else
3360                         result = -ENOTDIR;
3361                 skip_iam = 1;
3362         } else if (unlikely(feat == &dt_otable_features)) {
3363                 dt->do_index_ops = &osd_otable_ops;
3364                 return 0;
3365         } else if (unlikely(feat == &dt_acct_features)) {
3366                 dt->do_index_ops = &osd_acct_index_ops;
3367                 result = 0;
3368                 skip_iam = 1;
3369         } else if (!osd_has_index(obj)) {
3370                 struct osd_directory *dir;
3371
3372                 OBD_ALLOC_PTR(dir);
3373                 if (dir != NULL) {
3374
3375                         spin_lock(&obj->oo_guard);
3376                         if (obj->oo_dir == NULL)
3377                                 obj->oo_dir = dir;
3378                         else
3379                                 /*
3380                                  * Concurrent thread allocated container data.
3381                                  */
3382                                 OBD_FREE_PTR(dir);
3383                         spin_unlock(&obj->oo_guard);
3384                         /*
3385                          * Now, that we have container data, serialize its
3386                          * initialization.
3387                          */
3388                         down_write(&obj->oo_ext_idx_sem);
3389                         /*
3390                          * recheck under lock.
3391                          */
3392                         if (!osd_has_index(obj))
3393                                 result = osd_iam_container_init(env, obj,
3394                                                                 obj->oo_dir);
3395                         else
3396                                 result = 0;
3397                         up_write(&obj->oo_ext_idx_sem);
3398                 } else {
3399                         result = -ENOMEM;
3400                 }
3401         } else {
3402                 result = 0;
3403         }
3404
3405         if (result == 0 && skip_iam == 0) {
3406                 if (!osd_iam_index_probe(env, obj, feat))
3407                         result = -ENOTDIR;
3408         }
3409         LINVRNT(osd_invariant(obj));
3410
3411         if (result == 0 && is_quota_glb_feat(feat) &&
3412             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3413                 result = osd_quota_migration(env, dt, feat);
3414
3415         return result;
3416 }
3417
3418 static int osd_otable_it_attr_get(const struct lu_env *env,
3419                                  struct dt_object *dt,
3420                                  struct lu_attr *attr,
3421                                  struct lustre_capa *capa)
3422 {
3423         attr->la_valid = 0;
3424         return 0;
3425 }
3426
3427 static const struct dt_object_operations osd_obj_ops = {
3428         .do_read_lock         = osd_object_read_lock,
3429         .do_write_lock        = osd_object_write_lock,
3430         .do_read_unlock       = osd_object_read_unlock,
3431         .do_write_unlock      = osd_object_write_unlock,
3432         .do_write_locked      = osd_object_write_locked,
3433         .do_attr_get          = osd_attr_get,
3434         .do_declare_attr_set  = osd_declare_attr_set,
3435         .do_attr_set          = osd_attr_set,
3436         .do_ah_init           = osd_ah_init,
3437         .do_declare_create    = osd_declare_object_create,
3438         .do_create            = osd_object_create,
3439         .do_declare_destroy   = osd_declare_object_destroy,
3440         .do_destroy           = osd_object_destroy,
3441         .do_index_try         = osd_index_try,
3442         .do_declare_ref_add   = osd_declare_object_ref_add,
3443         .do_ref_add           = osd_object_ref_add,
3444         .do_declare_ref_del   = osd_declare_object_ref_del,
3445         .do_ref_del           = osd_object_ref_del,
3446         .do_xattr_get         = osd_xattr_get,
3447         .do_declare_xattr_set = osd_declare_xattr_set,
3448         .do_xattr_set         = osd_xattr_set,
3449         .do_declare_xattr_del = osd_declare_xattr_del,
3450         .do_xattr_del         = osd_xattr_del,
3451         .do_xattr_list        = osd_xattr_list,
3452         .do_capa_get          = osd_capa_get,
3453         .do_object_sync       = osd_object_sync,
3454         .do_data_get          = osd_data_get,
3455 };
3456
3457 /**
3458  * dt_object_operations for interoperability mode
3459  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3460  */
3461 static const struct dt_object_operations osd_obj_ea_ops = {
3462         .do_read_lock         = osd_object_read_lock,
3463         .do_write_lock        = osd_object_write_lock,
3464         .do_read_unlock       = osd_object_read_unlock,
3465         .do_write_unlock      = osd_object_write_unlock,
3466         .do_write_locked      = osd_object_write_locked,
3467         .do_attr_get          = osd_attr_get,
3468         .do_declare_attr_set  = osd_declare_attr_set,
3469         .do_attr_set          = osd_attr_set,
3470         .do_ah_init           = osd_ah_init,
3471         .do_declare_create    = osd_declare_object_create,
3472         .do_create            = osd_object_ea_create,
3473         .do_declare_destroy   = osd_declare_object_destroy,
3474         .do_destroy           = osd_object_destroy,
3475         .do_index_try         = osd_index_try,
3476         .do_declare_ref_add   = osd_declare_object_ref_add,
3477         .do_ref_add           = osd_object_ref_add,
3478         .do_declare_ref_del   = osd_declare_object_ref_del,
3479         .do_ref_del           = osd_object_ref_del,
3480         .do_xattr_get         = osd_xattr_get,
3481         .do_declare_xattr_set = osd_declare_xattr_set,
3482         .do_xattr_set         = osd_xattr_set,
3483         .do_declare_xattr_del = osd_declare_xattr_del,
3484         .do_xattr_del         = osd_xattr_del,
3485         .do_xattr_list        = osd_xattr_list,
3486         .do_capa_get          = osd_capa_get,
3487         .do_object_sync       = osd_object_sync,
3488         .do_data_get          = osd_data_get,
3489 };
3490
3491 static const struct dt_object_operations osd_obj_otable_it_ops = {
3492         .do_attr_get    = osd_otable_it_attr_get,
3493         .do_index_try   = osd_index_try,
3494 };
3495
3496 static int osd_index_declare_iam_delete(const struct lu_env *env,
3497                                         struct dt_object *dt,
3498                                         const struct dt_key *key,
3499                                         struct thandle *handle)
3500 {
3501         struct osd_thandle    *oh;
3502
3503         oh = container_of0(handle, struct osd_thandle, ot_super);
3504         LASSERT(oh->ot_handle == NULL);
3505
3506         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3507                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3508
3509         return 0;
3510 }
3511
3512 /**
3513  *      delete a (key, value) pair from index \a dt specified by \a key
3514  *
3515  *      \param  dt      osd index object
3516  *      \param  key     key for index
3517  *      \param  rec     record reference
3518  *      \param  handle  transaction handler
3519  *
3520  *      \retval  0  success
3521  *      \retval -ve   failure
3522  */
3523 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3524                                 const struct dt_key *key,
3525                                 struct thandle *handle,
3526                                 struct lustre_capa *capa)
3527 {
3528         struct osd_thread_info *oti = osd_oti_get(env);
3529         struct osd_object      *obj = osd_dt_obj(dt);
3530         struct osd_thandle     *oh;
3531         struct iam_path_descr  *ipd;
3532         struct iam_container   *bag = &obj->oo_dir->od_container;
3533         int                     rc;
3534         ENTRY;
3535
3536         LINVRNT(osd_invariant(obj));
3537         LASSERT(dt_object_exists(dt));
3538         LASSERT(!dt_object_remote(dt));
3539         LASSERT(bag->ic_object == obj->oo_inode);
3540         LASSERT(handle != NULL);
3541
3542         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3543                 RETURN(-EACCES);
3544
3545         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3546
3547         ipd = osd_idx_ipd_get(env, bag);
3548         if (unlikely(ipd == NULL))
3549                 RETURN(-ENOMEM);
3550
3551         oh = container_of0(handle, struct osd_thandle, ot_super);
3552         LASSERT(oh->ot_handle != NULL);
3553         LASSERT(oh->ot_handle->h_transaction != NULL);
3554
3555         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3556                 /* swab quota uid/gid provided by caller */
3557                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3558                 key = (const struct dt_key *)&oti->oti_quota_id;
3559         }
3560
3561         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3562         osd_ipd_put(env, bag, ipd);
3563         LINVRNT(osd_invariant(obj));
3564         RETURN(rc);
3565 }
3566
3567 static int osd_index_declare_ea_delete(const struct lu_env *env,
3568                                        struct dt_object *dt,
3569                                        const struct dt_key *key,
3570                                        struct thandle *handle)
3571 {
3572         struct osd_thandle *oh;
3573         struct inode       *inode;
3574         int                 rc;
3575         ENTRY;
3576
3577         LASSERT(dt_object_exists(dt));
3578         LASSERT(!dt_object_remote(dt));
3579         LASSERT(handle != NULL);
3580
3581         oh = container_of0(handle, struct osd_thandle, ot_super);
3582         LASSERT(oh->ot_handle == NULL);
3583
3584         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3585                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3586
3587         inode = osd_dt_obj(dt)->oo_inode;
3588         LASSERT(inode);
3589
3590         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3591                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3592         RETURN(rc);
3593 }
3594
3595 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3596                                           struct dt_rec *fid)
3597 {
3598         struct osd_fid_pack *rec;
3599         int                  rc = -ENODATA;
3600
3601         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3602                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3603                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3604         }
3605         return rc;
3606 }
3607
3608 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3609                           const struct lu_fid *fid)
3610 {
3611         struct seq_server_site  *ss = osd_seq_site(osd);
3612         ENTRY;
3613
3614         /* FID seqs not in FLDB, must be local seq */
3615         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3616                 RETURN(0);
3617
3618         /* If FLD is not being initialized yet, it only happens during the
3619          * initialization, likely during mgs initialization, and we assume
3620          * this is local FID. */
3621         if (ss == NULL || ss->ss_server_fld == NULL)
3622                 RETURN(0);
3623
3624         /* Only check the local FLDB here */
3625         if (osd_seq_exists(env, osd, fid_seq(fid)))
3626                 RETURN(0);
3627
3628         RETURN(1);
3629 }
3630
3631 /**
3632  * Index delete function for interoperability mode (b11826).
3633  * It will remove the directory entry added by osd_index_ea_insert().
3634  * This entry is needed to maintain name->fid mapping.
3635  *
3636  * \param key,  key i.e. file entry to be deleted
3637  *
3638  * \retval   0, on success
3639  * \retval -ve, on error
3640  */
3641 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3642                                const struct dt_key *key, struct thandle *handle,
3643                                struct lustre_capa *capa)
3644 {
3645         struct osd_object          *obj = osd_dt_obj(dt);
3646         struct inode               *dir = obj->oo_inode;
3647         struct dentry              *dentry;
3648         struct osd_thandle         *oh;
3649         struct ldiskfs_dir_entry_2 *de = NULL;
3650         struct buffer_head         *bh;
3651         struct htree_lock          *hlock = NULL;
3652         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3653         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3654         int                        rc;
3655         ENTRY;
3656
3657         LINVRNT(osd_invariant(obj));
3658         LASSERT(dt_object_exists(dt));
3659         LASSERT(!dt_object_remote(dt));
3660         LASSERT(handle != NULL);
3661
3662         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3663
3664         oh = container_of(handle, struct osd_thandle, ot_super);
3665         LASSERT(oh->ot_handle != NULL);
3666         LASSERT(oh->ot_handle->h_transaction != NULL);
3667
3668         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3669                 RETURN(-EACCES);
3670
3671         ll_vfs_dq_init(dir);
3672         dentry = osd_child_dentry_get(env, obj,
3673                                       (char *)key, strlen((char *)key));
3674
3675         if (obj->oo_hl_head != NULL) {
3676                 hlock = osd_oti_get(env)->oti_hlock;
3677                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3678                                    dir, LDISKFS_HLOCK_DEL);
3679         } else {
3680                 down_write(&obj->oo_ext_idx_sem);
3681         }
3682
3683         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3684         if (bh) {
3685                 __u32 ino = 0;
3686
3687                 /* If this is not the ".." entry, it might be a remote DNE
3688                  * entry and  we need to check if the FID is for a remote
3689                  * MDT.  If the FID is  not in the directory entry (e.g.
3690                  * upgraded 1.8 filesystem without dirdata enabled) then
3691                  * we need to get the FID from the LMA. For a remote directory
3692                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3693                  * case.
3694                  *
3695                  * Delete the entry before the agent inode in order to
3696                  * simplify error handling.  At worst an error after deleting
3697                  * the entry first might leak the agent inode afterward. The
3698                  * reverse would need filesystem abort in case of error deleting
3699                  * the entry after the agent had been removed, or leave a
3700                  * dangling entry pointing at a random inode. */
3701                 if (strcmp((char *)key, dotdot) != 0) {
3702                         LASSERT(de != NULL);
3703                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3704                         /* If Fid is not in dentry, try to get it from LMA */
3705                         if (rc == -ENODATA) {
3706                                 struct osd_inode_id *id;
3707                                 struct inode *inode;
3708
3709                                 /* Before trying to get fid from the inode,
3710                                  * check whether the inode is valid.
3711                                  *
3712                                  * If the inode has been deleted, do not go
3713                                  * ahead to do osd_ea_fid_get, which will set
3714                                  * the inode to bad inode, which might cause
3715                                  * the inode to be deleted uncorrectly */
3716                                 inode = ldiskfs_iget(osd_sb(osd),
3717                                                      le32_to_cpu(de->inode));
3718                                 if (IS_ERR(inode)) {
3719                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3720                                                "error.\n", osd_name(osd),
3721                                                PFID(fid));
3722                                         rc = PTR_ERR(inode);
3723                                 } else {
3724                                         if (likely(inode->i_nlink != 0)) {
3725                                                 id = &osd_oti_get(env)->oti_id;
3726                                                 rc = osd_ea_fid_get(env, obj,
3727                                                         le32_to_cpu(de->inode),
3728                                                                     fid, id);
3729                                         } else {
3730                                                 CDEBUG(D_INFO, "%s: %u "DFID
3731                                                        "deleted.\n",
3732                                                        osd_name(osd),
3733                                                        le32_to_cpu(de->inode),
3734                                                        PFID(fid));
3735                                                 rc = -ESTALE;
3736                                         }
3737                                         iput(inode);
3738                                 }
3739                         }
3740                         if (rc == 0 &&
3741                             unlikely(osd_remote_fid(env, osd, fid)))
3742                                 /* Need to delete agent inode */
3743                                 ino = le32_to_cpu(de->inode);
3744                 }
3745                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3746                 brelse(bh);
3747                 if (rc == 0 && unlikely(ino != 0)) {
3748                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3749                                                           oh);
3750                         if (rc != 0)
3751                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3752                                        osd_name(osd), PFID(fid), rc);
3753                 }
3754         } else {
3755                 rc = -ENOENT;
3756         }
3757         if (hlock != NULL)
3758                 ldiskfs_htree_unlock(hlock);
3759         else
3760                 up_write(&obj->oo_ext_idx_sem);
3761
3762         if (rc != 0)
3763                 GOTO(out, rc);
3764
3765         /* For inode on the remote MDT, .. will point to
3766          * /Agent directory, Check whether it needs to delete
3767          * from agent directory */
3768         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3769                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3770                                                    oh);
3771                 if (rc != 0 && rc != -ENOENT) {
3772                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3773                                osd_name(osd), PFID(fid), rc);
3774                 }
3775
3776                 if (rc == -ENOENT)
3777                         rc = 0;
3778
3779                 GOTO(out, rc);
3780         }
3781 out:
3782
3783         LASSERT(osd_invariant(obj));
3784         RETURN(rc);
3785 }
3786
3787 /**
3788  *      Lookup index for \a key and copy record to \a rec.
3789  *
3790  *      \param  dt      osd index object
3791  *      \param  key     key for index
3792  *      \param  rec     record reference
3793  *
3794  *      \retval  +ve  success : exact mach
3795  *      \retval  0    return record with key not greater than \a key
3796  *      \retval -ve   failure
3797  */
3798 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3799                                 struct dt_rec *rec, const struct dt_key *key,
3800                                 struct lustre_capa *capa)
3801 {
3802         struct osd_object      *obj = osd_dt_obj(dt);
3803         struct iam_path_descr  *ipd;
3804         struct iam_container   *bag = &obj->oo_dir->od_container;
3805         struct osd_thread_info *oti = osd_oti_get(env);
3806         struct iam_iterator    *it = &oti->oti_idx_it;
3807         struct iam_rec         *iam_rec;
3808         int                     rc;
3809         ENTRY;
3810
3811         LASSERT(osd_invariant(obj));
3812         LASSERT(dt_object_exists(dt));
3813         LASSERT(!dt_object_remote(dt));
3814         LASSERT(bag->ic_object == obj->oo_inode);
3815
3816         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3817                 RETURN(-EACCES);
3818
3819         ipd = osd_idx_ipd_get(env, bag);
3820         if (IS_ERR(ipd))
3821                 RETURN(-ENOMEM);
3822
3823         /* got ipd now we can start iterator. */
3824         iam_it_init(it, bag, 0, ipd);
3825
3826         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3827                 /* swab quota uid/gid provided by caller */
3828                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3829                 key = (const struct dt_key *)&oti->oti_quota_id;
3830         }
3831
3832         rc = iam_it_get(it, (struct iam_key *)key);
3833         if (rc >= 0) {
3834                 if (S_ISDIR(obj->oo_inode->i_mode))
3835                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3836                 else
3837                         iam_rec = (struct iam_rec *) rec;
3838
3839                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3840
3841                 if (S_ISDIR(obj->oo_inode->i_mode))
3842                         osd_fid_unpack((struct lu_fid *) rec,
3843                                        (struct osd_fid_pack *)iam_rec);
3844                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3845                         osd_quota_unpack(obj, rec);
3846         }
3847
3848         iam_it_put(it);
3849         iam_it_fini(it);
3850         osd_ipd_put(env, bag, ipd);
3851
3852         LINVRNT(osd_invariant(obj));
3853
3854         RETURN(rc);
3855 }
3856
3857 static int osd_index_declare_iam_insert(const struct lu_env *env,
3858                                         struct dt_object *dt,
3859                                         const struct dt_rec *rec,
3860                                         const struct dt_key *key,
3861                                         struct thandle *handle)
3862 {
3863         struct osd_thandle *oh;
3864
3865         LASSERT(handle != NULL);
3866
3867         oh = container_of0(handle, struct osd_thandle, ot_super);
3868         LASSERT(oh->ot_handle == NULL);
3869
3870         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3871                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3872
3873         return 0;
3874 }
3875
3876 /**
3877  *      Inserts (key, value) pair in \a dt index object.
3878  *
3879  *      \param  dt      osd index object
3880  *      \param  key     key for index
3881  *      \param  rec     record reference
3882  *      \param  th      transaction handler
3883  *
3884  *      \retval  0  success
3885  *      \retval -ve failure
3886  */
3887 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3888                                 const struct dt_rec *rec,
3889                                 const struct dt_key *key, struct thandle *th,
3890                                 struct lustre_capa *capa, int ignore_quota)
3891 {
3892         struct osd_object     *obj = osd_dt_obj(dt);
3893         struct iam_path_descr *ipd;
3894         struct osd_thandle    *oh;
3895         struct iam_container  *bag = &obj->oo_dir->od_container;
3896         struct osd_thread_info *oti = osd_oti_get(env);
3897         struct iam_rec         *iam_rec;
3898         int                     rc;
3899         ENTRY;
3900
3901         LINVRNT(osd_invariant(obj));
3902         LASSERT(dt_object_exists(dt));
3903         LASSERT(!dt_object_remote(dt));
3904         LASSERT(bag->ic_object == obj->oo_inode);
3905         LASSERT(th != NULL);
3906
3907         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3908                 RETURN(-EACCES);
3909
3910         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3911
3912         ipd = osd_idx_ipd_get(env, bag);
3913         if (unlikely(ipd == NULL))
3914                 RETURN(-ENOMEM);
3915
3916         oh = container_of0(th, struct osd_thandle, ot_super);
3917         LASSERT(oh->ot_handle != NULL);
3918         LASSERT(oh->ot_handle->h_transaction != NULL);
3919         if (S_ISDIR(obj->oo_inode->i_mode)) {
3920                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3921                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3922         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3923                 /* pack quota uid/gid */
3924                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3925                 key = (const struct dt_key *)&oti->oti_quota_id;
3926                 /* pack quota record */
3927                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3928                 iam_rec = (struct iam_rec *)rec;
3929         } else {
3930                 iam_rec = (struct iam_rec *)rec;
3931         }
3932
3933         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3934                         iam_rec, ipd);
3935         osd_ipd_put(env, bag, ipd);
3936         LINVRNT(osd_invariant(obj));
3937         RETURN(rc);
3938 }
3939
3940 /**
3941  * Calls ldiskfs_add_entry() to add directory entry
3942  * into the directory. This is required for
3943  * interoperability mode (b11826)
3944  *
3945  * \retval   0, on success
3946  * \retval -ve, on error
3947  */
3948 static int __osd_ea_add_rec(struct osd_thread_info *info,
3949                             struct osd_object *pobj, struct inode  *cinode,
3950                             const char *name, const struct lu_fid *fid,
3951                             struct htree_lock *hlock, struct thandle *th)
3952 {
3953         struct ldiskfs_dentry_param *ldp;
3954         struct dentry               *child;
3955         struct osd_thandle          *oth;
3956         int                          rc;
3957
3958         oth = container_of(th, struct osd_thandle, ot_super);
3959         LASSERT(oth->ot_handle != NULL);
3960         LASSERT(oth->ot_handle->h_transaction != NULL);
3961         LASSERT(pobj->oo_inode);
3962
3963         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3964         if (unlikely(pobj->oo_inode ==
3965                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3966                 ldp->edp_magic = 0;
3967         else
3968                 osd_get_ldiskfs_dirent_param(ldp, fid);
3969         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3970         child->d_fsdata = (void *)ldp;
3971         ll_vfs_dq_init(pobj->oo_inode);
3972         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3973         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3974                 struct ldiskfs_dir_entry_2      *de;
3975                 struct buffer_head              *bh;
3976                 int                              rc1;
3977
3978                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3979                                             NULL, hlock);
3980                 if (bh != NULL) {
3981                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3982                                                                bh);
3983                         if (rc1 == 0) {
3984                                 if (S_ISDIR(cinode->i_mode))
3985                                         de->file_type = LDISKFS_DIRENT_LUFID |
3986                                                         LDISKFS_FT_REG_FILE;
3987                                 else
3988                                         de->file_type = LDISKFS_DIRENT_LUFID |
3989                                                         LDISKFS_FT_DIR;
3990                                 ldiskfs_journal_dirty_metadata(oth->ot_handle,
3991                                                                bh);
3992                                 brelse(bh);
3993                         }
3994                 }
3995         }
3996
3997         RETURN(rc);
3998 }
3999
4000 /**
4001  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4002  * into the directory.Also sets flags into osd object to
4003  * indicate dot and dotdot are created. This is required for
4004  * interoperability mode (b11826)
4005  *
4006  * \param dir   directory for dot and dotdot fixup.
4007  * \param obj   child object for linking
4008  *
4009  * \retval   0, on success
4010  * \retval -ve, on error
4011  */
4012 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4013                               struct osd_object *dir,
4014                               struct inode *parent_dir, const char *name,
4015                               const struct lu_fid *dot_fid,
4016                               const struct lu_fid *dot_dot_fid,
4017                               struct thandle *th)
4018 {
4019         struct inode                *inode = dir->oo_inode;
4020         struct osd_thandle          *oth;
4021         int result = 0;
4022
4023         oth = container_of(th, struct osd_thandle, ot_super);
4024         LASSERT(oth->ot_handle->h_transaction != NULL);
4025         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4026
4027         if (strcmp(name, dot) == 0) {
4028                 if (dir->oo_compat_dot_created) {
4029                         result = -EEXIST;
4030                 } else {
4031                         LASSERT(inode == parent_dir);
4032                         dir->oo_compat_dot_created = 1;
4033                         result = 0;
4034                 }
4035         } else if (strcmp(name, dotdot) == 0) {
4036                 if (!dir->oo_compat_dot_created)
4037                         return -EINVAL;
4038                 /* in case of rename, dotdot is already created */
4039                 if (dir->oo_compat_dotdot_created) {
4040                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4041                                                 dot_dot_fid, NULL, th);
4042                 }
4043
4044                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4045                         struct lu_fid tfid = *dot_dot_fid;
4046
4047                         tfid.f_oid--;
4048                         result = osd_add_dot_dotdot_internal(info,
4049                                         dir->oo_inode, parent_dir, dot_fid,
4050                                         &tfid, oth);
4051                 } else {
4052                         result = osd_add_dot_dotdot_internal(info,
4053                                         dir->oo_inode, parent_dir, dot_fid,
4054                                         dot_dot_fid, oth);
4055                 }
4056
4057                 if (result == 0)
4058                         dir->oo_compat_dotdot_created = 1;
4059         }
4060
4061         return result;
4062 }
4063
4064
4065 /**
4066  * It will call the appropriate osd_add* function and return the
4067  * value, return by respective functions.
4068  */
4069 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4070                           struct inode *cinode, const char *name,
4071                           const struct lu_fid *fid, struct thandle *th)
4072 {
4073         struct osd_thread_info *info   = osd_oti_get(env);
4074         struct htree_lock      *hlock;
4075         int                     rc;
4076
4077         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4078
4079         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4080                                                    name[2] =='\0'))) {
4081                 if (hlock != NULL) {
4082                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4083                                            pobj->oo_inode, 0);
4084                 } else {
4085                         down_write(&pobj->oo_ext_idx_sem);
4086                 }
4087
4088                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4089                                         lu_object_fid(&pobj->oo_dt.do_lu),
4090                                         fid, th);
4091         } else {
4092                 if (hlock != NULL) {
4093                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4094                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4095                 } else {
4096                         down_write(&pobj->oo_ext_idx_sem);
4097                 }
4098
4099                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4100                         struct lu_fid *tfid = &info->oti_fid;
4101
4102                         *tfid = *fid;
4103                         tfid->f_ver = ~0;
4104                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4105                                               tfid, hlock, th);
4106                 } else {
4107                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4108                                               hlock, th);
4109                 }
4110         }
4111         if (hlock != NULL)
4112                 ldiskfs_htree_unlock(hlock);
4113         else
4114                 up_write(&pobj->oo_ext_idx_sem);
4115
4116         return rc;
4117 }
4118
4119 static void
4120 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4121                       struct osd_idmap_cache *oic)
4122 {
4123         struct osd_scrub    *scrub = &dev->od_scrub;
4124         struct lu_fid       *fid   = &oic->oic_fid;
4125         struct osd_inode_id *id    = &oti->oti_id;
4126         int                  once  = 0;
4127         int                  rc;
4128         ENTRY;
4129
4130         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4131                 RETURN_EXIT;
4132
4133 again:
4134         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4135         if (rc != 0 && rc != -ENOENT)
4136                 RETURN_EXIT;
4137
4138         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4139                 RETURN_EXIT;
4140
4141         if (thread_is_running(&scrub->os_thread)) {
4142                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4143                 /* There is race condition between osd_oi_lookup and OI scrub.
4144                  * The OI scrub finished just after osd_oi_lookup() failure.
4145                  * Under such case, it is unnecessary to trigger OI scrub again,
4146                  * but try to call osd_oi_lookup() again. */
4147                 if (unlikely(rc == -EAGAIN))
4148                         goto again;
4149
4150                 RETURN_EXIT;
4151         }
4152
4153         if (!dev->od_noscrub && ++once == 1) {
4154                 rc = osd_scrub_start(dev);
4155                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4156                               ", rc = %d [2]\n",
4157                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4158                               PFID(fid), rc);
4159                 if (rc == 0)
4160                         goto again;
4161         }
4162
4163         EXIT;
4164 }
4165
4166 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4167                                struct osd_device *dev,
4168                                struct osd_idmap_cache *oic,
4169                                struct lu_fid *fid, __u32 ino)
4170 {
4171         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4172         struct inode            *inode;
4173         int                      rc;
4174
4175         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4176         inode = osd_iget(oti, dev, &oic->oic_lid);
4177         if (IS_ERR(inode)) {
4178                 fid_zero(&oic->oic_fid);
4179                 return PTR_ERR(inode);
4180         }
4181
4182         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4183         iput(inode);
4184         if (rc != 0)
4185                 fid_zero(&oic->oic_fid);
4186         else
4187                 *fid = oic->oic_fid = lma->lma_self_fid;
4188         return rc;
4189 }
4190
4191 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4192                      struct osd_inode_id *id, const struct lu_fid *fid)
4193 {
4194         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4195                id->oii_ino, id->oii_gen, info);
4196         info->oti_cache.oic_lid = *id;
4197         info->oti_cache.oic_fid = *fid;
4198         info->oti_cache.oic_dev = osd;
4199
4200         return 0;
4201 }
4202
4203 /**
4204  * Get parent FID from the linkEA.
4205  *
4206  * For a directory which parent resides on remote MDT, to satisfy the
4207  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4208  * the other hand, to make the lookup(..) on the directory can return
4209  * the real parent FID, we append the real parent FID after its ".."
4210  * name entry in the /REMOTE_PARENT_DIR.
4211  *
4212  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4213  * backup. So after the restore, we cannot get the right parent FID from
4214  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4215  * we have stored the real parent FID in the directory object's linkEA,
4216  * we can parse the linkEA for the real parent FID.
4217  *
4218  * \param[in] env       pointer to the thread context
4219  * \param[in] obj       pointer to the object to be handled
4220  * \param[out]fid       pointer to the buffer to hold the parent FID
4221  *
4222  * \retval              0 for getting the real parent FID successfully
4223  * \retval              negative error number on failure
4224  */
4225 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4226                                     struct osd_object *obj,
4227                                     struct lu_fid *fid)
4228 {
4229         struct osd_thread_info  *oti    = osd_oti_get(env);
4230         struct lu_buf           *buf    = &oti->oti_big_buf;
4231         struct dentry           *dentry = &oti->oti_obj_dentry;
4232         struct inode            *inode  = obj->oo_inode;
4233         struct linkea_data       ldata  = { 0 };
4234         int                      rc;
4235         ENTRY;
4236
4237         fid_zero(fid);
4238         if (!S_ISDIR(inode->i_mode))
4239                 RETURN(-EIO);
4240
4241 again:
4242         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4243                              buf->lb_buf, buf->lb_len);
4244         if (rc == -ERANGE) {
4245                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4246                                      NULL, 0);
4247                 if (rc > 0) {
4248                         lu_buf_realloc(buf, rc);
4249                         if (buf->lb_buf == NULL)
4250                                 RETURN(-ENOMEM);
4251
4252                         goto again;
4253                 }
4254         }
4255
4256         if (unlikely(rc == 0))
4257                 RETURN(-ENODATA);
4258
4259         if (rc < 0)
4260                 RETURN(rc);
4261
4262         if (unlikely(buf->lb_buf == NULL)) {
4263                 lu_buf_realloc(buf, rc);
4264                 if (buf->lb_buf == NULL)
4265                         RETURN(-ENOMEM);
4266
4267                 goto again;
4268         }
4269
4270         ldata.ld_buf = buf;
4271         rc = linkea_init(&ldata);
4272         if (rc == 0) {
4273                 linkea_first_entry(&ldata);
4274                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4275         }
4276
4277         RETURN(rc);
4278 }
4279
4280 /**
4281  * Calls ->lookup() to find dentry. From dentry get inode and
4282  * read inode's ea to get fid. This is required for  interoperability
4283  * mode (b11826)
4284  *
4285  * \retval   0, on success
4286  * \retval -ve, on error
4287  */
4288 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4289                              struct dt_rec *rec, const struct dt_key *key)
4290 {
4291         struct inode                    *dir    = obj->oo_inode;
4292         struct dentry                   *dentry;
4293         struct ldiskfs_dir_entry_2      *de;
4294         struct buffer_head              *bh;
4295         struct lu_fid                   *fid = (struct lu_fid *) rec;
4296         struct htree_lock               *hlock = NULL;
4297         int                             ino;
4298         int                             rc;
4299         ENTRY;
4300
4301         LASSERT(dir->i_op != NULL);
4302         LASSERT(dir->i_op->lookup != NULL);
4303
4304         dentry = osd_child_dentry_get(env, obj,
4305                                       (char *)key, strlen((char *)key));
4306
4307         if (obj->oo_hl_head != NULL) {
4308                 hlock = osd_oti_get(env)->oti_hlock;
4309                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4310                                    dir, LDISKFS_HLOCK_LOOKUP);
4311         } else {
4312                 down_read(&obj->oo_ext_idx_sem);
4313         }
4314
4315         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4316         if (bh) {
4317                 struct osd_thread_info *oti = osd_oti_get(env);
4318                 struct osd_inode_id *id = &oti->oti_id;
4319                 struct osd_idmap_cache *oic = &oti->oti_cache;
4320                 struct osd_device *dev = osd_obj2dev(obj);
4321                 struct osd_scrub *scrub = &dev->od_scrub;
4322                 struct scrub_file *sf = &scrub->os_file;
4323
4324                 ino = le32_to_cpu(de->inode);
4325                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4326                         brelse(bh);
4327                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4328                         GOTO(out, rc);
4329                 }
4330
4331                 rc = osd_get_fid_from_dentry(de, rec);
4332
4333                 /* done with de, release bh */
4334                 brelse(bh);
4335                 if (rc != 0) {
4336                         if (unlikely(ino == osd_remote_parent_ino(dev)))
4337                                 /* If the parent is on remote MDT, and there
4338                                  * is no FID-in-dirent, then we have to get
4339                                  * the parent FID from the linkEA.  */
4340                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
4341                         else
4342                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4343                 } else {
4344                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4345                 }
4346
4347                 if (rc != 0) {
4348                         fid_zero(&oic->oic_fid);
4349                         GOTO(out, rc);
4350                 }
4351
4352                 if (osd_remote_fid(env, dev, fid))
4353                         GOTO(out, rc = 0);
4354
4355                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4356                                       fid);
4357                 if (rc != 0)
4358                         GOTO(out, rc);
4359                 if ((scrub->os_pos_current <= ino) &&
4360                     ((sf->sf_flags & SF_INCONSISTENT) ||
4361                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4362                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4363                                       sf->sf_oi_bitmap)))
4364                         osd_consistency_check(oti, dev, oic);
4365         } else {
4366                 rc = -ENOENT;
4367         }
4368
4369         GOTO(out, rc);
4370
4371 out:
4372         if (hlock != NULL)
4373                 ldiskfs_htree_unlock(hlock);
4374         else
4375                 up_read(&obj->oo_ext_idx_sem);
4376         return rc;
4377 }
4378
4379 /**
4380  * Find the osd object for given fid.
4381  *
4382  * \param fid need to find the osd object having this fid
4383  *
4384  * \retval osd_object on success
4385  * \retval        -ve on error
4386  */
4387 struct osd_object *osd_object_find(const struct lu_env *env,
4388                                    struct dt_object *dt,
4389                                    const struct lu_fid *fid)
4390 {
4391         struct lu_device  *ludev = dt->do_lu.lo_dev;
4392         struct osd_object *child = NULL;
4393         struct lu_object  *luch;
4394         struct lu_object  *lo;
4395
4396         /*
4397          * at this point topdev might not exist yet
4398          * (i.e. MGS is preparing profiles). so we can
4399          * not rely on topdev and instead lookup with
4400          * our device passed as topdev. this can't work
4401          * if the object isn't cached yet (as osd doesn't
4402          * allocate lu_header). IOW, the object must be
4403          * in the cache, otherwise lu_object_alloc() crashes
4404          * -bzzz
4405          */
4406         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4407                                  ludev : ludev->ld_site->ls_top_dev,
4408                                  fid, NULL);
4409         if (!IS_ERR(luch)) {
4410                 if (lu_object_exists(luch)) {
4411                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4412                         if (lo != NULL)
4413                                 child = osd_obj(lo);
4414                         else
4415                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4416                                                 "lu_object can't be located"
4417                                                 DFID"\n", PFID(fid));
4418
4419                         if (child == NULL) {
4420                                 lu_object_put(env, luch);
4421                                 CERROR("Unable to get osd_object\n");
4422                                 child = ERR_PTR(-ENOENT);
4423                         }
4424                 } else {
4425                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4426                                         "lu_object does not exists "DFID"\n",
4427                                         PFID(fid));
4428                         lu_object_put(env, luch);
4429                         child = ERR_PTR(-ENOENT);
4430                 }
4431         } else {
4432                 child = ERR_CAST(luch);
4433         }
4434
4435         return child;
4436 }
4437
4438 /**
4439  * Put the osd object once done with it.
4440  *
4441  * \param obj osd object that needs to be put
4442  */
4443 static inline void osd_object_put(const struct lu_env *env,
4444                                   struct osd_object *obj)
4445 {
4446         lu_object_put(env, &obj->oo_dt.do_lu);
4447 }
4448
4449 static int osd_index_declare_ea_insert(const struct lu_env *env,
4450                                        struct dt_object *dt,
4451                                        const struct dt_rec *rec,
4452                                        const struct dt_key *key,
4453                                        struct thandle *handle)
4454 {
4455         struct osd_thandle      *oh;
4456         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4457         struct lu_fid           *fid = (struct lu_fid *)rec;
4458         int                     rc;
4459         ENTRY;
4460
4461         LASSERT(!dt_object_remote(dt));
4462         LASSERT(handle != NULL);
4463
4464         oh = container_of0(handle, struct osd_thandle, ot_super);
4465         LASSERT(oh->ot_handle == NULL);
4466
4467         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4468                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4469
4470         if (osd_dt_obj(dt)->oo_inode != NULL) {
4471                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4472
4473                 /* We ignore block quota on meta pool (MDTs), so needn't
4474                  * calculate how many blocks will be consumed by this index
4475                  * insert */
4476                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4477                                            i_gid_read(inode), 0, oh,
4478                                            osd_dt_obj(dt), true, NULL, false);
4479         }
4480
4481         if (fid == NULL)
4482                 RETURN(0);
4483
4484         rc = osd_remote_fid(env, osd, fid);
4485         if (rc <= 0)
4486                 RETURN(rc);
4487
4488         rc = 0;
4489
4490         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4491                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4492         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4493                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4494         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4495                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4496
4497         RETURN(rc);
4498 }
4499
4500 /**
4501  * Index add function for interoperability mode (b11826).
4502  * It will add the directory entry.This entry is needed to
4503  * maintain name->fid mapping.
4504  *
4505  * \param key it is key i.e. file entry to be inserted
4506  * \param rec it is value of given key i.e. fid
4507  *
4508  * \retval   0, on success
4509  * \retval -ve, on error
4510  */
4511 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4512                                const struct dt_rec *rec,
4513                                const struct dt_key *key, struct thandle *th,
4514                                struct lustre_capa *capa, int ignore_quota)
4515 {
4516         struct osd_object       *obj = osd_dt_obj(dt);
4517         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4518         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4519         const struct lu_fid     *fid    = rec1->rec_fid;
4520         const char              *name = (const char *)key;
4521         struct osd_thread_info  *oti   = osd_oti_get(env);
4522         struct osd_inode_id     *id    = &oti->oti_id;
4523         struct inode            *child_inode = NULL;
4524         struct osd_object       *child = NULL;
4525         int                     rc;
4526         ENTRY;
4527
4528         LASSERT(osd_invariant(obj));
4529         LASSERT(dt_object_exists(dt));
4530         LASSERT(!dt_object_remote(dt));
4531         LASSERT(th != NULL);
4532
4533         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4534
4535         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4536                 RETURN(-EACCES);
4537
4538         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4539
4540         rc = osd_remote_fid(env, osd, fid);
4541         if (rc < 0) {
4542                 CERROR("%s: Can not find object "DFID" rc %d\n",
4543                        osd_name(osd), PFID(fid), rc);
4544                 RETURN(rc);
4545         }
4546
4547         if (rc == 1) {
4548                 /* Insert remote entry */
4549                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4550                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4551                         struct osd_thandle      *oh;
4552
4553                         /* If parent on remote MDT, we need put this object
4554                          * under AGENT */
4555                         oh = container_of(th, typeof(*oh), ot_super);
4556                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4557                         if (rc != 0) {
4558                                 CERROR("%s: add "DFID" error: rc = %d\n",
4559                                        osd_name(osd),
4560                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4561                                 RETURN(rc);
4562                         }
4563
4564                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4565                 } else {
4566                         child_inode = osd_create_local_agent_inode(env, osd,
4567                                         obj, fid, rec1->rec_type & S_IFMT, th);
4568                         if (IS_ERR(child_inode))
4569                                 RETURN(PTR_ERR(child_inode));
4570                 }
4571         } else {
4572                 /* Insert local entry */
4573                 child = osd_object_find(env, dt, fid);
4574                 if (IS_ERR(child)) {
4575                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4576                                osd_name(osd), PFID(fid),
4577                                id->oii_ino, id->oii_gen,
4578                                (int)PTR_ERR(child));
4579                         RETURN(PTR_ERR(child));
4580                 }
4581                 child_inode = igrab(child->oo_inode);
4582         }
4583
4584         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4585
4586         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4587                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4588
4589         iput(child_inode);
4590         if (child != NULL)
4591                 osd_object_put(env, child);
4592         LASSERT(osd_invariant(obj));
4593         RETURN(rc);
4594 }
4595
4596 /**
4597  *  Initialize osd Iterator for given osd index object.
4598  *
4599  *  \param  dt      osd index object
4600  */
4601
4602 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4603                                      struct dt_object *dt,
4604                                      __u32 unused,
4605                                      struct lustre_capa *capa)
4606 {
4607         struct osd_it_iam      *it;
4608         struct osd_thread_info *oti = osd_oti_get(env);
4609         struct osd_object      *obj = osd_dt_obj(dt);
4610         struct lu_object       *lo  = &dt->do_lu;
4611         struct iam_path_descr  *ipd;
4612         struct iam_container   *bag = &obj->oo_dir->od_container;
4613
4614         LASSERT(lu_object_exists(lo));
4615
4616         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4617                 return ERR_PTR(-EACCES);
4618
4619         if (oti->oti_it_inline) {
4620                 OBD_ALLOC_PTR(it);
4621                 if (it == NULL)
4622                         return ERR_PTR(-ENOMEM);
4623         } else {
4624                 it = &oti->oti_it;
4625                 oti->oti_it_inline = 1;
4626         }
4627
4628         ipd = osd_it_ipd_get(env, bag);
4629         if (likely(ipd != NULL)) {
4630                 it->oi_obj = obj;
4631                 it->oi_ipd = ipd;
4632                 lu_object_get(lo);
4633                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4634                 return (struct dt_it *)it;
4635         } else {
4636                 if (it != &oti->oti_it)
4637                         OBD_FREE_PTR(it);
4638                 else
4639                         oti->oti_it_inline = 0;
4640
4641                 return ERR_PTR(-ENOMEM);
4642         }
4643 }
4644
4645 /**
4646  * free given Iterator.
4647  */
4648
4649 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4650 {
4651         struct osd_thread_info  *oti = osd_oti_get(env);
4652         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4653         struct osd_object       *obj = it->oi_obj;
4654
4655         iam_it_fini(&it->oi_it);
4656         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4657         lu_object_put(env, &obj->oo_dt.do_lu);
4658         if (it != &oti->oti_it)
4659                 OBD_FREE_PTR(it);
4660         else
4661                 oti->oti_it_inline = 0;
4662 }
4663
4664 /**
4665  *  Move Iterator to record specified by \a key
4666  *
4667  *  \param  di      osd iterator
4668  *  \param  key     key for index
4669  *
4670  *  \retval +ve  di points to record with least key not larger than key
4671  *  \retval  0   di points to exact matched key
4672  *  \retval -ve  failure
4673  */
4674
4675 static int osd_it_iam_get(const struct lu_env *env,
4676                           struct dt_it *di, const struct dt_key *key)
4677 {
4678         struct osd_thread_info  *oti = osd_oti_get(env);
4679         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4680
4681         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4682                 /* swab quota uid/gid */
4683                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4684                 key = (struct dt_key *)&oti->oti_quota_id;
4685         }
4686
4687         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4688 }
4689
4690 /**
4691  *  Release Iterator
4692  *
4693  *  \param  di      osd iterator
4694  */
4695 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4696 {
4697         struct osd_it_iam *it = (struct osd_it_iam *)di;
4698
4699         iam_it_put(&it->oi_it);
4700 }
4701
4702 /**
4703  *  Move iterator by one record
4704  *
4705  *  \param  di      osd iterator
4706  *
4707  *  \retval +1   end of container reached
4708  *  \retval  0   success
4709  *  \retval -ve  failure
4710  */
4711
4712 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4713 {
4714         struct osd_it_iam *it = (struct osd_it_iam *)di;
4715
4716         return iam_it_next(&it->oi_it);
4717 }
4718
4719 /**
4720  * Return pointer to the key under iterator.
4721  */
4722
4723 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4724                                  const struct dt_it *di)
4725 {
4726         struct osd_thread_info *oti = osd_oti_get(env);
4727         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4728         struct osd_object      *obj = it->oi_obj;
4729         struct dt_key          *key;
4730
4731         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4732
4733         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4734                 /* swab quota uid/gid */
4735                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4736                 key = (struct dt_key *)&oti->oti_quota_id;
4737         }
4738
4739         return key;
4740 }
4741
4742 /**
4743  * Return size of key under iterator (in bytes)
4744  */
4745
4746 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4747 {
4748         struct osd_it_iam *it = (struct osd_it_iam *)di;
4749
4750         return iam_it_key_size(&it->oi_it);
4751 }
4752
4753 static inline void
4754 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4755 {
4756         /* check if file type is required */
4757         if (ent->lde_attrs & LUDA_TYPE) {
4758                 struct luda_type *lt;
4759                 int align = sizeof(*lt) - 1;
4760
4761                 len = (len + align) & ~align;
4762                 lt = (struct luda_type *)(ent->lde_name + len);
4763                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4764         }
4765
4766         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4767 }
4768
4769 /**
4770  * build lu direct from backend fs dirent.
4771  */
4772
4773 static inline void
4774 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4775                    char *name, __u16 namelen, __u16 type, __u32 attr)
4776 {
4777         ent->lde_attrs = attr | LUDA_FID;
4778         fid_cpu_to_le(&ent->lde_fid, fid);
4779
4780         ent->lde_hash = cpu_to_le64(offset);
4781         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4782
4783         strncpy(ent->lde_name, name, namelen);
4784         ent->lde_name[namelen] = '\0';
4785         ent->lde_namelen = cpu_to_le16(namelen);
4786
4787         /* append lustre attributes */
4788         osd_it_append_attrs(ent, namelen, type);
4789 }
4790
4791 /**
4792  * Return pointer to the record under iterator.
4793  */
4794 static int osd_it_iam_rec(const struct lu_env *env,
4795                           const struct dt_it *di,
4796                           struct dt_rec *dtrec, __u32 attr)
4797 {
4798         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4799         struct osd_thread_info *info = osd_oti_get(env);
4800         ENTRY;
4801
4802         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4803                 const struct osd_fid_pack *rec;
4804                 struct lu_fid             *fid = &info->oti_fid;
4805                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4806                 char                      *name;
4807                 int                        namelen;
4808                 __u64                      hash;
4809                 int                        rc;
4810
4811                 name = (char *)iam_it_key_get(&it->oi_it);
4812                 if (IS_ERR(name))
4813                         RETURN(PTR_ERR(name));
4814
4815                 namelen = iam_it_key_size(&it->oi_it);
4816
4817                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4818                 if (IS_ERR(rec))
4819                         RETURN(PTR_ERR(rec));
4820
4821                 rc = osd_fid_unpack(fid, rec);
4822                 if (rc)
4823                         RETURN(rc);
4824
4825                 hash = iam_it_store(&it->oi_it);
4826
4827                 /* IAM does not store object type in IAM index (dir) */
4828                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4829                                    0, LUDA_FID);
4830         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4831                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4832                            (struct iam_rec *)dtrec);
4833                 osd_quota_unpack(it->oi_obj, dtrec);
4834         } else {
4835                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4836                            (struct iam_rec *)dtrec);
4837         }
4838
4839         RETURN(0);
4840 }
4841
4842 /**
4843  * Returns cookie for current Iterator position.
4844  */
4845 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4846 {
4847         struct osd_it_iam *it = (struct osd_it_iam *)di;
4848
4849         return iam_it_store(&it->oi_it);
4850 }
4851
4852 /**
4853  * Restore iterator from cookie.
4854  *
4855  * \param  di      osd iterator
4856  * \param  hash    Iterator location cookie
4857  *
4858  * \retval +ve  di points to record with least key not larger than key.
4859  * \retval  0   di points to exact matched key
4860  * \retval -ve  failure
4861  */
4862
4863 static int osd_it_iam_load(const struct lu_env *env,
4864                            const struct dt_it *di, __u64 hash)
4865 {
4866         struct osd_it_iam *it = (struct osd_it_iam *)di;
4867
4868         return iam_it_load(&it->oi_it, hash);
4869 }
4870
4871 static const struct dt_index_operations osd_index_iam_ops = {
4872         .dio_lookup         = osd_index_iam_lookup,
4873         .dio_declare_insert = osd_index_declare_iam_insert,
4874         .dio_insert         = osd_index_iam_insert,
4875         .dio_declare_delete = osd_index_declare_iam_delete,
4876         .dio_delete         = osd_index_iam_delete,
4877         .dio_it     = {
4878                 .init     = osd_it_iam_init,
4879                 .fini     = osd_it_iam_fini,
4880                 .get      = osd_it_iam_get,
4881                 .put      = osd_it_iam_put,
4882                 .next     = osd_it_iam_next,
4883                 .key      = osd_it_iam_key,
4884                 .key_size = osd_it_iam_key_size,
4885                 .rec      = osd_it_iam_rec,
4886                 .store    = osd_it_iam_store,
4887                 .load     = osd_it_iam_load
4888         }
4889 };
4890
4891
4892 /**
4893  * Creates or initializes iterator context.
4894  *
4895  * \retval struct osd_it_ea, iterator structure on success
4896  *
4897  */
4898 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4899                                     struct dt_object *dt,
4900                                     __u32 attr,
4901                                     struct lustre_capa *capa)
4902 {
4903         struct osd_object       *obj  = osd_dt_obj(dt);
4904         struct osd_thread_info  *info = osd_oti_get(env);
4905         struct osd_it_ea        *it;
4906         struct file             *file;
4907         struct lu_object        *lo   = &dt->do_lu;
4908         struct dentry           *obj_dentry = &info->oti_it_dentry;
4909         ENTRY;
4910
4911         LASSERT(lu_object_exists(lo));
4912
4913         if (info->oti_it_inline) {
4914                 OBD_ALLOC_PTR(it);
4915                 if (it == NULL)
4916                         RETURN(ERR_PTR(-ENOMEM));
4917         } else {
4918                 it = &info->oti_it_ea;
4919                 info->oti_it_inline = 1;
4920         }
4921
4922         obj_dentry->d_inode = obj->oo_inode;
4923         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4924         obj_dentry->d_name.hash = 0;
4925
4926         it->oie_rd_dirent       = 0;
4927         it->oie_it_dirent       = 0;
4928         it->oie_dirent          = NULL;
4929         it->oie_buf             = info->oti_it_ea_buf;
4930         it->oie_obj             = obj;
4931
4932         file = &it->oie_file;
4933         /* Reset the "file" totally to avoid to reuse any old value from
4934          * former readdir handling, the "file->f_pos" should be zero. */
4935         memset(file, 0, sizeof(*file));
4936         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4937         if (attr & LUDA_64BITHASH)
4938                 file->f_mode    = FMODE_64BITHASH;
4939         else
4940                 file->f_mode    = FMODE_32BITHASH;
4941         file->f_dentry          = obj_dentry;
4942         file->f_mapping         = obj->oo_inode->i_mapping;
4943         file->f_op              = obj->oo_inode->i_fop;
4944         set_file_inode(file, obj->oo_inode);
4945
4946         lu_object_get(lo);
4947         RETURN((struct dt_it *) it);
4948 }
4949
4950 /**
4951  * Destroy or finishes iterator context.
4952  *
4953  * \param di iterator structure to be destroyed
4954  */
4955 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4956 {
4957         struct osd_thread_info  *info   = osd_oti_get(env);
4958         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4959         struct osd_object       *obj    = it->oie_obj;
4960         struct inode            *inode  = obj->oo_inode;
4961
4962         ENTRY;
4963         it->oie_file.f_op->release(inode, &it->oie_file);
4964         lu_object_put(env, &obj->oo_dt.do_lu);
4965         if (it != &info->oti_it_ea)
4966                 OBD_FREE_PTR(it);
4967         else
4968                 info->oti_it_inline = 0;
4969         EXIT;
4970 }
4971
4972 /**
4973  * It position the iterator at given key, so that next lookup continues from
4974  * that key Or it is similar to dio_it->load() but based on a key,
4975  * rather than file position.
4976  *
4977  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4978  * to the beginning.
4979  *
4980  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4981  */
4982 static int osd_it_ea_get(const struct lu_env *env,
4983                          struct dt_it *di, const struct dt_key *key)
4984 {
4985         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4986
4987         ENTRY;
4988         LASSERT(((const char *)key)[0] == '\0');
4989         it->oie_file.f_pos      = 0;
4990         it->oie_rd_dirent       = 0;
4991         it->oie_it_dirent       = 0;
4992         it->oie_dirent          = NULL;
4993
4994         RETURN(+1);
4995 }
4996
4997 /**
4998  * Does nothing
4999  */
5000 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5001 {
5002 }
5003
5004 struct osd_filldir_cbs {
5005 #ifdef HAVE_DIR_CONTEXT
5006         struct dir_context ctx;
5007 #endif
5008         struct osd_it_ea  *it;
5009 };
5010 /**
5011  * It is called internally by ->readdir(). It fills the
5012  * iterator's in-memory data structure with required
5013  * information i.e. name, namelen, rec_size etc.
5014  *
5015  * \param buf in which information to be filled in.
5016  * \param name name of the file in given dir
5017  *
5018  * \retval 0 on success
5019  * \retval 1 on buffer full
5020  */
5021 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5022                                loff_t offset, __u64 ino,
5023                                unsigned d_type)
5024 {
5025         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5026         struct osd_object       *obj  = it->oie_obj;
5027         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5028         struct lu_fid           *fid  = &ent->oied_fid;
5029         struct osd_fid_pack     *rec;
5030         ENTRY;
5031
5032         /* this should never happen */
5033         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5034                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5035                 RETURN(-EIO);
5036         }
5037
5038         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5039             OSD_IT_EA_BUFSIZE)
5040                 RETURN(1);
5041
5042         /* "." is just the object itself. */
5043         if (namelen == 1 && name[0] == '.') {
5044                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5045         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5046                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5047                 if (osd_fid_unpack(fid, rec) != 0)
5048                         fid_zero(fid);
5049         } else {
5050                 fid_zero(fid);
5051         }
5052         d_type &= ~LDISKFS_DIRENT_LUFID;
5053
5054         /* NOT export local root. */
5055         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5056                 ino = obj->oo_inode->i_ino;
5057                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5058         }
5059
5060         ent->oied_ino     = ino;
5061         ent->oied_off     = offset;
5062         ent->oied_namelen = namelen;
5063         ent->oied_type    = d_type;
5064
5065         memcpy(ent->oied_name, name, namelen);
5066
5067         it->oie_rd_dirent++;
5068         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5069         RETURN(0);
5070 }
5071
5072 /**
5073  * Calls ->readdir() to load a directory entry at a time
5074  * and stored it in iterator's in-memory data structure.
5075  *
5076  * \param di iterator's in memory structure
5077  *
5078  * \retval   0 on success
5079  * \retval -ve on error
5080  * \retval +1 reach the end of entry
5081  */
5082 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5083                                const struct dt_it *di)
5084 {
5085         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5086         struct osd_object  *obj   = it->oie_obj;
5087         struct inode       *inode = obj->oo_inode;
5088         struct htree_lock  *hlock = NULL;
5089         struct file        *filp  = &it->oie_file;
5090         int                 rc = 0;
5091         struct osd_filldir_cbs buf = {
5092 #ifdef HAVE_DIR_CONTEXT
5093                 .ctx.actor = osd_ldiskfs_filldir,
5094 #endif
5095                 .it = it
5096         };
5097
5098         ENTRY;
5099         it->oie_dirent = it->oie_buf;
5100         it->oie_rd_dirent = 0;
5101
5102         if (obj->oo_hl_head != NULL) {
5103                 hlock = osd_oti_get(env)->oti_hlock;
5104                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5105                                    inode, LDISKFS_HLOCK_READDIR);
5106         } else {
5107                 down_read(&obj->oo_ext_idx_sem);
5108         }
5109
5110 #ifdef HAVE_DIR_CONTEXT
5111         buf.ctx.pos = filp->f_pos;
5112         rc = inode->i_fop->iterate(filp, &buf.ctx);
5113         filp->f_pos = buf.ctx.pos;
5114 #else
5115         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5116 #endif
5117
5118         if (hlock != NULL)
5119                 ldiskfs_htree_unlock(hlock);
5120         else
5121                 up_read(&obj->oo_ext_idx_sem);
5122
5123         if (it->oie_rd_dirent == 0) {
5124                 /*If it does not get any dirent, it means it has been reached
5125                  *to the end of the dir */
5126                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5127                 if (rc == 0)
5128                         rc = 1;
5129         } else {
5130                 it->oie_dirent = it->oie_buf;
5131                 it->oie_it_dirent = 1;
5132         }
5133
5134         RETURN(rc);
5135 }
5136
5137 /**
5138  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5139  * to load a directory entry at a time and stored it in
5140  * iterator's in-memory data structure.
5141  *
5142  * \param di iterator's in memory structure
5143  *
5144  * \retval +ve iterator reached to end
5145  * \retval   0 iterator not reached to end
5146  * \retval -ve on error
5147  */
5148 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5149 {
5150         struct osd_it_ea *it = (struct osd_it_ea *)di;
5151         int rc;
5152
5153         ENTRY;
5154
5155         if (it->oie_it_dirent < it->oie_rd_dirent) {
5156                 it->oie_dirent =
5157                         (void *) it->oie_dirent +
5158                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5159                                        it->oie_dirent->oied_namelen);
5160                 it->oie_it_dirent++;
5161                 RETURN(0);
5162         } else {
5163                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5164                         rc = +1;
5165                 else
5166                         rc = osd_ldiskfs_it_fill(env, di);
5167         }
5168
5169         RETURN(rc);
5170 }
5171
5172 /**
5173  * Returns the key at current position from iterator's in memory structure.
5174  *
5175  * \param di iterator's in memory structure
5176  *
5177  * \retval key i.e. struct dt_key on success
5178  */
5179 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5180                                     const struct dt_it *di)
5181 {
5182         struct osd_it_ea *it = (struct osd_it_ea *)di;
5183
5184         return (struct dt_key *)it->oie_dirent->oied_name;
5185 }
5186
5187 /**
5188  * Returns the key's size at current position from iterator's in memory structure.
5189  *
5190  * \param di iterator's in memory structure
5191  *
5192  * \retval key_size i.e. struct dt_key on success
5193  */
5194 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5195 {
5196         struct osd_it_ea *it = (struct osd_it_ea *)di;
5197
5198         return it->oie_dirent->oied_namelen;
5199 }
5200
5201 static int
5202 osd_dirent_update(handle_t *jh, struct super_block *sb,
5203                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5204                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5205 {
5206         struct osd_fid_pack *rec;
5207         int                  rc;
5208         ENTRY;
5209
5210         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5211         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5212
5213         rc = ldiskfs_journal_get_write_access(jh, bh);
5214         if (rc != 0)
5215                 RETURN(rc);
5216
5217         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5218         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5219         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5220
5221         RETURN(rc);
5222 }
5223
5224 static inline int
5225 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5226 {
5227         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5228             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5229                 return 1;
5230         else
5231                 return 0;
5232 }
5233
5234 static inline int
5235 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5236 {
5237         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5238                  "dot_dotdot = %d\n", dot_dotdot);
5239
5240         if (LDISKFS_DIR_REC_LEN(de) >=
5241             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5242                 return 1;
5243         else
5244                 return 0;
5245 }
5246
5247 static int
5248 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5249                     struct inode *dir, struct inode *inode,
5250                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5251                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5252                     struct htree_lock *hlock)
5253 {
5254         struct dentry               *dentry;
5255         struct osd_fid_pack         *rec;
5256         struct ldiskfs_dentry_param *ldp;
5257         int                          rc;
5258         ENTRY;
5259
5260         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5261                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5262                 RETURN(0);
5263
5264         /* There is enough space to hold the FID-in-dirent. */
5265         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5266                                  dir->i_sb->s_blocksize)) {
5267                 rc = ldiskfs_journal_get_write_access(jh, bh);
5268                 if (rc != 0)
5269                         RETURN(rc);
5270
5271                 de->name[de->name_len] = 0;
5272                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5273                 rec->fp_len = sizeof(struct lu_fid) + 1;
5274                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5275                 de->file_type |= LDISKFS_DIRENT_LUFID;
5276
5277                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5278
5279                 RETURN(rc);
5280         }
5281
5282         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5283         if (rc != 0)
5284                 RETURN(rc);
5285
5286         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5287                                            ent->oied_namelen);
5288         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5289         osd_get_ldiskfs_dirent_param(ldp, fid);
5290         dentry->d_fsdata = (void *)ldp;
5291         ll_vfs_dq_init(dir);
5292         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5293         /* It is too bad, we cannot reinsert the name entry back.
5294          * That means we lose it! */
5295         if (rc != 0)
5296                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5297                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5298                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5299                        dir->i_ino, dir->i_generation,
5300                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5301
5302         RETURN(rc);
5303 }
5304
5305 static int
5306 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5307                         struct osd_it_ea *it, struct lu_fid *fid,
5308                         struct osd_inode_id *id, __u32 *attr)
5309 {
5310         struct osd_thread_info     *info        = osd_oti_get(env);
5311         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5312         struct osd_device          *dev         = osd_obj2dev(obj);
5313         struct super_block         *sb          = osd_sb(dev);
5314         const char                 *devname     =
5315                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5316         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5317         struct inode               *dir         = obj->oo_inode;
5318         struct htree_lock          *hlock       = NULL;
5319         struct buffer_head         *bh          = NULL;
5320         handle_t                   *jh          = NULL;
5321         struct ldiskfs_dir_entry_2 *de;
5322         struct dentry              *dentry;
5323         struct inode               *inode;
5324         int                         credits;
5325         int                         rc;
5326         int                         dot_dotdot  = 0;
5327         bool                        dirty       = false;
5328         ENTRY;
5329
5330         if (ent->oied_name[0] == '.') {
5331                 if (ent->oied_namelen == 1)
5332                         dot_dotdot = 1;
5333                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5334                         dot_dotdot = 2;
5335         }
5336
5337         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5338                                       ent->oied_namelen);
5339
5340         /* We need to ensure that the name entry is still valid.
5341          * Because it may be removed or renamed by other already.
5342          *
5343          * The unlink or rename operation will start journal before PDO lock,
5344          * so to avoid deadlock, here we need to start journal handle before
5345          * related PDO lock also. But because we do not know whether there
5346          * will be something to be repaired before PDO lock, we just start
5347          * journal without conditions.
5348          *
5349          * We may need to remove the name entry firstly, then insert back.
5350          * One credit is for user quota file update.
5351          * One credit is for group quota file update.
5352          * Two credits are for dirty inode. */
5353         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5354                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5355
5356 again:
5357         if (dev->od_dirent_journal) {
5358                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5359                 if (IS_ERR(jh)) {
5360                         rc = PTR_ERR(jh);
5361                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5362                                "check_repair, dir = %lu/%u, credits = %d, "
5363                                "name = %.*s: rc = %d\n",
5364                                devname, dir->i_ino, dir->i_generation, credits,
5365                                ent->oied_namelen, ent->oied_name, rc);
5366                         RETURN(rc);
5367                 }
5368
5369                 if (obj->oo_hl_head != NULL) {
5370                         hlock = osd_oti_get(env)->oti_hlock;
5371                         /* "0" means exclusive lock for the whole directory.
5372                          * We need to prevent others access such name entry
5373                          * during the delete + insert. Neither HLOCK_ADD nor
5374                          * HLOCK_DEL cannot guarantee the atomicity. */
5375                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5376                 } else {
5377                         down_write(&obj->oo_ext_idx_sem);
5378                 }
5379         } else {
5380                 if (obj->oo_hl_head != NULL) {
5381                         hlock = osd_oti_get(env)->oti_hlock;
5382                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5383                                            LDISKFS_HLOCK_LOOKUP);
5384                 } else {
5385                         down_read(&obj->oo_ext_idx_sem);
5386                 }
5387         }
5388
5389         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5390         /* For dot/dotdot entry, if there is not enough space to hold the
5391          * FID-in-dirent, just keep them there. It only happens when the
5392          * device upgraded from 1.8 or restored from MDT file-level backup.
5393          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5394          * and needs to get FID from LMA when readdir, it will not affect the
5395          * performance much. */
5396         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5397             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5398                 *attr |= LUDA_IGNORE;
5399                 GOTO(out_journal, rc = 0);
5400         }
5401
5402         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5403         inode = osd_iget(info, dev, id);
5404         if (IS_ERR(inode)) {
5405                 rc = PTR_ERR(inode);
5406                 if (rc == -ENOENT || rc == -ESTALE)
5407                         rc = 1;
5408                 else
5409                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5410                                "check_repair, dir = %lu/%u, name = %.*s: "
5411                                "rc = %d\n",
5412                                devname, dir->i_ino, dir->i_generation,
5413                                ent->oied_namelen, ent->oied_name, rc);
5414
5415                 GOTO(out_journal, rc);
5416         }
5417
5418         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5419         if (rc == 0) {
5420                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5421
5422                 if (fid_is_sane(fid)) {
5423                         /* FID-in-dirent is valid. */
5424                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5425                                 GOTO(out_inode, rc = 0);
5426
5427                         /* Do not repair under dryrun mode. */
5428                         if (*attr & LUDA_VERIFY_DRYRUN) {
5429                                 *attr |= LUDA_REPAIR;
5430                                 GOTO(out_inode, rc = 0);
5431                         }
5432
5433                         if (!dev->od_dirent_journal) {
5434                                 iput(inode);
5435                                 brelse(bh);
5436                                 if (hlock != NULL)
5437                                         ldiskfs_htree_unlock(hlock);
5438                                 else
5439                                         up_read(&obj->oo_ext_idx_sem);
5440                                 dev->od_dirent_journal = 1;
5441                                 goto again;
5442                         }
5443
5444                         *fid = lma->lma_self_fid;
5445                         dirty = true;
5446                         /* Update the FID-in-dirent. */
5447                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5448                         if (rc == 0)
5449                                 *attr |= LUDA_REPAIR;
5450                         else
5451                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5452                                        "in the dirent, dir = %lu/%u, "
5453                                        "name = %.*s, "DFID": rc = %d\n",
5454                                        devname, dir->i_ino, dir->i_generation,
5455                                        ent->oied_namelen, ent->oied_name,
5456                                        PFID(fid), rc);
5457                 } else {
5458                         /* Do not repair under dryrun mode. */
5459                         if (*attr & LUDA_VERIFY_DRYRUN) {
5460                                 *fid = lma->lma_self_fid;
5461                                 *attr |= LUDA_REPAIR;
5462                                 GOTO(out_inode, rc = 0);
5463                         }
5464
5465                         if (!dev->od_dirent_journal) {
5466                                 iput(inode);
5467                                 brelse(bh);
5468                                 if (hlock != NULL)
5469                                         ldiskfs_htree_unlock(hlock);
5470                                 else
5471                                         up_read(&obj->oo_ext_idx_sem);
5472                                 dev->od_dirent_journal = 1;
5473                                 goto again;
5474                         }
5475
5476                         *fid = lma->lma_self_fid;
5477                         dirty = true;
5478                         /* Append the FID-in-dirent. */
5479                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5480                                                  fid, bh, de, hlock);
5481                         if (rc == 0)
5482                                 *attr |= LUDA_REPAIR;
5483                         else
5484                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5485                                        "after the dirent, dir = %lu/%u, "
5486                                        "name = %.*s, "DFID": rc = %d\n",
5487                                        devname, dir->i_ino, dir->i_generation,
5488                                        ent->oied_namelen, ent->oied_name,
5489                                        PFID(fid), rc);
5490                 }
5491         } else if (rc == -ENODATA) {
5492                 /* Do not repair under dryrun mode. */
5493                 if (*attr & LUDA_VERIFY_DRYRUN) {
5494                         if (fid_is_sane(fid)) {
5495                                 *attr |= LUDA_REPAIR;
5496                         } else {
5497                                 lu_igif_build(fid, inode->i_ino,
5498                                               inode->i_generation);
5499                                 *attr |= LUDA_UPGRADE;
5500                         }
5501                         GOTO(out_inode, rc = 0);
5502                 }
5503
5504                 if (!dev->od_dirent_journal) {
5505                         iput(inode);
5506                         brelse(bh);
5507                         if (hlock != NULL)
5508                                 ldiskfs_htree_unlock(hlock);
5509                         else
5510                                 up_read(&obj->oo_ext_idx_sem);
5511                         dev->od_dirent_journal = 1;
5512                         goto again;
5513                 }
5514
5515                 dirty = true;
5516                 if (unlikely(fid_is_sane(fid))) {
5517                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5518                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5519                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5520                         if (rc == 0)
5521                                 *attr |= LUDA_REPAIR;
5522                         else
5523                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5524                                        "update dirent, dir = %lu/%u, "
5525                                        "name = %.*s, "DFID": rc = %d\n",
5526                                        devname, dir->i_ino, dir->i_generation,
5527                                        ent->oied_namelen, ent->oied_name,
5528                                        PFID(fid), rc);
5529                 } else {
5530                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5531                         /* It is probably IGIF object. Only aappend the
5532                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5533                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5534                                                  fid, bh, de, hlock);
5535                         if (rc == 0)
5536                                 *attr |= LUDA_UPGRADE;
5537                         else
5538                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5539                                        "after the dirent, dir = %lu/%u, "
5540                                        "name = %.*s, "DFID": rc = %d\n",
5541                                        devname, dir->i_ino, dir->i_generation,
5542                                        ent->oied_namelen, ent->oied_name,
5543                                        PFID(fid), rc);
5544                 }
5545         }
5546
5547         GOTO(out_inode, rc);
5548
5549 out_inode:
5550         iput(inode);
5551
5552 out_journal:
5553         brelse(bh);
5554         if (hlock != NULL) {
5555                 ldiskfs_htree_unlock(hlock);
5556         } else {
5557                 if (dev->od_dirent_journal)
5558                         up_write(&obj->oo_ext_idx_sem);
5559                 else
5560                         up_read(&obj->oo_ext_idx_sem);
5561         }
5562         if (jh != NULL)
5563                 ldiskfs_journal_stop(jh);
5564         if (rc >= 0 && !dirty)
5565                 dev->od_dirent_journal = 0;
5566         return rc;
5567 }
5568
5569 /**
5570  * Returns the value at current position from iterator's in memory structure.
5571  *
5572  * \param di struct osd_it_ea, iterator's in memory structure
5573  * \param attr attr requested for dirent.
5574  * \param lde lustre dirent
5575  *
5576  * \retval   0 no error and \param lde has correct lustre dirent.
5577  * \retval -ve on error
5578  */
5579 static inline int osd_it_ea_rec(const struct lu_env *env,
5580                                 const struct dt_it *di,
5581                                 struct dt_rec *dtrec, __u32 attr)
5582 {
5583         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5584         struct osd_object      *obj   = it->oie_obj;
5585         struct osd_device      *dev   = osd_obj2dev(obj);
5586         struct osd_scrub       *scrub = &dev->od_scrub;
5587         struct scrub_file      *sf    = &scrub->os_file;
5588         struct osd_thread_info *oti   = osd_oti_get(env);
5589         struct osd_inode_id    *id    = &oti->oti_id;
5590         struct osd_idmap_cache *oic   = &oti->oti_cache;
5591         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5592         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5593         __u32                   ino   = it->oie_dirent->oied_ino;
5594         int                     rc    = 0;
5595         ENTRY;
5596
5597         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5598
5599         if (attr & LUDA_VERIFY) {
5600                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5601                         attr |= LUDA_IGNORE;
5602                         /* If the parent is on remote MDT, and there
5603                          * is no FID-in-dirent, then we have to get
5604                          * the parent FID from the linkEA.  */
5605                         osd_get_pfid_from_linkea(env, obj, fid);
5606                 } else {
5607                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5608                                                      &attr);
5609                 }
5610         } else {
5611                 attr &= ~LU_DIRENT_ATTRS_MASK;
5612                 if (!fid_is_sane(fid)) {
5613                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5614                             likely(it->oie_dirent->oied_namelen != 2 ||
5615                                    it->oie_dirent->oied_name[0] != '.' ||
5616                                    it->oie_dirent->oied_name[1] != '.'))
5617                                 RETURN(-ENOENT);
5618
5619                         if (unlikely(ino == osd_remote_parent_ino(dev)))
5620                                 /* If the parent is on remote MDT, and there
5621                                  * is no FID-in-dirent, then we have to get
5622                                  * the parent FID from the linkEA.  */
5623                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5624                         else
5625                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5626                 } else {
5627                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5628                 }
5629         }
5630
5631         /* Pack the entry anyway, at least the offset is right. */
5632         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5633                            it->oie_dirent->oied_name,
5634                            it->oie_dirent->oied_namelen,
5635                            it->oie_dirent->oied_type, attr);
5636
5637         if (rc < 0)
5638                 RETURN(rc);
5639
5640         if (osd_remote_fid(env, dev, fid))
5641                 RETURN(0);
5642
5643         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5644                 rc = osd_add_oi_cache(oti, dev, id, fid);
5645
5646         if (!(attr & LUDA_VERIFY) &&
5647             (scrub->os_pos_current <= ino) &&
5648             ((sf->sf_flags & SF_INCONSISTENT) ||
5649              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5650              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5651                 osd_consistency_check(oti, dev, oic);
5652
5653         RETURN(rc > 0 ? 0 : rc);
5654 }
5655
5656 /**
5657  * Returns the record size size at current position.
5658  *
5659  * This function will return record(lu_dirent) size in bytes.
5660  *
5661  * \param[in] env       execution environment
5662  * \param[in] di        iterator's in memory structure
5663  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5664  *                      calculate the lu_dirent size.
5665  *
5666  * \retval      record size(in bytes & in memory) of the current lu_dirent
5667  *              entry.
5668  */
5669 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5670                               __u32 attr)
5671 {
5672         struct osd_it_ea *it = (struct osd_it_ea *)di;
5673
5674         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5675 }
5676
5677 /**
5678  * Returns a cookie for current position of the iterator head, so that
5679  * user can use this cookie to load/start the iterator next time.
5680  *
5681  * \param di iterator's in memory structure
5682  *
5683  * \retval cookie for current position, on success
5684  */
5685 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5686 {
5687         struct osd_it_ea *it = (struct osd_it_ea *)di;
5688
5689         return it->oie_dirent->oied_off;
5690 }
5691
5692 /**
5693  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5694  * to load a directory entry at a time and stored it i inn,
5695  * in iterator's in-memory data structure.
5696  *
5697  * \param di struct osd_it_ea, iterator's in memory structure
5698  *
5699  * \retval +ve on success
5700  * \retval -ve on error
5701  */
5702 static int osd_it_ea_load(const struct lu_env *env,
5703                           const struct dt_it *di, __u64 hash)
5704 {
5705         struct osd_it_ea *it = (struct osd_it_ea *)di;
5706         int rc;
5707
5708         ENTRY;
5709         it->oie_file.f_pos = hash;
5710
5711         rc =  osd_ldiskfs_it_fill(env, di);
5712         if (rc > 0)
5713                 rc = -ENODATA;
5714
5715         if (rc == 0)
5716                 rc = +1;
5717
5718         RETURN(rc);
5719 }
5720
5721 /**
5722  * Index lookup function for interoperability mode (b11826).
5723  *
5724  * \param key,  key i.e. file name to be searched
5725  *
5726  * \retval +ve, on success
5727  * \retval -ve, on error
5728  */
5729 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5730                                struct dt_rec *rec, const struct dt_key *key,
5731                                struct lustre_capa *capa)
5732 {
5733         struct osd_object *obj = osd_dt_obj(dt);
5734         int rc = 0;
5735
5736         ENTRY;
5737
5738         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5739         LINVRNT(osd_invariant(obj));
5740
5741         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5742                 return -EACCES;
5743
5744         rc = osd_ea_lookup_rec(env, obj, rec, key);
5745         if (rc == 0)
5746                 rc = +1;
5747         RETURN(rc);
5748 }
5749
5750 /**
5751  * Index and Iterator operations for interoperability
5752  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5753  */
5754 static const struct dt_index_operations osd_index_ea_ops = {
5755         .dio_lookup         = osd_index_ea_lookup,
5756         .dio_declare_insert = osd_index_declare_ea_insert,
5757         .dio_insert         = osd_index_ea_insert,
5758         .dio_declare_delete = osd_index_declare_ea_delete,
5759         .dio_delete         = osd_index_ea_delete,
5760         .dio_it     = {
5761                 .init     = osd_it_ea_init,
5762                 .fini     = osd_it_ea_fini,
5763                 .get      = osd_it_ea_get,
5764                 .put      = osd_it_ea_put,
5765                 .next     = osd_it_ea_next,
5766                 .key      = osd_it_ea_key,
5767                 .key_size = osd_it_ea_key_size,
5768                 .rec      = osd_it_ea_rec,
5769                 .rec_size = osd_it_ea_rec_size,
5770                 .store    = osd_it_ea_store,
5771                 .load     = osd_it_ea_load
5772         }
5773 };
5774
5775 static void *osd_key_init(const struct lu_context *ctx,
5776                           struct lu_context_key *key)
5777 {
5778         struct osd_thread_info *info;
5779
5780         OBD_ALLOC_PTR(info);
5781         if (info == NULL)
5782                 return ERR_PTR(-ENOMEM);
5783
5784         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5785         if (info->oti_it_ea_buf == NULL)
5786                 goto out_free_info;
5787
5788         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5789
5790         info->oti_hlock = ldiskfs_htree_lock_alloc();
5791         if (info->oti_hlock == NULL)
5792                 goto out_free_ea;
5793
5794         return info;
5795
5796  out_free_ea:
5797         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5798  out_free_info:
5799         OBD_FREE_PTR(info);
5800         return ERR_PTR(-ENOMEM);
5801 }
5802
5803 static void osd_key_fini(const struct lu_context *ctx,
5804                          struct lu_context_key *key, void* data)
5805 {
5806         struct osd_thread_info *info = data;
5807
5808         if (info->oti_hlock != NULL)
5809                 ldiskfs_htree_lock_free(info->oti_hlock);
5810         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5811         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5812         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5813         lu_buf_free(&info->oti_big_buf);
5814         OBD_FREE_PTR(info);
5815 }
5816
5817 static void osd_key_exit(const struct lu_context *ctx,
5818                          struct lu_context_key *key, void *data)
5819 {
5820         struct osd_thread_info *info = data;
5821
5822         LASSERT(info->oti_r_locks == 0);
5823         LASSERT(info->oti_w_locks == 0);
5824         LASSERT(info->oti_txns    == 0);
5825 }
5826
5827 /* type constructor/destructor: osd_type_init, osd_type_fini */
5828 LU_TYPE_INIT_FINI(osd, &osd_key);
5829
5830 struct lu_context_key osd_key = {
5831         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5832         .lct_init = osd_key_init,
5833         .lct_fini = osd_key_fini,
5834         .lct_exit = osd_key_exit
5835 };
5836
5837
5838 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5839                            const char *name, struct lu_device *next)
5840 {
5841         struct osd_device *osd = osd_dev(d);
5842
5843         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5844             >= sizeof(osd->od_svname))
5845                 return -E2BIG;
5846         return osd_procfs_init(osd, name);
5847 }
5848
5849 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5850 {
5851         struct seq_server_site  *ss = osd_seq_site(osd);
5852         int                     rc;
5853         ENTRY;
5854
5855         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5856                 RETURN(0);
5857
5858         if (unlikely(ss == NULL))
5859                 RETURN(-ENODEV);
5860
5861         OBD_ALLOC_PTR(osd->od_cl_seq);
5862         if (osd->od_cl_seq == NULL)
5863                 RETURN(-ENOMEM);
5864
5865         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5866                              osd->od_svname, ss->ss_server_seq);
5867
5868         if (rc != 0) {
5869                 OBD_FREE_PTR(osd->od_cl_seq);
5870                 osd->od_cl_seq = NULL;
5871         }
5872
5873         RETURN(rc);
5874 }
5875
5876 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5877 {
5878         if (osd->od_cl_seq == NULL)
5879                 return;
5880
5881         seq_client_fini(osd->od_cl_seq);
5882         OBD_FREE_PTR(osd->od_cl_seq);
5883         osd->od_cl_seq = NULL;
5884 }
5885
5886 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5887 {
5888         ENTRY;
5889
5890         /* shutdown quota slave instance associated with the device */
5891         if (o->od_quota_slave != NULL) {
5892                 qsd_fini(env, o->od_quota_slave);
5893                 o->od_quota_slave = NULL;
5894         }
5895
5896         osd_fid_fini(env, o);
5897
5898         RETURN(0);
5899 }
5900
5901 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5902 {
5903         ENTRY;
5904
5905         if (o->od_mnt != NULL) {
5906                 shrink_dcache_sb(osd_sb(o));
5907                 osd_sync(env, &o->od_dt_dev);
5908
5909                 mntput(o->od_mnt);
5910                 o->od_mnt = NULL;
5911         }
5912
5913         EXIT;
5914 }
5915
5916 static int osd_mount(const struct lu_env *env,
5917                      struct osd_device *o, struct lustre_cfg *cfg)
5918 {
5919         const char              *name  = lustre_cfg_string(cfg, 0);
5920         const char              *dev  = lustre_cfg_string(cfg, 1);
5921         const char              *opts;
5922         unsigned long            page, s_flags, lmd_flags = 0;
5923         struct page             *__page;
5924         struct file_system_type *type;
5925         char                    *options = NULL;
5926         char                    *str;
5927         struct osd_thread_info  *info = osd_oti_get(env);
5928         struct lu_fid           *fid = &info->oti_fid;
5929         struct inode            *inode;
5930         int                      rc = 0;
5931         ENTRY;
5932
5933         if (o->od_mnt != NULL)
5934                 RETURN(0);
5935
5936         if (strlen(dev) >= sizeof(o->od_mntdev))
5937                 RETURN(-E2BIG);
5938         strcpy(o->od_mntdev, dev);
5939
5940         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5941         if (__page == NULL)
5942                 GOTO(out, rc = -ENOMEM);
5943
5944         str = lustre_cfg_string(cfg, 2);
5945         s_flags = simple_strtoul(str, NULL, 0);
5946         str = strstr(str, ":");
5947         if (str)
5948                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5949         opts = lustre_cfg_string(cfg, 3);
5950         page = (unsigned long)page_address(__page);
5951         options = (char *)page;
5952         *options = '\0';
5953         if (opts == NULL)
5954                 strcat(options, "user_xattr,acl");
5955         else
5956                 strcat(options, opts);
5957
5958         /* Glom up mount options */
5959         if (*options != '\0')
5960                 strcat(options, ",");
5961         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5962
5963         type = get_fs_type("ldiskfs");
5964         if (!type) {
5965                 CERROR("%s: cannot find ldiskfs module\n", name);
5966                 GOTO(out, rc = -ENODEV);
5967         }
5968
5969         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5970         module_put(type->owner);
5971
5972         if (IS_ERR(o->od_mnt)) {
5973                 rc = PTR_ERR(o->od_mnt);
5974                 o->od_mnt = NULL;
5975                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5976                 GOTO(out, rc);
5977         }
5978
5979 #ifdef HAVE_DEV_SET_RDONLY
5980         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5981                 CERROR("%s: underlying device %s is marked as read-only. "
5982                        "Setup failed\n", name, dev);
5983                 GOTO(out_mnt, rc = -EROFS);
5984         }
5985 #endif
5986
5987         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5988                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5989                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5990                 GOTO(out_mnt, rc = -EINVAL);
5991         }
5992
5993 #ifdef LDISKFS_MOUNT_DIRDATA
5994         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5995                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5996                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5997         else if (!o->od_is_ost)
5998                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5999                       "enabling the dirdata feature. If you do not want to "
6000                       "downgrade to Lustre-1.x again, you can enable it via "
6001                       "'tune2fs -O dirdata device'\n", name, dev);
6002 #endif
6003         inode = osd_sb(o)->s_root->d_inode;
6004         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6005         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6006         if (rc != 0) {
6007                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6008                 GOTO(out_mnt, rc);
6009         }
6010
6011         if (lmd_flags & LMD_FLG_NOSCRUB)
6012                 o->od_noscrub = 1;
6013
6014         GOTO(out, rc = 0);
6015
6016 out_mnt:
6017         mntput(o->od_mnt);
6018         o->od_mnt = NULL;
6019
6020 out:
6021         if (__page)
6022                 OBD_PAGE_FREE(__page);
6023
6024         return rc;
6025 }
6026
6027 static struct lu_device *osd_device_fini(const struct lu_env *env,
6028                                          struct lu_device *d)
6029 {
6030         struct osd_device *o = osd_dev(d);
6031         ENTRY;
6032
6033         osd_shutdown(env, o);
6034         osd_procfs_fini(o);
6035         osd_scrub_cleanup(env, o);
6036         osd_obj_map_fini(o);
6037         osd_umount(env, o);
6038
6039         RETURN(NULL);
6040 }
6041
6042 static int osd_device_init0(const struct lu_env *env,
6043                             struct osd_device *o,
6044                             struct lustre_cfg *cfg)
6045 {
6046         struct lu_device        *l = osd2lu_dev(o);
6047         struct osd_thread_info *info;
6048         int                     rc;
6049         int                     cplen = 0;
6050
6051         /* if the module was re-loaded, env can loose its keys */
6052         rc = lu_env_refill((struct lu_env *) env);
6053         if (rc)
6054                 GOTO(out, rc);
6055         info = osd_oti_get(env);
6056         LASSERT(info);
6057
6058         l->ld_ops = &osd_lu_ops;
6059         o->od_dt_dev.dd_ops = &osd_dt_ops;
6060
6061         spin_lock_init(&o->od_osfs_lock);
6062         mutex_init(&o->od_otable_mutex);
6063
6064         o->od_capa_hash = init_capa_hash();
6065         if (o->od_capa_hash == NULL)
6066                 GOTO(out, rc = -ENOMEM);
6067
6068         o->od_read_cache = 1;
6069         o->od_writethrough_cache = 1;
6070         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6071
6072         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6073                         sizeof(o->od_svname));
6074         if (cplen >= sizeof(o->od_svname)) {
6075                 rc = -E2BIG;
6076                 GOTO(out_capa, rc);
6077         }
6078
6079         if (server_name_is_ost(o->od_svname))
6080                 o->od_is_ost = 1;
6081
6082         rc = osd_mount(env, o, cfg);
6083         if (rc != 0)
6084                 GOTO(out_capa, rc);
6085
6086         rc = osd_obj_map_init(env, o);
6087         if (rc != 0)
6088                 GOTO(out_mnt, rc);
6089
6090         rc = lu_site_init(&o->od_site, l);
6091         if (rc != 0)
6092                 GOTO(out_compat, rc);
6093         o->od_site.ls_bottom_dev = l;
6094
6095         rc = lu_site_init_finish(&o->od_site);
6096         if (rc != 0)
6097                 GOTO(out_site, rc);
6098
6099         /* self-repair LMA by default */
6100         o->od_lma_self_repair = 1;
6101
6102         INIT_LIST_HEAD(&o->od_ios_list);
6103         /* setup scrub, including OI files initialization */
6104         rc = osd_scrub_setup(env, o);
6105         if (rc < 0)
6106                 GOTO(out_site, rc);
6107
6108         rc = osd_procfs_init(o, o->od_svname);
6109         if (rc != 0) {
6110                 CERROR("%s: can't initialize procfs: rc = %d\n",
6111                        o->od_svname, rc);
6112                 GOTO(out_scrub, rc);
6113         }
6114
6115         LASSERT(l->ld_site->ls_linkage.next != NULL);
6116         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6117
6118         /* initialize quota slave instance */
6119         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6120                                      o->od_proc_entry);
6121         if (IS_ERR(o->od_quota_slave)) {
6122                 rc = PTR_ERR(o->od_quota_slave);
6123                 o->od_quota_slave = NULL;
6124                 GOTO(out_procfs, rc);
6125         }
6126
6127         RETURN(0);
6128
6129 out_procfs:
6130         osd_procfs_fini(o);
6131 out_scrub:
6132         osd_scrub_cleanup(env, o);
6133 out_site:
6134         lu_site_fini(&o->od_site);
6135 out_compat:
6136         osd_obj_map_fini(o);
6137 out_mnt:
6138         osd_umount(env, o);
6139 out_capa:
6140         cleanup_capa_hash(o->od_capa_hash);
6141 out:
6142         return rc;
6143 }
6144
6145 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6146                                           struct lu_device_type *t,
6147                                           struct lustre_cfg *cfg)
6148 {
6149         struct osd_device *o;
6150         int                rc;
6151
6152         OBD_ALLOC_PTR(o);
6153         if (o == NULL)
6154                 return ERR_PTR(-ENOMEM);
6155
6156         rc = dt_device_init(&o->od_dt_dev, t);
6157         if (rc == 0) {
6158                 /* Because the ctx might be revived in dt_device_init,
6159                  * refill the env here */
6160                 lu_env_refill((struct lu_env *)env);
6161                 rc = osd_device_init0(env, o, cfg);
6162                 if (rc)
6163                         dt_device_fini(&o->od_dt_dev);
6164         }
6165
6166         if (unlikely(rc != 0))
6167                 OBD_FREE_PTR(o);
6168
6169         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6170 }
6171
6172 static struct lu_device *osd_device_free(const struct lu_env *env,
6173                                          struct lu_device *d)
6174 {
6175         struct osd_device *o = osd_dev(d);
6176         ENTRY;
6177
6178         cleanup_capa_hash(o->od_capa_hash);
6179         /* XXX: make osd top device in order to release reference */
6180         d->ld_site->ls_top_dev = d;
6181         lu_site_purge(env, d->ld_site, -1);
6182         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6183                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6184                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6185         }
6186         lu_site_fini(&o->od_site);
6187         dt_device_fini(&o->od_dt_dev);
6188         OBD_FREE_PTR(o);
6189         RETURN(NULL);
6190 }
6191
6192 static int osd_process_config(const struct lu_env *env,
6193                               struct lu_device *d, struct lustre_cfg *cfg)
6194 {
6195         struct osd_device               *o = osd_dev(d);
6196         int                             rc;
6197         ENTRY;
6198
6199         switch (cfg->lcfg_command) {
6200         case LCFG_SETUP:
6201                 rc = osd_mount(env, o, cfg);
6202                 break;
6203         case LCFG_CLEANUP:
6204                 lu_dev_del_linkage(d->ld_site, d);
6205                 rc = osd_shutdown(env, o);
6206                 break;
6207         case LCFG_PARAM:
6208                 LASSERT(&o->od_dt_dev);
6209                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6210                                               cfg, &o->od_dt_dev);
6211                 if (rc > 0 || rc == -ENOSYS)
6212                         rc = class_process_proc_param(PARAM_OST,
6213                                                       lprocfs_osd_obd_vars,
6214                                                       cfg, &o->od_dt_dev);
6215                 break;
6216         default:
6217                 rc = -ENOSYS;
6218         }
6219
6220         RETURN(rc);
6221 }
6222
6223 static int osd_recovery_complete(const struct lu_env *env,
6224                                  struct lu_device *d)
6225 {
6226         struct osd_device       *osd = osd_dev(d);
6227         int                      rc = 0;
6228         ENTRY;
6229
6230         if (osd->od_quota_slave == NULL)
6231                 RETURN(0);
6232
6233         /* start qsd instance on recovery completion, this notifies the quota
6234          * slave code that we are about to process new requests now */
6235         rc = qsd_start(env, osd->od_quota_slave);
6236         RETURN(rc);
6237 }
6238
6239 /*
6240  * we use exports to track all osd users
6241  */
6242 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6243                            struct obd_device *obd, struct obd_uuid *cluuid,
6244                            struct obd_connect_data *data, void *localdata)
6245 {
6246         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6247         struct lustre_handle  conn;
6248         int                   rc;
6249         ENTRY;
6250
6251         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6252
6253         rc = class_connect(&conn, obd, cluuid);
6254         if (rc)
6255                 RETURN(rc);
6256
6257         *exp = class_conn2export(&conn);
6258
6259         spin_lock(&osd->od_osfs_lock);
6260         osd->od_connects++;
6261         spin_unlock(&osd->od_osfs_lock);
6262
6263         RETURN(0);
6264 }
6265
6266 /*
6267  * once last export (we don't count self-export) disappeared
6268  * osd can be released
6269  */
6270 static int osd_obd_disconnect(struct obd_export *exp)
6271 {
6272         struct obd_device *obd = exp->exp_obd;
6273         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6274         int                rc, release = 0;
6275         ENTRY;
6276
6277         /* Only disconnect the underlying layers on the final disconnect. */
6278         spin_lock(&osd->od_osfs_lock);
6279         osd->od_connects--;
6280         if (osd->od_connects == 0)
6281                 release = 1;
6282         spin_unlock(&osd->od_osfs_lock);
6283
6284         rc = class_disconnect(exp); /* bz 9811 */
6285
6286         if (rc == 0 && release)
6287                 class_manual_cleanup(obd);
6288         RETURN(rc);
6289 }
6290
6291 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6292                        struct lu_device *dev)
6293 {
6294         struct osd_device *osd = osd_dev(dev);
6295         int                result = 0;
6296         ENTRY;
6297
6298         if (osd->od_quota_slave != NULL) {
6299                 /* set up quota slave objects */
6300                 result = qsd_prepare(env, osd->od_quota_slave);
6301                 if (result != 0)
6302                         RETURN(result);
6303         }
6304
6305         result = osd_fid_init(env, osd);
6306
6307         RETURN(result);
6308 }
6309
6310 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6311                   struct lu_fid *fid, struct md_op_data *op_data)
6312 {
6313         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6314
6315         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6316 }
6317
6318 static const struct lu_object_operations osd_lu_obj_ops = {
6319         .loo_object_init      = osd_object_init,
6320         .loo_object_delete    = osd_object_delete,
6321         .loo_object_release   = osd_object_release,
6322         .loo_object_free      = osd_object_free,
6323         .loo_object_print     = osd_object_print,
6324         .loo_object_invariant = osd_object_invariant
6325 };
6326
6327 const struct lu_device_operations osd_lu_ops = {
6328         .ldo_object_alloc      = osd_object_alloc,
6329         .ldo_process_config    = osd_process_config,
6330         .ldo_recovery_complete = osd_recovery_complete,
6331         .ldo_prepare           = osd_prepare,
6332 };
6333
6334 static const struct lu_device_type_operations osd_device_type_ops = {
6335         .ldto_init = osd_type_init,
6336         .ldto_fini = osd_type_fini,
6337
6338         .ldto_start = osd_type_start,
6339         .ldto_stop  = osd_type_stop,
6340
6341         .ldto_device_alloc = osd_device_alloc,
6342         .ldto_device_free  = osd_device_free,
6343
6344         .ldto_device_init    = osd_device_init,
6345         .ldto_device_fini    = osd_device_fini
6346 };
6347
6348 struct lu_device_type osd_device_type = {
6349         .ldt_tags     = LU_DEVICE_DT,
6350         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6351         .ldt_ops      = &osd_device_type_ops,
6352         .ldt_ctx_tags = LCT_LOCAL,
6353 };
6354
6355 /*
6356  * lprocfs legacy support.
6357  */
6358 static struct obd_ops osd_obd_device_ops = {
6359         .o_owner = THIS_MODULE,
6360         .o_connect      = osd_obd_connect,
6361         .o_disconnect   = osd_obd_disconnect,
6362         .o_fid_alloc    = osd_fid_alloc,
6363 };
6364
6365 static int __init osd_mod_init(void)
6366 {
6367         int rc;
6368
6369         osd_oi_mod_init();
6370
6371         rc = lu_kmem_init(ldiskfs_caches);
6372         if (rc)
6373                 return rc;
6374
6375         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6376                                  lprocfs_osd_module_vars,
6377                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6378         if (rc)
6379                 lu_kmem_fini(ldiskfs_caches);
6380         return rc;
6381 }
6382
6383 static void __exit osd_mod_exit(void)
6384 {
6385         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6386         lu_kmem_fini(ldiskfs_caches);
6387 }
6388
6389 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6390 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6391 MODULE_LICENSE("GPL");
6392
6393 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);