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         int rc;
1360
1361         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1362
1363         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1364
1365         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1366                LUSTRE_OSD_LDISKFS_NAME, rc);
1367
1368         return rc;
1369 }
1370
1371 /**
1372  * Start commit for OSD device.
1373  *
1374  * An implementation of dt_commit_async method for OSD device.
1375  * Asychronously starts underlayng fs sync and thereby a transaction
1376  * commit.
1377  *
1378  * \param env environment
1379  * \param d dt device
1380  *
1381  * \see dt_device_operations
1382  */
1383 static int osd_commit_async(const struct lu_env *env,
1384                             struct dt_device *d)
1385 {
1386         struct super_block *s = osd_sb(osd_dt_dev(d));
1387         ENTRY;
1388
1389         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1390         RETURN(s->s_op->sync_fs(s, 0));
1391 }
1392
1393 /*
1394  * Concurrency: shouldn't matter.
1395  */
1396
1397 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1398 {
1399         struct super_block *sb = osd_sb(osd_dt_dev(d));
1400         struct block_device *dev = sb->s_bdev;
1401 #ifdef HAVE_DEV_SET_RDONLY
1402         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1403         int rc = 0;
1404 #else
1405         int rc = -EOPNOTSUPP;
1406 #endif
1407         ENTRY;
1408
1409 #ifdef HAVE_DEV_SET_RDONLY
1410         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1411
1412         if (jdev && (jdev != dev)) {
1413                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1414                        (long)jdev);
1415                 dev_set_rdonly(jdev);
1416         }
1417         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1418         dev_set_rdonly(dev);
1419 #else
1420         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1421                osd_dt_dev(d)->od_svname, (long)dev, rc);
1422 #endif
1423         RETURN(rc);
1424 }
1425
1426 /*
1427  * Concurrency: serialization provided by callers.
1428  */
1429 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1430                               int mode, unsigned long timeout, __u32 alg,
1431                               struct lustre_capa_key *keys)
1432 {
1433         struct osd_device *dev = osd_dt_dev(d);
1434         ENTRY;
1435
1436         dev->od_fl_capa = mode;
1437         dev->od_capa_timeout = timeout;
1438         dev->od_capa_alg = alg;
1439         dev->od_capa_keys = keys;
1440         RETURN(0);
1441 }
1442
1443 /**
1444  * Note: we do not count into QUOTA here.
1445  * If we mount with --data_journal we may need more.
1446  */
1447 const int osd_dto_credits_noquota[DTO_NR] = {
1448         /**
1449          * Insert.
1450          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1451          * SINGLEDATA_TRANS_BLOCKS(8)
1452          * XXX Note: maybe iam need more, since iam have more level than
1453          *           EXT3 htree.
1454          */
1455         [DTO_INDEX_INSERT]  = 16,
1456         /**
1457          * Delete
1458          * just modify a single entry, probably merge few within a block
1459          */
1460         [DTO_INDEX_DELETE]  = 1,
1461         /**
1462          * Used for OI scrub
1463          */
1464         [DTO_INDEX_UPDATE]  = 16,
1465         /**
1466          * 4(inode, inode bits, groups, GDT)
1467          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1468          */
1469         [DTO_OBJECT_CREATE] = 4,
1470         /**
1471          * 4(inode, inode bits, groups, GDT)
1472          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1473          */
1474         [DTO_OBJECT_DELETE] = 4,
1475         /**
1476          * Attr set credits (inode)
1477          */
1478         [DTO_ATTR_SET_BASE] = 1,
1479         /**
1480          * Xattr set. The same as xattr of EXT3.
1481          * DATA_TRANS_BLOCKS(14)
1482          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1483          * are also counted in. Do not know why?
1484          */
1485         [DTO_XATTR_SET]     = 14,
1486         /**
1487          * credits for inode change during write.
1488          */
1489         [DTO_WRITE_BASE]    = 3,
1490         /**
1491          * credits for single block write.
1492          */
1493         [DTO_WRITE_BLOCK]   = 14,
1494         /**
1495          * Attr set credits for chown.
1496          * This is extra credits for setattr, and it is null without quota
1497          */
1498         [DTO_ATTR_SET_CHOWN] = 0
1499 };
1500
1501 static const struct dt_device_operations osd_dt_ops = {
1502         .dt_root_get       = osd_root_get,
1503         .dt_statfs         = osd_statfs,
1504         .dt_trans_create   = osd_trans_create,
1505         .dt_trans_start    = osd_trans_start,
1506         .dt_trans_stop     = osd_trans_stop,
1507         .dt_trans_cb_add   = osd_trans_cb_add,
1508         .dt_conf_get       = osd_conf_get,
1509         .dt_sync           = osd_sync,
1510         .dt_ro             = osd_ro,
1511         .dt_commit_async   = osd_commit_async,
1512         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1513 };
1514
1515 static void osd_object_read_lock(const struct lu_env *env,
1516                                  struct dt_object *dt, unsigned role)
1517 {
1518         struct osd_object *obj = osd_dt_obj(dt);
1519         struct osd_thread_info *oti = osd_oti_get(env);
1520
1521         LINVRNT(osd_invariant(obj));
1522
1523         LASSERT(obj->oo_owner != env);
1524         down_read_nested(&obj->oo_sem, role);
1525
1526         LASSERT(obj->oo_owner == NULL);
1527         oti->oti_r_locks++;
1528 }
1529
1530 static void osd_object_write_lock(const struct lu_env *env,
1531                                   struct dt_object *dt, unsigned role)
1532 {
1533         struct osd_object *obj = osd_dt_obj(dt);
1534         struct osd_thread_info *oti = osd_oti_get(env);
1535
1536         LINVRNT(osd_invariant(obj));
1537
1538         LASSERT(obj->oo_owner != env);
1539         down_write_nested(&obj->oo_sem, role);
1540
1541         LASSERT(obj->oo_owner == NULL);
1542         obj->oo_owner = env;
1543         oti->oti_w_locks++;
1544 }
1545
1546 static void osd_object_read_unlock(const struct lu_env *env,
1547                                    struct dt_object *dt)
1548 {
1549         struct osd_object *obj = osd_dt_obj(dt);
1550         struct osd_thread_info *oti = osd_oti_get(env);
1551
1552         LINVRNT(osd_invariant(obj));
1553
1554         LASSERT(oti->oti_r_locks > 0);
1555         oti->oti_r_locks--;
1556         up_read(&obj->oo_sem);
1557 }
1558
1559 static void osd_object_write_unlock(const struct lu_env *env,
1560                                     struct dt_object *dt)
1561 {
1562         struct osd_object *obj = osd_dt_obj(dt);
1563         struct osd_thread_info *oti = osd_oti_get(env);
1564
1565         LINVRNT(osd_invariant(obj));
1566
1567         LASSERT(obj->oo_owner == env);
1568         LASSERT(oti->oti_w_locks > 0);
1569         oti->oti_w_locks--;
1570         obj->oo_owner = NULL;
1571         up_write(&obj->oo_sem);
1572 }
1573
1574 static int osd_object_write_locked(const struct lu_env *env,
1575                                    struct dt_object *dt)
1576 {
1577         struct osd_object *obj = osd_dt_obj(dt);
1578
1579         LINVRNT(osd_invariant(obj));
1580
1581         return obj->oo_owner == env;
1582 }
1583
1584 static int capa_is_sane(const struct lu_env *env,
1585                         struct osd_device *dev,
1586                         struct lustre_capa *capa,
1587                         struct lustre_capa_key *keys)
1588 {
1589         struct osd_thread_info *oti = osd_oti_get(env);
1590         struct lustre_capa *tcapa = &oti->oti_capa;
1591         struct obd_capa *oc;
1592         int i, rc = 0;
1593         ENTRY;
1594
1595         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1596         if (oc) {
1597                 if (capa_is_expired(oc)) {
1598                         DEBUG_CAPA(D_ERROR, capa, "expired");
1599                         rc = -ESTALE;
1600                 }
1601                 capa_put(oc);
1602                 RETURN(rc);
1603         }
1604
1605         if (capa_is_expired_sec(capa)) {
1606                 DEBUG_CAPA(D_ERROR, capa, "expired");
1607                 RETURN(-ESTALE);
1608         }
1609
1610         spin_lock(&capa_lock);
1611         for (i = 0; i < 2; i++) {
1612                 if (keys[i].lk_keyid == capa->lc_keyid) {
1613                         oti->oti_capa_key = keys[i];
1614                         break;
1615                 }
1616         }
1617         spin_unlock(&capa_lock);
1618
1619         if (i == 2) {
1620                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1621                 RETURN(-ESTALE);
1622         }
1623
1624         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1625         if (rc)
1626                 RETURN(rc);
1627
1628         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1629                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1630                 RETURN(-EACCES);
1631         }
1632
1633         oc = capa_add(dev->od_capa_hash, capa);
1634         capa_put(oc);
1635
1636         RETURN(0);
1637 }
1638
1639 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1640                     struct lustre_capa *capa, __u64 opc)
1641 {
1642         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1643         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1644         struct lu_capainfo *lci;
1645         int rc;
1646
1647         if (!osd->od_fl_capa)
1648                 return 0;
1649
1650         if (capa == BYPASS_CAPA)
1651                 return 0;
1652
1653         lci = lu_capainfo_get(env);
1654         if (unlikely(lci == NULL))
1655                 return 0;
1656
1657         if (lci->lci_auth == LC_ID_NONE)
1658                 return 0;
1659
1660         if (capa == NULL) {
1661                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1662                        osd_name(osd), PFID(fid), -EACCES);
1663                 return -EACCES;
1664         }
1665
1666         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1667                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1668                            PFID(fid));
1669                 return -EACCES;
1670         }
1671
1672         if (!capa_opc_supported(capa, opc)) {
1673                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1674                 return -EACCES;
1675         }
1676
1677         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1678         if (rc != 0) {
1679                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1680                 return -EACCES;
1681         }
1682
1683         return 0;
1684 }
1685
1686 static struct timespec *osd_inode_time(const struct lu_env *env,
1687                                        struct inode *inode, __u64 seconds)
1688 {
1689         struct osd_thread_info  *oti = osd_oti_get(env);
1690         struct timespec         *t   = &oti->oti_time;
1691
1692         t->tv_sec = seconds;
1693         t->tv_nsec = 0;
1694         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1695         return t;
1696 }
1697
1698
1699 static void osd_inode_getattr(const struct lu_env *env,
1700                               struct inode *inode, struct lu_attr *attr)
1701 {
1702         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1703                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1704                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1705                            LA_TYPE;
1706
1707         attr->la_atime   = LTIME_S(inode->i_atime);
1708         attr->la_mtime   = LTIME_S(inode->i_mtime);
1709         attr->la_ctime   = LTIME_S(inode->i_ctime);
1710         attr->la_mode    = inode->i_mode;
1711         attr->la_size    = i_size_read(inode);
1712         attr->la_blocks  = inode->i_blocks;
1713         attr->la_uid     = i_uid_read(inode);
1714         attr->la_gid     = i_gid_read(inode);
1715         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1716         attr->la_nlink   = inode->i_nlink;
1717         attr->la_rdev    = inode->i_rdev;
1718         attr->la_blksize = 1 << inode->i_blkbits;
1719         attr->la_blkbits = inode->i_blkbits;
1720 }
1721
1722 static int osd_attr_get(const struct lu_env *env,
1723                         struct dt_object *dt,
1724                         struct lu_attr *attr,
1725                         struct lustre_capa *capa)
1726 {
1727         struct osd_object *obj = osd_dt_obj(dt);
1728
1729         LASSERT(dt_object_exists(dt));
1730         LASSERT(!dt_object_remote(dt));
1731         LINVRNT(osd_invariant(obj));
1732
1733         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1734                 return -EACCES;
1735
1736         spin_lock(&obj->oo_guard);
1737         osd_inode_getattr(env, obj->oo_inode, attr);
1738         spin_unlock(&obj->oo_guard);
1739         return 0;
1740 }
1741
1742 static int osd_declare_attr_set(const struct lu_env *env,
1743                                 struct dt_object *dt,
1744                                 const struct lu_attr *attr,
1745                                 struct thandle *handle)
1746 {
1747         struct osd_thandle     *oh;
1748         struct osd_object      *obj;
1749         struct osd_thread_info *info = osd_oti_get(env);
1750         struct lquota_id_info  *qi = &info->oti_qi;
1751         qid_t                   uid;
1752         qid_t                   gid;
1753         long long               bspace;
1754         int                     rc = 0;
1755         bool                    enforce;
1756         ENTRY;
1757
1758         LASSERT(dt != NULL);
1759         LASSERT(handle != NULL);
1760
1761         obj = osd_dt_obj(dt);
1762         LASSERT(osd_invariant(obj));
1763
1764         oh = container_of0(handle, struct osd_thandle, ot_super);
1765         LASSERT(oh->ot_handle == NULL);
1766
1767         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1768                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1769
1770         if (attr == NULL || obj->oo_inode == NULL)
1771                 RETURN(rc);
1772
1773         bspace   = obj->oo_inode->i_blocks;
1774         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1775         bspace   = toqb(bspace);
1776
1777         /* Changing ownership is always preformed by super user, it should not
1778          * fail with EDQUOT.
1779          *
1780          * We still need to call the osd_declare_qid() to calculate the journal
1781          * credits for updating quota accounting files and to trigger quota
1782          * space adjustment once the operation is completed.*/
1783         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1784                 /* USERQUOTA */
1785                 uid = i_uid_read(obj->oo_inode);
1786                 qi->lqi_type = USRQUOTA;
1787                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1788                 /* inode accounting */
1789                 qi->lqi_is_blk = false;
1790
1791                 /* one more inode for the new uid ... */
1792                 qi->lqi_id.qid_uid = attr->la_uid;
1793                 qi->lqi_space      = 1;
1794                 /* Reserve credits for the new uid */
1795                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1796                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1797                         rc = 0;
1798                 if (rc)
1799                         RETURN(rc);
1800
1801                 /* and one less inode for the current uid */
1802                 qi->lqi_id.qid_uid = uid;
1803                 qi->lqi_space      = -1;
1804                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1805                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1806                         rc = 0;
1807                 if (rc)
1808                         RETURN(rc);
1809
1810                 /* block accounting */
1811                 qi->lqi_is_blk = true;
1812
1813                 /* more blocks for the new uid ... */
1814                 qi->lqi_id.qid_uid = attr->la_uid;
1815                 qi->lqi_space      = bspace;
1816                 /*
1817                  * Credits for the new uid has been reserved, re-use "obj"
1818                  * to save credit reservation.
1819                  */
1820                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1821                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1822                         rc = 0;
1823                 if (rc)
1824                         RETURN(rc);
1825
1826                 /* and finally less blocks for the current uid */
1827                 qi->lqi_id.qid_uid = uid;
1828                 qi->lqi_space      = -bspace;
1829                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1830                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1831                         rc = 0;
1832                 if (rc)
1833                         RETURN(rc);
1834
1835                 /* GROUP QUOTA */
1836                 gid = i_gid_read(obj->oo_inode);
1837                 qi->lqi_type = GRPQUOTA;
1838                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1839
1840                 /* inode accounting */
1841                 qi->lqi_is_blk = false;
1842
1843                 /* one more inode for the new gid ... */
1844                 qi->lqi_id.qid_gid = attr->la_gid;
1845                 qi->lqi_space      = 1;
1846                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1847                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1848                         rc = 0;
1849                 if (rc)
1850                         RETURN(rc);
1851
1852                 /* and one less inode for the current gid */
1853                 qi->lqi_id.qid_gid = gid;
1854                 qi->lqi_space      = -1;
1855                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1856                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1857                         rc = 0;
1858                 if (rc)
1859                         RETURN(rc);
1860
1861                 /* block accounting */
1862                 qi->lqi_is_blk = true;
1863
1864                 /* more blocks for the new gid ... */
1865                 qi->lqi_id.qid_gid = attr->la_gid;
1866                 qi->lqi_space      = bspace;
1867                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1868                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1869                         rc = 0;
1870                 if (rc)
1871                         RETURN(rc);
1872
1873                 /* and finally less blocks for the current gid */
1874                 qi->lqi_id.qid_gid = gid;
1875                 qi->lqi_space      = -bspace;
1876                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1877                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1878                         rc = 0;
1879                 if (rc)
1880                         RETURN(rc);
1881         }
1882
1883         RETURN(rc);
1884 }
1885
1886 static int osd_inode_setattr(const struct lu_env *env,
1887                              struct inode *inode, const struct lu_attr *attr)
1888 {
1889         __u64 bits = attr->la_valid;
1890
1891         /* Only allow set size for regular file */
1892         if (!S_ISREG(inode->i_mode))
1893                 bits &= ~(LA_SIZE | LA_BLOCKS);
1894
1895         if (bits == 0)
1896                 return 0;
1897
1898         if (bits & LA_ATIME)
1899                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1900         if (bits & LA_CTIME)
1901                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1902         if (bits & LA_MTIME)
1903                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1904         if (bits & LA_SIZE) {
1905                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1906                 i_size_write(inode, attr->la_size);
1907         }
1908
1909 #if 0
1910         /* OSD should not change "i_blocks" which is used by quota.
1911          * "i_blocks" should be changed by ldiskfs only. */
1912         if (bits & LA_BLOCKS)
1913                 inode->i_blocks = attr->la_blocks;
1914 #endif
1915         if (bits & LA_MODE)
1916                 inode->i_mode = (inode->i_mode & S_IFMT) |
1917                                 (attr->la_mode & ~S_IFMT);
1918         if (bits & LA_UID)
1919                 i_uid_write(inode, attr->la_uid);
1920         if (bits & LA_GID)
1921                 i_gid_write(inode, attr->la_gid);
1922         if (bits & LA_NLINK)
1923                 set_nlink(inode, attr->la_nlink);
1924         if (bits & LA_RDEV)
1925                 inode->i_rdev = attr->la_rdev;
1926
1927         if (bits & LA_FLAGS) {
1928                 /* always keep S_NOCMTIME */
1929                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1930                                  S_NOCMTIME;
1931         }
1932         return 0;
1933 }
1934
1935 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1936 {
1937         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1938             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1939                 struct iattr    iattr;
1940                 int             rc;
1941
1942                 ll_vfs_dq_init(inode);
1943                 iattr.ia_valid = 0;
1944                 if (attr->la_valid & LA_UID)
1945                         iattr.ia_valid |= ATTR_UID;
1946                 if (attr->la_valid & LA_GID)
1947                         iattr.ia_valid |= ATTR_GID;
1948                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1949                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1950
1951                 rc = ll_vfs_dq_transfer(inode, &iattr);
1952                 if (rc) {
1953                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1954                                "enforcement enabled on the ldiskfs "
1955                                "filesystem?\n", inode->i_sb->s_id, rc);
1956                         return rc;
1957                 }
1958         }
1959         return 0;
1960 }
1961
1962 static int osd_attr_set(const struct lu_env *env,
1963                         struct dt_object *dt,
1964                         const struct lu_attr *attr,
1965                         struct thandle *handle,
1966                         struct lustre_capa *capa)
1967 {
1968         struct osd_object *obj = osd_dt_obj(dt);
1969         struct inode      *inode;
1970         int rc;
1971
1972         LASSERT(handle != NULL);
1973         LASSERT(dt_object_exists(dt));
1974         LASSERT(!dt_object_remote(dt));
1975         LASSERT(osd_invariant(obj));
1976
1977         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1978                 return -EACCES;
1979
1980         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1981
1982         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1983                 struct osd_thread_info  *oti  = osd_oti_get(env);
1984                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1985                 struct lu_fid           *fid1 = &oti->oti_fid;
1986                 struct osd_inode_id     *id   = &oti->oti_id;
1987                 struct iam_path_descr   *ipd;
1988                 struct iam_container    *bag;
1989                 struct osd_thandle      *oh;
1990                 int                      rc;
1991
1992                 fid_cpu_to_be(fid1, fid0);
1993                 memset(id, 1, sizeof(*id));
1994                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1995                                   fid0)->oi_dir.od_container;
1996                 ipd = osd_idx_ipd_get(env, bag);
1997                 if (unlikely(ipd == NULL))
1998                         RETURN(-ENOMEM);
1999
2000                 oh = container_of0(handle, struct osd_thandle, ot_super);
2001                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2002                                 (const struct iam_rec *)id, ipd);
2003                 osd_ipd_put(env, bag, ipd);
2004                 return(rc > 0 ? 0 : rc);
2005         }
2006
2007         inode = obj->oo_inode;
2008
2009         rc = osd_quota_transfer(inode, attr);
2010         if (rc)
2011                 return rc;
2012
2013         spin_lock(&obj->oo_guard);
2014         rc = osd_inode_setattr(env, inode, attr);
2015         spin_unlock(&obj->oo_guard);
2016
2017         if (!rc)
2018                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2019         return rc;
2020 }
2021
2022 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2023                                     struct osd_object *obj,
2024                                     const char *name, const int namelen)
2025 {
2026         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2027 }
2028
2029 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2030                       umode_t mode, struct dt_allocation_hint *hint,
2031                       struct thandle *th)
2032 {
2033         int result;
2034         struct osd_device  *osd = osd_obj2dev(obj);
2035         struct osd_thandle *oth;
2036         struct dt_object   *parent = NULL;
2037         struct inode       *inode;
2038
2039         LINVRNT(osd_invariant(obj));
2040         LASSERT(obj->oo_inode == NULL);
2041         LASSERT(obj->oo_hl_head == NULL);
2042
2043         if (S_ISDIR(mode) && ldiskfs_pdo) {
2044                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2045                 if (obj->oo_hl_head == NULL)
2046                         return -ENOMEM;
2047         }
2048
2049         oth = container_of(th, struct osd_thandle, ot_super);
2050         LASSERT(oth->ot_handle->h_transaction != NULL);
2051
2052         if (hint && hint->dah_parent)
2053                 parent = hint->dah_parent;
2054
2055         inode = ldiskfs_create_inode(oth->ot_handle,
2056                                      parent ? osd_dt_obj(parent)->oo_inode :
2057                                               osd_sb(osd)->s_root->d_inode,
2058                                      mode);
2059         if (!IS_ERR(inode)) {
2060                 /* Do not update file c/mtime in ldiskfs.
2061                  * NB: don't need any lock because no contention at this
2062                  * early stage */
2063                 inode->i_flags |= S_NOCMTIME;
2064
2065                 /* For new created object, it must be consistent,
2066                  * and it is unnecessary to scrub against it. */
2067                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2068                 obj->oo_inode = inode;
2069                 result = 0;
2070         } else {
2071                 if (obj->oo_hl_head != NULL) {
2072                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2073                         obj->oo_hl_head = NULL;
2074                 }
2075                 result = PTR_ERR(inode);
2076         }
2077         LINVRNT(osd_invariant(obj));
2078         return result;
2079 }
2080
2081 enum {
2082         OSD_NAME_LEN = 255
2083 };
2084
2085 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2086                      struct lu_attr *attr,
2087                      struct dt_allocation_hint *hint,
2088                      struct dt_object_format *dof,
2089                      struct thandle *th)
2090 {
2091         int result;
2092         struct osd_thandle *oth;
2093         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2094
2095         LASSERT(S_ISDIR(attr->la_mode));
2096
2097         oth = container_of(th, struct osd_thandle, ot_super);
2098         LASSERT(oth->ot_handle->h_transaction != NULL);
2099         result = osd_mkfile(info, obj, mode, hint, th);
2100
2101         return result;
2102 }
2103
2104 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2105                         struct lu_attr *attr,
2106                         struct dt_allocation_hint *hint,
2107                         struct dt_object_format *dof,
2108                         struct thandle *th)
2109 {
2110         int result;
2111         struct osd_thandle *oth;
2112         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2113
2114         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2115
2116         LASSERT(S_ISREG(attr->la_mode));
2117
2118         oth = container_of(th, struct osd_thandle, ot_super);
2119         LASSERT(oth->ot_handle->h_transaction != NULL);
2120
2121         result = osd_mkfile(info, obj, mode, hint, th);
2122         if (result == 0) {
2123                 LASSERT(obj->oo_inode != NULL);
2124                 if (feat->dif_flags & DT_IND_VARKEY)
2125                         result = iam_lvar_create(obj->oo_inode,
2126                                                  feat->dif_keysize_max,
2127                                                  feat->dif_ptrsize,
2128                                                  feat->dif_recsize_max,
2129                                                  oth->ot_handle);
2130                 else
2131                         result = iam_lfix_create(obj->oo_inode,
2132                                                  feat->dif_keysize_max,
2133                                                  feat->dif_ptrsize,
2134                                                  feat->dif_recsize_max,
2135                                                  oth->ot_handle);
2136
2137         }
2138         return result;
2139 }
2140
2141 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2142                      struct lu_attr *attr,
2143                      struct dt_allocation_hint *hint,
2144                      struct dt_object_format *dof,
2145                      struct thandle *th)
2146 {
2147         LASSERT(S_ISREG(attr->la_mode));
2148         return osd_mkfile(info, obj, (attr->la_mode &
2149                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2150 }
2151
2152 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2153                      struct lu_attr *attr,
2154                      struct dt_allocation_hint *hint,
2155                      struct dt_object_format *dof,
2156                      struct thandle *th)
2157 {
2158         LASSERT(S_ISLNK(attr->la_mode));
2159         return osd_mkfile(info, obj, (attr->la_mode &
2160                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2161 }
2162
2163 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2164                      struct lu_attr *attr,
2165                      struct dt_allocation_hint *hint,
2166                      struct dt_object_format *dof,
2167                      struct thandle *th)
2168 {
2169         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2170         int result;
2171
2172         LINVRNT(osd_invariant(obj));
2173         LASSERT(obj->oo_inode == NULL);
2174         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2175                 S_ISFIFO(mode) || S_ISSOCK(mode));
2176
2177         result = osd_mkfile(info, obj, mode, hint, th);
2178         if (result == 0) {
2179                 LASSERT(obj->oo_inode != NULL);
2180                 /*
2181                  * This inode should be marked dirty for i_rdev.  Currently
2182                  * that is done in the osd_attr_init().
2183                  */
2184                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2185                                    attr->la_rdev);
2186         }
2187         LINVRNT(osd_invariant(obj));
2188         return result;
2189 }
2190
2191 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2192                               struct lu_attr *,
2193                               struct dt_allocation_hint *hint,
2194                               struct dt_object_format *dof,
2195                               struct thandle *);
2196
2197 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2198 {
2199         osd_obj_type_f result;
2200
2201         switch (type) {
2202         case DFT_DIR:
2203                 result = osd_mkdir;
2204                 break;
2205         case DFT_REGULAR:
2206                 result = osd_mkreg;
2207                 break;
2208         case DFT_SYM:
2209                 result = osd_mksym;
2210                 break;
2211         case DFT_NODE:
2212                 result = osd_mknod;
2213                 break;
2214         case DFT_INDEX:
2215                 result = osd_mk_index;
2216                 break;
2217
2218         default:
2219                 LBUG();
2220                 break;
2221         }
2222         return result;
2223 }
2224
2225
2226 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2227                         struct dt_object *parent, struct dt_object *child,
2228                         umode_t child_mode)
2229 {
2230         LASSERT(ah);
2231
2232         ah->dah_parent = parent;
2233         ah->dah_mode = child_mode;
2234 }
2235
2236 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2237                           struct lu_attr *attr, struct dt_object_format *dof)
2238 {
2239         struct inode   *inode = obj->oo_inode;
2240         __u64           valid = attr->la_valid;
2241         int             result;
2242
2243         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2244
2245         if (dof->dof_type != DFT_NODE)
2246                 attr->la_valid &= ~LA_RDEV;
2247         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2248                 attr->la_valid &= ~LA_ATIME;
2249         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2250                 attr->la_valid &= ~LA_CTIME;
2251         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2252                 attr->la_valid &= ~LA_MTIME;
2253
2254         result = osd_quota_transfer(inode, attr);
2255         if (result)
2256                 return;
2257
2258         if (attr->la_valid != 0) {
2259                 result = osd_inode_setattr(info->oti_env, inode, attr);
2260                 /*
2261                  * The osd_inode_setattr() should always succeed here.  The
2262                  * only error that could be returned is EDQUOT when we are
2263                  * trying to change the UID or GID of the inode. However, this
2264                  * should not happen since quota enforcement is no longer
2265                  * enabled on ldiskfs (lquota takes care of it).
2266                  */
2267                 LASSERTF(result == 0, "%d\n", result);
2268                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2269         }
2270
2271         attr->la_valid = valid;
2272 }
2273
2274 /**
2275  * Helper function for osd_object_create()
2276  *
2277  * \retval 0, on success
2278  */
2279 static int __osd_object_create(struct osd_thread_info *info,
2280                                struct osd_object *obj, struct lu_attr *attr,
2281                                struct dt_allocation_hint *hint,
2282                                struct dt_object_format *dof,
2283                                struct thandle *th)
2284 {
2285         int     result;
2286         __u32   umask;
2287
2288         /* we drop umask so that permissions we pass are not affected */
2289         umask = current->fs->umask;
2290         current->fs->umask = 0;
2291
2292         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2293                                                   th);
2294         if (result == 0) {
2295                 osd_attr_init(info, obj, attr, dof);
2296                 osd_object_init0(obj);
2297                 /* bz 24037 */
2298                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2299                         unlock_new_inode(obj->oo_inode);
2300         }
2301
2302         /* restore previous umask value */
2303         current->fs->umask = umask;
2304
2305         return result;
2306 }
2307
2308 /**
2309  * Helper function for osd_object_create()
2310  *
2311  * \retval 0, on success
2312  */
2313 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2314                            const struct lu_fid *fid, struct thandle *th)
2315 {
2316         struct osd_thread_info *info = osd_oti_get(env);
2317         struct osd_inode_id    *id   = &info->oti_id;
2318         struct osd_device      *osd  = osd_obj2dev(obj);
2319         struct osd_thandle     *oh;
2320
2321         LASSERT(obj->oo_inode != NULL);
2322
2323         oh = container_of0(th, struct osd_thandle, ot_super);
2324         LASSERT(oh->ot_handle);
2325
2326         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2327         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2328 }
2329
2330 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2331                    obd_seq seq, struct lu_seq_range *range)
2332 {
2333         struct seq_server_site  *ss = osd_seq_site(osd);
2334
2335         if (fid_seq_is_idif(seq)) {
2336                 fld_range_set_ost(range);
2337                 range->lsr_index = idif_ost_idx(seq);
2338                 return 0;
2339         }
2340
2341         if (!fid_seq_in_fldb(seq)) {
2342                 fld_range_set_mdt(range);
2343                 if (ss != NULL)
2344                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2345                          * at all */
2346                         range->lsr_index = ss->ss_node_id;
2347                 return 0;
2348         }
2349
2350         LASSERT(ss != NULL);
2351         fld_range_set_any(range);
2352         /* OSD will only do local fld lookup */
2353         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2354 }
2355
2356 /*
2357  * Concurrency: no external locking is necessary.
2358  */
2359 static int osd_declare_object_create(const struct lu_env *env,
2360                                      struct dt_object *dt,
2361                                      struct lu_attr *attr,
2362                                      struct dt_allocation_hint *hint,
2363                                      struct dt_object_format *dof,
2364                                      struct thandle *handle)
2365 {
2366         struct osd_thandle      *oh;
2367         int                      rc;
2368         ENTRY;
2369
2370         LASSERT(handle != NULL);
2371
2372         oh = container_of0(handle, struct osd_thandle, ot_super);
2373         LASSERT(oh->ot_handle == NULL);
2374
2375         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2376                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2377         /* Reuse idle OI block may cause additional one OI block
2378          * to be changed. */
2379         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2380                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2381
2382         /* If this is directory, then we expect . and .. to be inserted as
2383          * well. The one directory block always needs to be created for the
2384          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2385          * block), there is no danger of needing a tree for the first block.
2386          */
2387         if (attr && S_ISDIR(attr->la_mode)) {
2388                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2389                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2390                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2391         }
2392
2393         if (!attr)
2394                 RETURN(0);
2395
2396         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2397                                    osd_dt_obj(dt), false, NULL, false);
2398         if (rc != 0)
2399                 RETURN(rc);
2400
2401         RETURN(rc);
2402 }
2403
2404 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2405                              struct lu_attr *attr,
2406                              struct dt_allocation_hint *hint,
2407                              struct dt_object_format *dof, struct thandle *th)
2408 {
2409         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2410         struct osd_object       *obj    = osd_dt_obj(dt);
2411         struct osd_thread_info  *info   = osd_oti_get(env);
2412         int result;
2413         ENTRY;
2414
2415         LINVRNT(osd_invariant(obj));
2416         LASSERT(!dt_object_exists(dt));
2417         LASSERT(!dt_object_remote(dt));
2418         LASSERT(osd_write_locked(env, obj));
2419         LASSERT(th != NULL);
2420
2421         if (unlikely(fid_is_acct(fid)))
2422                 /* Quota files can't be created from the kernel any more,
2423                  * 'tune2fs -O quota' will take care of creating them */
2424                 RETURN(-EPERM);
2425
2426         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2427         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2428
2429         result = __osd_object_create(info, obj, attr, hint, dof, th);
2430         if (result == 0)
2431                 result = __osd_oi_insert(env, obj, fid, th);
2432
2433         LASSERT(ergo(result == 0,
2434                      dt_object_exists(dt) && !dt_object_remote(dt)));
2435
2436         LASSERT(osd_invariant(obj));
2437         RETURN(result);
2438 }
2439
2440 /**
2441  * Called to destroy on-disk representation of the object
2442  *
2443  * Concurrency: must be locked
2444  */
2445 static int osd_declare_object_destroy(const struct lu_env *env,
2446                                       struct dt_object *dt,
2447                                       struct thandle *th)
2448 {
2449         struct osd_object  *obj = osd_dt_obj(dt);
2450         struct inode       *inode = obj->oo_inode;
2451         struct osd_thandle *oh;
2452         int                 rc;
2453         ENTRY;
2454
2455         oh = container_of0(th, struct osd_thandle, ot_super);
2456         LASSERT(oh->ot_handle == NULL);
2457         LASSERT(inode);
2458
2459         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2460                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2461         /* Recycle idle OI leaf may cause additional three OI blocks
2462          * to be changed. */
2463         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2464                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2465         /* one less inode */
2466         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2467                                    -1, oh, obj, false, NULL, false);
2468         if (rc)
2469                 RETURN(rc);
2470         /* data to be truncated */
2471         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2472                                    0, oh, obj, true, NULL, false);
2473         RETURN(rc);
2474 }
2475
2476 static int osd_object_destroy(const struct lu_env *env,
2477                               struct dt_object *dt,
2478                               struct thandle *th)
2479 {
2480         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2481         struct osd_object      *obj = osd_dt_obj(dt);
2482         struct inode           *inode = obj->oo_inode;
2483         struct osd_device      *osd = osd_obj2dev(obj);
2484         struct osd_thandle     *oh;
2485         int                     result;
2486         ENTRY;
2487
2488         oh = container_of0(th, struct osd_thandle, ot_super);
2489         LASSERT(oh->ot_handle);
2490         LASSERT(inode);
2491         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2492
2493         if (unlikely(fid_is_acct(fid)))
2494                 RETURN(-EPERM);
2495
2496         if (S_ISDIR(inode->i_mode)) {
2497                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2498                         inode->i_nlink == 2);
2499                 /* it will check/delete the inode from remote parent,
2500                  * how to optimize it? unlink performance impaction XXX */
2501                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2502                 if (result != 0 && result != -ENOENT) {
2503                         CERROR("%s: delete inode "DFID": rc = %d\n",
2504                                osd_name(osd), PFID(fid), result);
2505                 }
2506                 spin_lock(&obj->oo_guard);
2507                 clear_nlink(inode);
2508                 spin_unlock(&obj->oo_guard);
2509                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2510         }
2511
2512         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2513
2514         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2515                                OI_CHECK_FLD);
2516
2517         /* XXX: add to ext3 orphan list */
2518         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2519
2520         /* not needed in the cache anymore */
2521         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2522
2523         RETURN(0);
2524 }
2525
2526 /**
2527  * Put the fid into lustre_mdt_attrs, and then place the structure
2528  * inode's ea. This fid should not be altered during the life time
2529  * of the inode.
2530  *
2531  * \retval +ve, on success
2532  * \retval -ve, on error
2533  *
2534  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2535  */
2536 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2537                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2538 {
2539         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2540         int                      rc;
2541         ENTRY;
2542
2543         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2544                 RETURN(0);
2545
2546         lustre_lma_init(lma, fid, compat, incompat);
2547         lustre_lma_swab(lma);
2548
2549         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2550                              XATTR_CREATE);
2551         /* LMA may already exist, but we need to check that all the
2552          * desired compat/incompat flags have been added. */
2553         if (unlikely(rc == -EEXIST)) {
2554                 if (compat == 0 && incompat == 0)
2555                         RETURN(0);
2556
2557                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2558                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2559                                      LMA_OLD_SIZE);
2560                 if (rc <= 0)
2561                         RETURN(-EINVAL);
2562
2563                 lustre_lma_swab(lma);
2564                 if (!(~lma->lma_compat & compat) &&
2565                     !(~lma->lma_incompat & incompat))
2566                         RETURN(0);
2567
2568                 lma->lma_compat |= compat;
2569                 lma->lma_incompat |= incompat;
2570                 lustre_lma_swab(lma);
2571                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2572                                      sizeof(*lma), XATTR_REPLACE);
2573         }
2574
2575         RETURN(rc);
2576 }
2577
2578 /**
2579  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2580  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2581  * To have compatilibility with 1.8 ldiskfs driver we need to have
2582  * magic number at start of fid data.
2583  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2584  * its inmemory API.
2585  */
2586 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2587                                   const struct lu_fid *fid)
2588 {
2589         if (!fid_is_namespace_visible(fid) ||
2590             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2591                 param->edp_magic = 0;
2592                 return;
2593         }
2594
2595         param->edp_magic = LDISKFS_LUFID_MAGIC;
2596         param->edp_len =  sizeof(struct lu_fid) + 1;
2597         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2598 }
2599
2600 /**
2601  * Try to read the fid from inode ea into dt_rec.
2602  *
2603  * \param fid object fid.
2604  *
2605  * \retval 0 on success
2606  */
2607 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2608                           __u32 ino, struct lu_fid *fid,
2609                           struct osd_inode_id *id)
2610 {
2611         struct osd_thread_info *info  = osd_oti_get(env);
2612         struct inode           *inode;
2613         ENTRY;
2614
2615         osd_id_gen(id, ino, OSD_OII_NOGEN);
2616         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2617         if (IS_ERR(inode))
2618                 RETURN(PTR_ERR(inode));
2619
2620         iput(inode);
2621         RETURN(0);
2622 }
2623
2624 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2625                                         struct inode *dir,
2626                                         struct inode *parent_dir,
2627                                         const struct lu_fid *dot_fid,
2628                                         const struct lu_fid *dot_dot_fid,
2629                                         struct osd_thandle *oth)
2630 {
2631         struct ldiskfs_dentry_param *dot_ldp;
2632         struct ldiskfs_dentry_param *dot_dot_ldp;
2633
2634         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2635         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2636
2637         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2638         dot_ldp->edp_magic = 0;
2639         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2640                                         dir, dot_ldp, dot_dot_ldp);
2641 }
2642
2643 /**
2644  * Create an local agent inode for remote entry
2645  */
2646 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2647                                                   struct osd_device *osd,
2648                                                   struct osd_object *pobj,
2649                                                   const struct lu_fid *fid,
2650                                                   __u32 type,
2651                                                   struct thandle *th)
2652 {
2653         struct osd_thread_info  *info = osd_oti_get(env);
2654         struct inode            *local;
2655         struct osd_thandle      *oh;
2656         int                     rc;
2657         ENTRY;
2658
2659         LASSERT(th);
2660         oh = container_of(th, struct osd_thandle, ot_super);
2661         LASSERT(oh->ot_handle->h_transaction != NULL);
2662
2663         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2664         if (IS_ERR(local)) {
2665                 CERROR("%s: create local error %d\n", osd_name(osd),
2666                        (int)PTR_ERR(local));
2667                 RETURN(local);
2668         }
2669
2670         /* Set special LMA flag for local agent inode */
2671         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2672         if (rc != 0) {
2673                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2674                        osd_name(osd), PFID(fid), rc);
2675                 RETURN(ERR_PTR(rc));
2676         }
2677
2678         if (!S_ISDIR(type))
2679                 RETURN(local);
2680
2681         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2682                                          lu_object_fid(&pobj->oo_dt.do_lu),
2683                                          fid, oh);
2684         if (rc != 0) {
2685                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2686                         osd_name(osd), PFID(fid), rc);
2687                 RETURN(ERR_PTR(rc));
2688         }
2689
2690         RETURN(local);
2691 }
2692
2693 /**
2694  * Delete local agent inode for remote entry
2695  */
2696 static int osd_delete_local_agent_inode(const struct lu_env *env,
2697                                         struct osd_device *osd,
2698                                         const struct lu_fid *fid,
2699                                         __u32 ino, struct osd_thandle *oh)
2700 {
2701         struct osd_thread_info  *oti = osd_oti_get(env);
2702         struct osd_inode_id     *id = &oti->oti_id;
2703         struct inode            *inode;
2704         ENTRY;
2705
2706         id->oii_ino = le32_to_cpu(ino);
2707         id->oii_gen = OSD_OII_NOGEN;
2708         inode = osd_iget(oti, osd, id);
2709         if (IS_ERR(inode)) {
2710                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2711                        PFID(fid), id->oii_ino, id->oii_gen);
2712                 RETURN(PTR_ERR(inode));
2713         }
2714
2715         clear_nlink(inode);
2716         mark_inode_dirty(inode);
2717         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2718                 osd_name(osd), PFID(fid), inode->i_ino);
2719         iput(inode);
2720         RETURN(0);
2721 }
2722
2723 /**
2724  * OSD layer object create function for interoperability mode (b11826).
2725  * This is mostly similar to osd_object_create(). Only difference being, fid is
2726  * inserted into inode ea here.
2727  *
2728  * \retval   0, on success
2729  * \retval -ve, on error
2730  */
2731 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2732                                 struct lu_attr *attr,
2733                                 struct dt_allocation_hint *hint,
2734                                 struct dt_object_format *dof,
2735                                 struct thandle *th)
2736 {
2737         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2738         struct osd_object       *obj    = osd_dt_obj(dt);
2739         struct osd_thread_info  *info   = osd_oti_get(env);
2740         int                      result;
2741
2742         ENTRY;
2743
2744         LASSERT(osd_invariant(obj));
2745         LASSERT(!dt_object_exists(dt));
2746         LASSERT(!dt_object_remote(dt));
2747         LASSERT(osd_write_locked(env, obj));
2748         LASSERT(th != NULL);
2749
2750         if (unlikely(fid_is_acct(fid)))
2751                 /* Quota files can't be created from the kernel any more,
2752                  * 'tune2fs -O quota' will take care of creating them */
2753                 RETURN(-EPERM);
2754
2755         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2756         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2757
2758         result = __osd_object_create(info, obj, attr, hint, dof, th);
2759         if (result == 0)
2760                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2761                                 fid_is_on_ost(info, osd_obj2dev(obj),
2762                                               fid, OI_CHECK_FLD) ?
2763                                 LMAC_FID_ON_OST : 0, 0);
2764
2765         if (result == 0)
2766                 result = __osd_oi_insert(env, obj, fid, th);
2767
2768         LASSERT(ergo(result == 0,
2769                      dt_object_exists(dt) && !dt_object_remote(dt)));
2770         LINVRNT(osd_invariant(obj));
2771         RETURN(result);
2772 }
2773
2774 static int osd_declare_object_ref_add(const struct lu_env *env,
2775                                       struct dt_object *dt,
2776                                       struct thandle *handle)
2777 {
2778         struct osd_thandle       *oh;
2779
2780         /* it's possible that object doesn't exist yet */
2781         LASSERT(handle != NULL);
2782
2783         oh = container_of0(handle, struct osd_thandle, ot_super);
2784         LASSERT(oh->ot_handle == NULL);
2785
2786         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2787                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2788
2789         return 0;
2790 }
2791
2792 /*
2793  * Concurrency: @dt is write locked.
2794  */
2795 static int osd_object_ref_add(const struct lu_env *env,
2796                               struct dt_object *dt, struct thandle *th)
2797 {
2798         struct osd_object  *obj = osd_dt_obj(dt);
2799         struct inode       *inode = obj->oo_inode;
2800         struct osd_thandle *oh;
2801         int                 rc = 0;
2802
2803         LINVRNT(osd_invariant(obj));
2804         LASSERT(dt_object_exists(dt));
2805         LASSERT(!dt_object_remote(dt));
2806         LASSERT(osd_write_locked(env, obj));
2807         LASSERT(th != NULL);
2808
2809         oh = container_of0(th, struct osd_thandle, ot_super);
2810         LASSERT(oh->ot_handle != NULL);
2811
2812         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2813
2814         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2815                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2816         /*
2817          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2818          * (65000) subdirectories by storing "1" in i_nlink if the link count
2819          * would otherwise overflow. Directory tranversal tools understand
2820          * that (st_nlink == 1) indicates that the filesystem dose not track
2821          * hard links count on the directory, and will not abort subdirectory
2822          * scanning early once (st_nlink - 2) subdirs have been found.
2823          *
2824          * This also has to properly handle the case of inodes with nlink == 0
2825          * in case they are being linked into the PENDING directory
2826          */
2827         spin_lock(&obj->oo_guard);
2828         if (unlikely(inode->i_nlink == 0))
2829                 /* inc_nlink from 0 may cause WARN_ON */
2830                 set_nlink(inode, 1);
2831         else {
2832                 ldiskfs_inc_count(oh->ot_handle, inode);
2833                 if (!S_ISDIR(inode->i_mode))
2834                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2835         }
2836         spin_unlock(&obj->oo_guard);
2837
2838         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2839         LINVRNT(osd_invariant(obj));
2840
2841         return rc;
2842 }
2843
2844 static int osd_declare_object_ref_del(const struct lu_env *env,
2845                                       struct dt_object *dt,
2846                                       struct thandle *handle)
2847 {
2848         struct osd_thandle *oh;
2849
2850         LASSERT(dt_object_exists(dt));
2851         LASSERT(!dt_object_remote(dt));
2852         LASSERT(handle != NULL);
2853
2854         oh = container_of0(handle, struct osd_thandle, ot_super);
2855         LASSERT(oh->ot_handle == NULL);
2856
2857         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2858                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2859
2860         return 0;
2861 }
2862
2863 /*
2864  * Concurrency: @dt is write locked.
2865  */
2866 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2867                               struct thandle *th)
2868 {
2869         struct osd_object       *obj = osd_dt_obj(dt);
2870         struct inode            *inode = obj->oo_inode;
2871         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2872         struct osd_thandle      *oh;
2873
2874         LINVRNT(osd_invariant(obj));
2875         LASSERT(dt_object_exists(dt));
2876         LASSERT(!dt_object_remote(dt));
2877         LASSERT(osd_write_locked(env, obj));
2878         LASSERT(th != NULL);
2879
2880         oh = container_of0(th, struct osd_thandle, ot_super);
2881         LASSERT(oh->ot_handle != NULL);
2882
2883         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2884
2885         spin_lock(&obj->oo_guard);
2886         /* That can be result of upgrade from old Lustre version and
2887          * applied only to local files.  Just skip this ref_del call.
2888          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2889         if (inode->i_nlink == 0) {
2890                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2891                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2892                              ", maybe an upgraded file? (LU-3915)\n",
2893                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2894                 spin_unlock(&obj->oo_guard);
2895                 return 0;
2896         }
2897
2898         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2899                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2900
2901         ldiskfs_dec_count(oh->ot_handle, inode);
2902         spin_unlock(&obj->oo_guard);
2903
2904         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2905         LINVRNT(osd_invariant(obj));
2906
2907         return 0;
2908 }
2909
2910 /*
2911  * Get the 64-bit version for an inode.
2912  */
2913 static int osd_object_version_get(const struct lu_env *env,
2914                                   struct dt_object *dt, dt_obj_version_t *ver)
2915 {
2916         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2917
2918         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2919                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2920         *ver = LDISKFS_I(inode)->i_fs_version;
2921         return 0;
2922 }
2923
2924 /*
2925  * Concurrency: @dt is read locked.
2926  */
2927 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2928                          struct lu_buf *buf, const char *name,
2929                          struct lustre_capa *capa)
2930 {
2931         struct osd_object      *obj    = osd_dt_obj(dt);
2932         struct inode           *inode  = obj->oo_inode;
2933         struct osd_thread_info *info   = osd_oti_get(env);
2934         struct dentry          *dentry = &info->oti_obj_dentry;
2935
2936         /* version get is not real XATTR but uses xattr API */
2937         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2938                 /* for version we are just using xattr API but change inode
2939                  * field instead */
2940                 if (buf->lb_len == 0)
2941                         return sizeof(dt_obj_version_t);
2942
2943                 if (buf->lb_len < sizeof(dt_obj_version_t))
2944                         return -ERANGE;
2945
2946                 osd_object_version_get(env, dt, buf->lb_buf);
2947
2948                 return sizeof(dt_obj_version_t);
2949         }
2950
2951         LASSERT(dt_object_exists(dt));
2952         LASSERT(!dt_object_remote(dt));
2953         LASSERT(inode->i_op != NULL);
2954         LASSERT(inode->i_op->getxattr != NULL);
2955
2956         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2957                 return -EACCES;
2958
2959         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2960 }
2961
2962
2963 static int osd_declare_xattr_set(const struct lu_env *env,
2964                                  struct dt_object *dt,
2965                                  const struct lu_buf *buf, const char *name,
2966                                  int fl, struct thandle *handle)
2967 {
2968         struct osd_thandle *oh;
2969         int credits;
2970         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2971
2972         LASSERT(handle != NULL);
2973
2974         oh = container_of0(handle, struct osd_thandle, ot_super);
2975         LASSERT(oh->ot_handle == NULL);
2976
2977         /* optimistic optimization: LMA is set first and usually fit inode */
2978         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2979                 if (dt_object_exists(dt))
2980                         credits = 0;
2981                 else
2982                         credits = 1;
2983         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2984                 credits = 1;
2985         } else {
2986                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2987                 if (buf && buf->lb_len > sb->s_blocksize) {
2988                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2989                                         sb->s_blocksize_bits;
2990                 }
2991                 /*
2992                  * xattr set may involve inode quota change, reserve credits for
2993                  * dquot_initialize()
2994                  */
2995                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2996         }
2997
2998         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2999
3000         return 0;
3001 }
3002
3003 /*
3004  * Set the 64-bit version for object
3005  */
3006 static void osd_object_version_set(const struct lu_env *env,
3007                                    struct dt_object *dt,
3008                                    dt_obj_version_t *new_version)
3009 {
3010         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3011
3012         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3013                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3014
3015         LDISKFS_I(inode)->i_fs_version = *new_version;
3016         /** Version is set after all inode operations are finished,
3017          *  so we should mark it dirty here */
3018         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3019 }
3020
3021 /*
3022  * Concurrency: @dt is write locked.
3023  */
3024 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3025                          const struct lu_buf *buf, const char *name, int fl,
3026                          struct thandle *handle, struct lustre_capa *capa)
3027 {
3028         struct osd_object      *obj      = osd_dt_obj(dt);
3029         struct inode           *inode    = obj->oo_inode;
3030         struct osd_thread_info *info     = osd_oti_get(env);
3031         int                     fs_flags = 0;
3032         ENTRY;
3033
3034         LASSERT(handle != NULL);
3035
3036         /* version set is not real XATTR */
3037         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3038                 /* for version we are just using xattr API but change inode
3039                  * field instead */
3040                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3041                 osd_object_version_set(env, dt, buf->lb_buf);
3042                 return sizeof(dt_obj_version_t);
3043         }
3044
3045         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3046                 return -EACCES;
3047
3048         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3049                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3050
3051         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3052         if (fl & LU_XATTR_REPLACE)
3053                 fs_flags |= XATTR_REPLACE;
3054
3055         if (fl & LU_XATTR_CREATE)
3056                 fs_flags |= XATTR_CREATE;
3057
3058         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3059                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3060                 int                      rc;
3061
3062                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3063                 if (rc != 0)
3064                         RETURN(rc);
3065
3066                 lma->lma_incompat |= LMAI_STRIPED;
3067                 lustre_lma_swab(lma);
3068                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3069                                      sizeof(*lma), XATTR_REPLACE);
3070                 if (rc != 0)
3071                         RETURN(rc);
3072         }
3073
3074         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3075             strcmp(name, XATTR_NAME_LINK) == 0)
3076                 return -ENOSPC;
3077
3078         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3079                                fs_flags);
3080 }
3081
3082 /*
3083  * Concurrency: @dt is read locked.
3084  */
3085 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3086                           struct lu_buf *buf, struct lustre_capa *capa)
3087 {
3088         struct osd_object      *obj    = osd_dt_obj(dt);
3089         struct inode           *inode  = obj->oo_inode;
3090         struct osd_thread_info *info   = osd_oti_get(env);
3091         struct dentry          *dentry = &info->oti_obj_dentry;
3092
3093         LASSERT(dt_object_exists(dt));
3094         LASSERT(!dt_object_remote(dt));
3095         LASSERT(inode->i_op != NULL);
3096         LASSERT(inode->i_op->listxattr != NULL);
3097
3098         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3099                 return -EACCES;
3100
3101         dentry->d_inode = inode;
3102         dentry->d_sb = inode->i_sb;
3103         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3104 }
3105
3106 static int osd_declare_xattr_del(const struct lu_env *env,
3107                                  struct dt_object *dt, const char *name,
3108                                  struct thandle *handle)
3109 {
3110         struct osd_thandle *oh;
3111         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3112
3113         LASSERT(dt_object_exists(dt));
3114         LASSERT(!dt_object_remote(dt));
3115         LASSERT(handle != NULL);
3116
3117         oh = container_of0(handle, struct osd_thandle, ot_super);
3118         LASSERT(oh->ot_handle == NULL);
3119
3120         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3121                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3122         /*
3123          * xattr del may involve inode quota change, reserve credits for
3124          * dquot_initialize()
3125          */
3126         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3127
3128         return 0;
3129 }
3130
3131 /*
3132  * Concurrency: @dt is write locked.
3133  */
3134 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3135                          const char *name, struct thandle *handle,
3136                          struct lustre_capa *capa)
3137 {
3138         struct osd_object      *obj    = osd_dt_obj(dt);
3139         struct inode           *inode  = obj->oo_inode;
3140         struct osd_thread_info *info   = osd_oti_get(env);
3141         struct dentry          *dentry = &info->oti_obj_dentry;
3142         int                     rc;
3143
3144         LASSERT(dt_object_exists(dt));
3145         LASSERT(!dt_object_remote(dt));
3146         LASSERT(inode->i_op != NULL);
3147         LASSERT(inode->i_op->removexattr != NULL);
3148         LASSERT(handle != NULL);
3149
3150         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3151                 return -EACCES;
3152
3153         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3154
3155         ll_vfs_dq_init(inode);
3156         dentry->d_inode = inode;
3157         dentry->d_sb = inode->i_sb;
3158         rc = inode->i_op->removexattr(dentry, name);
3159         return rc;
3160 }
3161
3162 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3163                                      struct dt_object *dt,
3164                                      struct lustre_capa *old, __u64 opc)
3165 {
3166         struct osd_thread_info *info = osd_oti_get(env);
3167         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3168         struct osd_object *obj = osd_dt_obj(dt);
3169         struct osd_device *osd = osd_obj2dev(obj);
3170         struct lustre_capa_key *key = &info->oti_capa_key;
3171         struct lustre_capa *capa = &info->oti_capa;
3172         struct obd_capa *oc;
3173         struct lu_capainfo *lci;
3174         int rc;
3175         ENTRY;
3176
3177         if (!osd->od_fl_capa)
3178                 RETURN(ERR_PTR(-ENOENT));
3179
3180         LASSERT(dt_object_exists(dt));
3181         LASSERT(!dt_object_remote(dt));
3182         LINVRNT(osd_invariant(obj));
3183
3184         /* renewal sanity check */
3185         if (old && osd_object_auth(env, dt, old, opc))
3186                 RETURN(ERR_PTR(-EACCES));
3187
3188         lci = lu_capainfo_get(env);
3189         if (unlikely(lci == NULL))
3190                 RETURN(ERR_PTR(-ENOENT));
3191
3192         switch (lci->lci_auth) {
3193         case LC_ID_NONE:
3194                 RETURN(NULL);
3195         case LC_ID_PLAIN:
3196                 capa->lc_uid = i_uid_read(obj->oo_inode);
3197                 capa->lc_gid = i_gid_read(obj->oo_inode);
3198                 capa->lc_flags = LC_ID_PLAIN;
3199                 break;
3200         case LC_ID_CONVERT: {
3201                 __u32 d[4], s[4];
3202
3203                 s[0] = i_uid_read(obj->oo_inode);
3204                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3205                 s[2] = i_uid_read(obj->oo_inode);
3206                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3207                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3208                 if (unlikely(rc))
3209                         RETURN(ERR_PTR(rc));
3210
3211                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3212                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3213                 capa->lc_flags = LC_ID_CONVERT;
3214                 break;
3215         }
3216         default:
3217                 RETURN(ERR_PTR(-EINVAL));
3218         }
3219
3220         capa->lc_fid = *fid;
3221         capa->lc_opc = opc;
3222         capa->lc_flags |= osd->od_capa_alg << 24;
3223         capa->lc_timeout = osd->od_capa_timeout;
3224         capa->lc_expiry = 0;
3225
3226         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3227         if (oc) {
3228                 LASSERT(!capa_is_expired(oc));
3229                 RETURN(oc);
3230         }
3231
3232         spin_lock(&capa_lock);
3233         *key = osd->od_capa_keys[1];
3234         spin_unlock(&capa_lock);
3235
3236         capa->lc_keyid = key->lk_keyid;
3237         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3238
3239         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3240         if (rc) {
3241                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3242                 RETURN(ERR_PTR(rc));
3243         }
3244
3245         oc = capa_add(osd->od_capa_hash, capa);
3246         RETURN(oc);
3247 }
3248
3249 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3250                            __u64 start, __u64 end)
3251 {
3252         struct osd_object       *obj    = osd_dt_obj(dt);
3253         struct inode            *inode  = obj->oo_inode;
3254         struct osd_thread_info  *info   = osd_oti_get(env);
3255         struct dentry           *dentry = &info->oti_obj_dentry;
3256         struct file             *file   = &info->oti_file;
3257         int                     rc;
3258
3259         ENTRY;
3260
3261         dentry->d_inode = inode;
3262         dentry->d_sb = inode->i_sb;
3263         file->f_dentry = dentry;
3264         file->f_mapping = inode->i_mapping;
3265         file->f_op = inode->i_fop;
3266         set_file_inode(file, inode);
3267
3268 #ifdef HAVE_FILE_FSYNC_4ARGS
3269         rc = file->f_op->fsync(file, start, end, 0);
3270 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3271         mutex_lock(&inode->i_mutex);
3272         rc = file->f_op->fsync(file, 0);
3273         mutex_unlock(&inode->i_mutex);
3274 #else
3275         mutex_lock(&inode->i_mutex);
3276         rc = file->f_op->fsync(file, dentry, 0);
3277         mutex_unlock(&inode->i_mutex);
3278 #endif
3279
3280         RETURN(rc);
3281 }
3282
3283 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3284                         void **data)
3285 {
3286         struct osd_object *obj = osd_dt_obj(dt);
3287         ENTRY;
3288
3289         *data = (void *)obj->oo_inode;
3290         RETURN(0);
3291 }
3292
3293 /*
3294  * Index operations.
3295  */
3296
3297 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3298                            const struct dt_index_features *feat)
3299 {
3300         struct iam_descr *descr;
3301
3302         if (osd_object_is_root(o))
3303                 return feat == &dt_directory_features;
3304
3305         LASSERT(o->oo_dir != NULL);
3306
3307         descr = o->oo_dir->od_container.ic_descr;
3308         if (feat == &dt_directory_features) {
3309                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3310                         return 1;
3311                 else
3312                         return 0;
3313         } else {
3314                 return
3315                         feat->dif_keysize_min <= descr->id_key_size &&
3316                         descr->id_key_size <= feat->dif_keysize_max &&
3317                         feat->dif_recsize_min <= descr->id_rec_size &&
3318                         descr->id_rec_size <= feat->dif_recsize_max &&
3319                         !(feat->dif_flags & (DT_IND_VARKEY |
3320                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3321                         ergo(feat->dif_flags & DT_IND_UPDATE,
3322                              1 /* XXX check that object (and file system) is
3323                                 * writable */);
3324         }
3325 }
3326
3327 static int osd_iam_container_init(const struct lu_env *env,
3328                                   struct osd_object *obj,
3329                                   struct osd_directory *dir)
3330 {
3331         struct iam_container *bag = &dir->od_container;
3332         int result;
3333
3334         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3335         if (result != 0)
3336                 return result;
3337
3338         result = iam_container_setup(bag);
3339         if (result == 0)
3340                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3341         else
3342                 iam_container_fini(bag);
3343
3344         return result;
3345 }
3346
3347
3348 /*
3349  * Concurrency: no external locking is necessary.
3350  */
3351 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3352                          const struct dt_index_features *feat)
3353 {
3354         int                      result;
3355         int                      skip_iam = 0;
3356         struct osd_object       *obj = osd_dt_obj(dt);
3357
3358         LINVRNT(osd_invariant(obj));
3359
3360         if (osd_object_is_root(obj)) {
3361                 dt->do_index_ops = &osd_index_ea_ops;
3362                 result = 0;
3363         } else if (feat == &dt_directory_features) {
3364                 dt->do_index_ops = &osd_index_ea_ops;
3365                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3366                         result = 0;
3367                 else
3368                         result = -ENOTDIR;
3369                 skip_iam = 1;
3370         } else if (unlikely(feat == &dt_otable_features)) {
3371                 dt->do_index_ops = &osd_otable_ops;
3372                 return 0;
3373         } else if (unlikely(feat == &dt_acct_features)) {
3374                 dt->do_index_ops = &osd_acct_index_ops;
3375                 result = 0;
3376                 skip_iam = 1;
3377         } else if (!osd_has_index(obj)) {
3378                 struct osd_directory *dir;
3379
3380                 OBD_ALLOC_PTR(dir);
3381                 if (dir != NULL) {
3382
3383                         spin_lock(&obj->oo_guard);
3384                         if (obj->oo_dir == NULL)
3385                                 obj->oo_dir = dir;
3386                         else
3387                                 /*
3388                                  * Concurrent thread allocated container data.
3389                                  */
3390                                 OBD_FREE_PTR(dir);
3391                         spin_unlock(&obj->oo_guard);
3392                         /*
3393                          * Now, that we have container data, serialize its
3394                          * initialization.
3395                          */
3396                         down_write(&obj->oo_ext_idx_sem);
3397                         /*
3398                          * recheck under lock.
3399                          */
3400                         if (!osd_has_index(obj))
3401                                 result = osd_iam_container_init(env, obj,
3402                                                                 obj->oo_dir);
3403                         else
3404                                 result = 0;
3405                         up_write(&obj->oo_ext_idx_sem);
3406                 } else {
3407                         result = -ENOMEM;
3408                 }
3409         } else {
3410                 result = 0;
3411         }
3412
3413         if (result == 0 && skip_iam == 0) {
3414                 if (!osd_iam_index_probe(env, obj, feat))
3415                         result = -ENOTDIR;
3416         }
3417         LINVRNT(osd_invariant(obj));
3418
3419         if (result == 0 && is_quota_glb_feat(feat) &&
3420             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3421                 result = osd_quota_migration(env, dt, feat);
3422
3423         return result;
3424 }
3425
3426 static int osd_otable_it_attr_get(const struct lu_env *env,
3427                                  struct dt_object *dt,
3428                                  struct lu_attr *attr,
3429                                  struct lustre_capa *capa)
3430 {
3431         attr->la_valid = 0;
3432         return 0;
3433 }
3434
3435 static const struct dt_object_operations osd_obj_ops = {
3436         .do_read_lock         = osd_object_read_lock,
3437         .do_write_lock        = osd_object_write_lock,
3438         .do_read_unlock       = osd_object_read_unlock,
3439         .do_write_unlock      = osd_object_write_unlock,
3440         .do_write_locked      = osd_object_write_locked,
3441         .do_attr_get          = osd_attr_get,
3442         .do_declare_attr_set  = osd_declare_attr_set,
3443         .do_attr_set          = osd_attr_set,
3444         .do_ah_init           = osd_ah_init,
3445         .do_declare_create    = osd_declare_object_create,
3446         .do_create            = osd_object_create,
3447         .do_declare_destroy   = osd_declare_object_destroy,
3448         .do_destroy           = osd_object_destroy,
3449         .do_index_try         = osd_index_try,
3450         .do_declare_ref_add   = osd_declare_object_ref_add,
3451         .do_ref_add           = osd_object_ref_add,
3452         .do_declare_ref_del   = osd_declare_object_ref_del,
3453         .do_ref_del           = osd_object_ref_del,
3454         .do_xattr_get         = osd_xattr_get,
3455         .do_declare_xattr_set = osd_declare_xattr_set,
3456         .do_xattr_set         = osd_xattr_set,
3457         .do_declare_xattr_del = osd_declare_xattr_del,
3458         .do_xattr_del         = osd_xattr_del,
3459         .do_xattr_list        = osd_xattr_list,
3460         .do_capa_get          = osd_capa_get,
3461         .do_object_sync       = osd_object_sync,
3462         .do_data_get          = osd_data_get,
3463 };
3464
3465 /**
3466  * dt_object_operations for interoperability mode
3467  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3468  */
3469 static const struct dt_object_operations osd_obj_ea_ops = {
3470         .do_read_lock         = osd_object_read_lock,
3471         .do_write_lock        = osd_object_write_lock,
3472         .do_read_unlock       = osd_object_read_unlock,
3473         .do_write_unlock      = osd_object_write_unlock,
3474         .do_write_locked      = osd_object_write_locked,
3475         .do_attr_get          = osd_attr_get,
3476         .do_declare_attr_set  = osd_declare_attr_set,
3477         .do_attr_set          = osd_attr_set,
3478         .do_ah_init           = osd_ah_init,
3479         .do_declare_create    = osd_declare_object_create,
3480         .do_create            = osd_object_ea_create,
3481         .do_declare_destroy   = osd_declare_object_destroy,
3482         .do_destroy           = osd_object_destroy,
3483         .do_index_try         = osd_index_try,
3484         .do_declare_ref_add   = osd_declare_object_ref_add,
3485         .do_ref_add           = osd_object_ref_add,
3486         .do_declare_ref_del   = osd_declare_object_ref_del,
3487         .do_ref_del           = osd_object_ref_del,
3488         .do_xattr_get         = osd_xattr_get,
3489         .do_declare_xattr_set = osd_declare_xattr_set,
3490         .do_xattr_set         = osd_xattr_set,
3491         .do_declare_xattr_del = osd_declare_xattr_del,
3492         .do_xattr_del         = osd_xattr_del,
3493         .do_xattr_list        = osd_xattr_list,
3494         .do_capa_get          = osd_capa_get,
3495         .do_object_sync       = osd_object_sync,
3496         .do_data_get          = osd_data_get,
3497 };
3498
3499 static const struct dt_object_operations osd_obj_otable_it_ops = {
3500         .do_attr_get    = osd_otable_it_attr_get,
3501         .do_index_try   = osd_index_try,
3502 };
3503
3504 static int osd_index_declare_iam_delete(const struct lu_env *env,
3505                                         struct dt_object *dt,
3506                                         const struct dt_key *key,
3507                                         struct thandle *handle)
3508 {
3509         struct osd_thandle    *oh;
3510
3511         oh = container_of0(handle, struct osd_thandle, ot_super);
3512         LASSERT(oh->ot_handle == NULL);
3513
3514         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3515                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3516
3517         return 0;
3518 }
3519
3520 /**
3521  *      delete a (key, value) pair from index \a dt specified by \a key
3522  *
3523  *      \param  dt      osd index object
3524  *      \param  key     key for index
3525  *      \param  rec     record reference
3526  *      \param  handle  transaction handler
3527  *
3528  *      \retval  0  success
3529  *      \retval -ve   failure
3530  */
3531 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3532                                 const struct dt_key *key,
3533                                 struct thandle *handle,
3534                                 struct lustre_capa *capa)
3535 {
3536         struct osd_thread_info *oti = osd_oti_get(env);
3537         struct osd_object      *obj = osd_dt_obj(dt);
3538         struct osd_thandle     *oh;
3539         struct iam_path_descr  *ipd;
3540         struct iam_container   *bag = &obj->oo_dir->od_container;
3541         int                     rc;
3542         ENTRY;
3543
3544         LINVRNT(osd_invariant(obj));
3545         LASSERT(dt_object_exists(dt));
3546         LASSERT(!dt_object_remote(dt));
3547         LASSERT(bag->ic_object == obj->oo_inode);
3548         LASSERT(handle != NULL);
3549
3550         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3551                 RETURN(-EACCES);
3552
3553         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3554
3555         ipd = osd_idx_ipd_get(env, bag);
3556         if (unlikely(ipd == NULL))
3557                 RETURN(-ENOMEM);
3558
3559         oh = container_of0(handle, struct osd_thandle, ot_super);
3560         LASSERT(oh->ot_handle != NULL);
3561         LASSERT(oh->ot_handle->h_transaction != NULL);
3562
3563         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3564                 /* swab quota uid/gid provided by caller */
3565                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3566                 key = (const struct dt_key *)&oti->oti_quota_id;
3567         }
3568
3569         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3570         osd_ipd_put(env, bag, ipd);
3571         LINVRNT(osd_invariant(obj));
3572         RETURN(rc);
3573 }
3574
3575 static int osd_index_declare_ea_delete(const struct lu_env *env,
3576                                        struct dt_object *dt,
3577                                        const struct dt_key *key,
3578                                        struct thandle *handle)
3579 {
3580         struct osd_thandle *oh;
3581         struct inode       *inode;
3582         int                 rc;
3583         ENTRY;
3584
3585         LASSERT(dt_object_exists(dt));
3586         LASSERT(!dt_object_remote(dt));
3587         LASSERT(handle != NULL);
3588
3589         oh = container_of0(handle, struct osd_thandle, ot_super);
3590         LASSERT(oh->ot_handle == NULL);
3591
3592         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3593                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3594
3595         inode = osd_dt_obj(dt)->oo_inode;
3596         LASSERT(inode);
3597
3598         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3599                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3600         RETURN(rc);
3601 }
3602
3603 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3604                                           struct dt_rec *fid)
3605 {
3606         struct osd_fid_pack *rec;
3607         int                  rc = -ENODATA;
3608
3609         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3610                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3611                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3612         }
3613         return rc;
3614 }
3615
3616 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3617                           const struct lu_fid *fid)
3618 {
3619         struct seq_server_site  *ss = osd_seq_site(osd);
3620         ENTRY;
3621
3622         /* FID seqs not in FLDB, must be local seq */
3623         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3624                 RETURN(0);
3625
3626         /* If FLD is not being initialized yet, it only happens during the
3627          * initialization, likely during mgs initialization, and we assume
3628          * this is local FID. */
3629         if (ss == NULL || ss->ss_server_fld == NULL)
3630                 RETURN(0);
3631
3632         /* Only check the local FLDB here */
3633         if (osd_seq_exists(env, osd, fid_seq(fid)))
3634                 RETURN(0);
3635
3636         RETURN(1);
3637 }
3638
3639 /**
3640  * Index delete function for interoperability mode (b11826).
3641  * It will remove the directory entry added by osd_index_ea_insert().
3642  * This entry is needed to maintain name->fid mapping.
3643  *
3644  * \param key,  key i.e. file entry to be deleted
3645  *
3646  * \retval   0, on success
3647  * \retval -ve, on error
3648  */
3649 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3650                                const struct dt_key *key, struct thandle *handle,
3651                                struct lustre_capa *capa)
3652 {
3653         struct osd_object          *obj = osd_dt_obj(dt);
3654         struct inode               *dir = obj->oo_inode;
3655         struct dentry              *dentry;
3656         struct osd_thandle         *oh;
3657         struct ldiskfs_dir_entry_2 *de = NULL;
3658         struct buffer_head         *bh;
3659         struct htree_lock          *hlock = NULL;
3660         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3661         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3662         int                        rc;
3663         ENTRY;
3664
3665         LINVRNT(osd_invariant(obj));
3666         LASSERT(dt_object_exists(dt));
3667         LASSERT(!dt_object_remote(dt));
3668         LASSERT(handle != NULL);
3669
3670         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3671
3672         oh = container_of(handle, struct osd_thandle, ot_super);
3673         LASSERT(oh->ot_handle != NULL);
3674         LASSERT(oh->ot_handle->h_transaction != NULL);
3675
3676         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3677                 RETURN(-EACCES);
3678
3679         ll_vfs_dq_init(dir);
3680         dentry = osd_child_dentry_get(env, obj,
3681                                       (char *)key, strlen((char *)key));
3682
3683         if (obj->oo_hl_head != NULL) {
3684                 hlock = osd_oti_get(env)->oti_hlock;
3685                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3686                                    dir, LDISKFS_HLOCK_DEL);
3687         } else {
3688                 down_write(&obj->oo_ext_idx_sem);
3689         }
3690
3691         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3692         if (bh) {
3693                 __u32 ino = 0;
3694
3695                 /* If this is not the ".." entry, it might be a remote DNE
3696                  * entry and  we need to check if the FID is for a remote
3697                  * MDT.  If the FID is  not in the directory entry (e.g.
3698                  * upgraded 1.8 filesystem without dirdata enabled) then
3699                  * we need to get the FID from the LMA. For a remote directory
3700                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3701                  * case.
3702                  *
3703                  * Delete the entry before the agent inode in order to
3704                  * simplify error handling.  At worst an error after deleting
3705                  * the entry first might leak the agent inode afterward. The
3706                  * reverse would need filesystem abort in case of error deleting
3707                  * the entry after the agent had been removed, or leave a
3708                  * dangling entry pointing at a random inode. */
3709                 if (strcmp((char *)key, dotdot) != 0) {
3710                         LASSERT(de != NULL);
3711                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3712                         /* If Fid is not in dentry, try to get it from LMA */
3713                         if (rc == -ENODATA) {
3714                                 struct osd_inode_id *id;
3715                                 struct inode *inode;
3716
3717                                 /* Before trying to get fid from the inode,
3718                                  * check whether the inode is valid.
3719                                  *
3720                                  * If the inode has been deleted, do not go
3721                                  * ahead to do osd_ea_fid_get, which will set
3722                                  * the inode to bad inode, which might cause
3723                                  * the inode to be deleted uncorrectly */
3724                                 inode = ldiskfs_iget(osd_sb(osd),
3725                                                      le32_to_cpu(de->inode));
3726                                 if (IS_ERR(inode)) {
3727                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3728                                                "error.\n", osd_name(osd),
3729                                                PFID(fid));
3730                                         rc = PTR_ERR(inode);
3731                                 } else {
3732                                         if (likely(inode->i_nlink != 0)) {
3733                                                 id = &osd_oti_get(env)->oti_id;
3734                                                 rc = osd_ea_fid_get(env, obj,
3735                                                         le32_to_cpu(de->inode),
3736                                                                     fid, id);
3737                                         } else {
3738                                                 CDEBUG(D_INFO, "%s: %u "DFID
3739                                                        "deleted.\n",
3740                                                        osd_name(osd),
3741                                                        le32_to_cpu(de->inode),
3742                                                        PFID(fid));
3743                                                 rc = -ESTALE;
3744                                         }
3745                                         iput(inode);
3746                                 }
3747                         }
3748                         if (rc == 0 &&
3749                             unlikely(osd_remote_fid(env, osd, fid)))
3750                                 /* Need to delete agent inode */
3751                                 ino = le32_to_cpu(de->inode);
3752                 }
3753                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3754                 brelse(bh);
3755                 if (rc == 0 && unlikely(ino != 0)) {
3756                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3757                                                           oh);
3758                         if (rc != 0)
3759                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3760                                        osd_name(osd), PFID(fid), rc);
3761                 }
3762         } else {
3763                 rc = -ENOENT;
3764         }
3765         if (hlock != NULL)
3766                 ldiskfs_htree_unlock(hlock);
3767         else
3768                 up_write(&obj->oo_ext_idx_sem);
3769
3770         if (rc != 0)
3771                 GOTO(out, rc);
3772
3773         /* For inode on the remote MDT, .. will point to
3774          * /Agent directory, Check whether it needs to delete
3775          * from agent directory */
3776         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3777                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3778                                                    oh);
3779                 if (rc != 0 && rc != -ENOENT) {
3780                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3781                                osd_name(osd), PFID(fid), rc);
3782                 }
3783
3784                 if (rc == -ENOENT)
3785                         rc = 0;
3786
3787                 GOTO(out, rc);
3788         }
3789 out:
3790
3791         LASSERT(osd_invariant(obj));
3792         RETURN(rc);
3793 }
3794
3795 /**
3796  *      Lookup index for \a key and copy record to \a rec.
3797  *
3798  *      \param  dt      osd index object
3799  *      \param  key     key for index
3800  *      \param  rec     record reference
3801  *
3802  *      \retval  +ve  success : exact mach
3803  *      \retval  0    return record with key not greater than \a key
3804  *      \retval -ve   failure
3805  */
3806 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3807                                 struct dt_rec *rec, const struct dt_key *key,
3808                                 struct lustre_capa *capa)
3809 {
3810         struct osd_object      *obj = osd_dt_obj(dt);
3811         struct iam_path_descr  *ipd;
3812         struct iam_container   *bag = &obj->oo_dir->od_container;
3813         struct osd_thread_info *oti = osd_oti_get(env);
3814         struct iam_iterator    *it = &oti->oti_idx_it;
3815         struct iam_rec         *iam_rec;
3816         int                     rc;
3817         ENTRY;
3818
3819         LASSERT(osd_invariant(obj));
3820         LASSERT(dt_object_exists(dt));
3821         LASSERT(!dt_object_remote(dt));
3822         LASSERT(bag->ic_object == obj->oo_inode);
3823
3824         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3825                 RETURN(-EACCES);
3826
3827         ipd = osd_idx_ipd_get(env, bag);
3828         if (IS_ERR(ipd))
3829                 RETURN(-ENOMEM);
3830
3831         /* got ipd now we can start iterator. */
3832         iam_it_init(it, bag, 0, ipd);
3833
3834         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3835                 /* swab quota uid/gid provided by caller */
3836                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3837                 key = (const struct dt_key *)&oti->oti_quota_id;
3838         }
3839
3840         rc = iam_it_get(it, (struct iam_key *)key);
3841         if (rc >= 0) {
3842                 if (S_ISDIR(obj->oo_inode->i_mode))
3843                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3844                 else
3845                         iam_rec = (struct iam_rec *) rec;
3846
3847                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3848
3849                 if (S_ISDIR(obj->oo_inode->i_mode))
3850                         osd_fid_unpack((struct lu_fid *) rec,
3851                                        (struct osd_fid_pack *)iam_rec);
3852                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3853                         osd_quota_unpack(obj, rec);
3854         }
3855
3856         iam_it_put(it);
3857         iam_it_fini(it);
3858         osd_ipd_put(env, bag, ipd);
3859
3860         LINVRNT(osd_invariant(obj));
3861
3862         RETURN(rc);
3863 }
3864
3865 static int osd_index_declare_iam_insert(const struct lu_env *env,
3866                                         struct dt_object *dt,
3867                                         const struct dt_rec *rec,
3868                                         const struct dt_key *key,
3869                                         struct thandle *handle)
3870 {
3871         struct osd_thandle *oh;
3872
3873         LASSERT(handle != NULL);
3874
3875         oh = container_of0(handle, struct osd_thandle, ot_super);
3876         LASSERT(oh->ot_handle == NULL);
3877
3878         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3879                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3880
3881         return 0;
3882 }
3883
3884 /**
3885  *      Inserts (key, value) pair in \a dt index object.
3886  *
3887  *      \param  dt      osd index object
3888  *      \param  key     key for index
3889  *      \param  rec     record reference
3890  *      \param  th      transaction handler
3891  *
3892  *      \retval  0  success
3893  *      \retval -ve failure
3894  */
3895 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3896                                 const struct dt_rec *rec,
3897                                 const struct dt_key *key, struct thandle *th,
3898                                 struct lustre_capa *capa, int ignore_quota)
3899 {
3900         struct osd_object     *obj = osd_dt_obj(dt);
3901         struct iam_path_descr *ipd;
3902         struct osd_thandle    *oh;
3903         struct iam_container  *bag = &obj->oo_dir->od_container;
3904         struct osd_thread_info *oti = osd_oti_get(env);
3905         struct iam_rec         *iam_rec;
3906         int                     rc;
3907         ENTRY;
3908
3909         LINVRNT(osd_invariant(obj));
3910         LASSERT(dt_object_exists(dt));
3911         LASSERT(!dt_object_remote(dt));
3912         LASSERT(bag->ic_object == obj->oo_inode);
3913         LASSERT(th != NULL);
3914
3915         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3916                 RETURN(-EACCES);
3917
3918         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3919
3920         ipd = osd_idx_ipd_get(env, bag);
3921         if (unlikely(ipd == NULL))
3922                 RETURN(-ENOMEM);
3923
3924         oh = container_of0(th, struct osd_thandle, ot_super);
3925         LASSERT(oh->ot_handle != NULL);
3926         LASSERT(oh->ot_handle->h_transaction != NULL);
3927         if (S_ISDIR(obj->oo_inode->i_mode)) {
3928                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3929                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3930         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3931                 /* pack quota uid/gid */
3932                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3933                 key = (const struct dt_key *)&oti->oti_quota_id;
3934                 /* pack quota record */
3935                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3936                 iam_rec = (struct iam_rec *)rec;
3937         } else {
3938                 iam_rec = (struct iam_rec *)rec;
3939         }
3940
3941         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3942                         iam_rec, ipd);
3943         osd_ipd_put(env, bag, ipd);
3944         LINVRNT(osd_invariant(obj));
3945         RETURN(rc);
3946 }
3947
3948 /**
3949  * Calls ldiskfs_add_entry() to add directory entry
3950  * into the directory. This is required for
3951  * interoperability mode (b11826)
3952  *
3953  * \retval   0, on success
3954  * \retval -ve, on error
3955  */
3956 static int __osd_ea_add_rec(struct osd_thread_info *info,
3957                             struct osd_object *pobj, struct inode  *cinode,
3958                             const char *name, const struct lu_fid *fid,
3959                             struct htree_lock *hlock, struct thandle *th)
3960 {
3961         struct ldiskfs_dentry_param *ldp;
3962         struct dentry               *child;
3963         struct osd_thandle          *oth;
3964         int                          rc;
3965
3966         oth = container_of(th, struct osd_thandle, ot_super);
3967         LASSERT(oth->ot_handle != NULL);
3968         LASSERT(oth->ot_handle->h_transaction != NULL);
3969         LASSERT(pobj->oo_inode);
3970
3971         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3972         if (unlikely(pobj->oo_inode ==
3973                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3974                 ldp->edp_magic = 0;
3975         else
3976                 osd_get_ldiskfs_dirent_param(ldp, fid);
3977         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3978         child->d_fsdata = (void *)ldp;
3979         ll_vfs_dq_init(pobj->oo_inode);
3980         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3981         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3982                 struct ldiskfs_dir_entry_2      *de;
3983                 struct buffer_head              *bh;
3984                 int                              rc1;
3985
3986                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3987                                             NULL, hlock);
3988                 if (bh != NULL) {
3989                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3990                                                                bh);
3991                         if (rc1 == 0) {
3992                                 if (S_ISDIR(cinode->i_mode))
3993                                         de->file_type = LDISKFS_DIRENT_LUFID |
3994                                                         LDISKFS_FT_REG_FILE;
3995                                 else
3996                                         de->file_type = LDISKFS_DIRENT_LUFID |
3997                                                         LDISKFS_FT_DIR;
3998                                 ldiskfs_journal_dirty_metadata(oth->ot_handle,
3999                                                                bh);
4000                                 brelse(bh);
4001                         }
4002                 }
4003         }
4004
4005         RETURN(rc);
4006 }
4007
4008 /**
4009  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4010  * into the directory.Also sets flags into osd object to
4011  * indicate dot and dotdot are created. This is required for
4012  * interoperability mode (b11826)
4013  *
4014  * \param dir   directory for dot and dotdot fixup.
4015  * \param obj   child object for linking
4016  *
4017  * \retval   0, on success
4018  * \retval -ve, on error
4019  */
4020 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4021                               struct osd_object *dir,
4022                               struct inode *parent_dir, const char *name,
4023                               const struct lu_fid *dot_fid,
4024                               const struct lu_fid *dot_dot_fid,
4025                               struct thandle *th)
4026 {
4027         struct inode                *inode = dir->oo_inode;
4028         struct osd_thandle          *oth;
4029         int result = 0;
4030
4031         oth = container_of(th, struct osd_thandle, ot_super);
4032         LASSERT(oth->ot_handle->h_transaction != NULL);
4033         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4034
4035         if (strcmp(name, dot) == 0) {
4036                 if (dir->oo_compat_dot_created) {
4037                         result = -EEXIST;
4038                 } else {
4039                         LASSERT(inode == parent_dir);
4040                         dir->oo_compat_dot_created = 1;
4041                         result = 0;
4042                 }
4043         } else if (strcmp(name, dotdot) == 0) {
4044                 if (!dir->oo_compat_dot_created)
4045                         return -EINVAL;
4046                 /* in case of rename, dotdot is already created */
4047                 if (dir->oo_compat_dotdot_created) {
4048                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4049                                                 dot_dot_fid, NULL, th);
4050                 }
4051
4052                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4053                         struct lu_fid tfid = *dot_dot_fid;
4054
4055                         tfid.f_oid--;
4056                         result = osd_add_dot_dotdot_internal(info,
4057                                         dir->oo_inode, parent_dir, dot_fid,
4058                                         &tfid, oth);
4059                 } else {
4060                         result = osd_add_dot_dotdot_internal(info,
4061                                         dir->oo_inode, parent_dir, dot_fid,
4062                                         dot_dot_fid, oth);
4063                 }
4064
4065                 if (result == 0)
4066                         dir->oo_compat_dotdot_created = 1;
4067         }
4068
4069         return result;
4070 }
4071
4072
4073 /**
4074  * It will call the appropriate osd_add* function and return the
4075  * value, return by respective functions.
4076  */
4077 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4078                           struct inode *cinode, const char *name,
4079                           const struct lu_fid *fid, struct thandle *th)
4080 {
4081         struct osd_thread_info *info   = osd_oti_get(env);
4082         struct htree_lock      *hlock;
4083         int                     rc;
4084
4085         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4086
4087         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4088                                                    name[2] =='\0'))) {
4089                 if (hlock != NULL) {
4090                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4091                                            pobj->oo_inode, 0);
4092                 } else {
4093                         down_write(&pobj->oo_ext_idx_sem);
4094                 }
4095
4096                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4097                                         lu_object_fid(&pobj->oo_dt.do_lu),
4098                                         fid, th);
4099         } else {
4100                 if (hlock != NULL) {
4101                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4102                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4103                 } else {
4104                         down_write(&pobj->oo_ext_idx_sem);
4105                 }
4106
4107                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4108                         struct lu_fid *tfid = &info->oti_fid;
4109
4110                         *tfid = *fid;
4111                         tfid->f_ver = ~0;
4112                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4113                                               tfid, hlock, th);
4114                 } else {
4115                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4116                                               hlock, th);
4117                 }
4118         }
4119         if (hlock != NULL)
4120                 ldiskfs_htree_unlock(hlock);
4121         else
4122                 up_write(&pobj->oo_ext_idx_sem);
4123
4124         return rc;
4125 }
4126
4127 static void
4128 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4129                       struct osd_idmap_cache *oic)
4130 {
4131         struct osd_scrub    *scrub = &dev->od_scrub;
4132         struct lu_fid       *fid   = &oic->oic_fid;
4133         struct osd_inode_id *id    = &oti->oti_id;
4134         int                  once  = 0;
4135         int                  rc;
4136         ENTRY;
4137
4138         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4139                 RETURN_EXIT;
4140
4141 again:
4142         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4143         if (rc != 0 && rc != -ENOENT)
4144                 RETURN_EXIT;
4145
4146         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4147                 RETURN_EXIT;
4148
4149         if (thread_is_running(&scrub->os_thread)) {
4150                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4151                 /* There is race condition between osd_oi_lookup and OI scrub.
4152                  * The OI scrub finished just after osd_oi_lookup() failure.
4153                  * Under such case, it is unnecessary to trigger OI scrub again,
4154                  * but try to call osd_oi_lookup() again. */
4155                 if (unlikely(rc == -EAGAIN))
4156                         goto again;
4157
4158                 RETURN_EXIT;
4159         }
4160
4161         if (!dev->od_noscrub && ++once == 1) {
4162                 rc = osd_scrub_start(dev);
4163                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4164                               ", rc = %d [2]\n",
4165                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4166                               PFID(fid), rc);
4167                 if (rc == 0)
4168                         goto again;
4169         }
4170
4171         EXIT;
4172 }
4173
4174 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4175                                struct osd_device *dev,
4176                                struct osd_idmap_cache *oic,
4177                                struct lu_fid *fid, __u32 ino)
4178 {
4179         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4180         struct inode            *inode;
4181         int                      rc;
4182
4183         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4184         inode = osd_iget(oti, dev, &oic->oic_lid);
4185         if (IS_ERR(inode)) {
4186                 fid_zero(&oic->oic_fid);
4187                 return PTR_ERR(inode);
4188         }
4189
4190         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4191         iput(inode);
4192         if (rc != 0)
4193                 fid_zero(&oic->oic_fid);
4194         else
4195                 *fid = oic->oic_fid = lma->lma_self_fid;
4196         return rc;
4197 }
4198
4199 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4200                      struct osd_inode_id *id, const struct lu_fid *fid)
4201 {
4202         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4203                id->oii_ino, id->oii_gen, info);
4204         info->oti_cache.oic_lid = *id;
4205         info->oti_cache.oic_fid = *fid;
4206         info->oti_cache.oic_dev = osd;
4207
4208         return 0;
4209 }
4210
4211 /**
4212  * Get parent FID from the linkEA.
4213  *
4214  * For a directory which parent resides on remote MDT, to satisfy the
4215  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4216  * the other hand, to make the lookup(..) on the directory can return
4217  * the real parent FID, we append the real parent FID after its ".."
4218  * name entry in the /REMOTE_PARENT_DIR.
4219  *
4220  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4221  * backup. So after the restore, we cannot get the right parent FID from
4222  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4223  * we have stored the real parent FID in the directory object's linkEA,
4224  * we can parse the linkEA for the real parent FID.
4225  *
4226  * \param[in] env       pointer to the thread context
4227  * \param[in] obj       pointer to the object to be handled
4228  * \param[out]fid       pointer to the buffer to hold the parent FID
4229  *
4230  * \retval              0 for getting the real parent FID successfully
4231  * \retval              negative error number on failure
4232  */
4233 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4234                                     struct osd_object *obj,
4235                                     struct lu_fid *fid)
4236 {
4237         struct osd_thread_info  *oti    = osd_oti_get(env);
4238         struct lu_buf           *buf    = &oti->oti_big_buf;
4239         struct dentry           *dentry = &oti->oti_obj_dentry;
4240         struct inode            *inode  = obj->oo_inode;
4241         struct linkea_data       ldata  = { 0 };
4242         int                      rc;
4243         ENTRY;
4244
4245         fid_zero(fid);
4246         if (!S_ISDIR(inode->i_mode))
4247                 RETURN(-EIO);
4248
4249 again:
4250         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4251                              buf->lb_buf, buf->lb_len);
4252         if (rc == -ERANGE) {
4253                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4254                                      NULL, 0);
4255                 if (rc > 0) {
4256                         lu_buf_realloc(buf, rc);
4257                         if (buf->lb_buf == NULL)
4258                                 RETURN(-ENOMEM);
4259
4260                         goto again;
4261                 }
4262         }
4263
4264         if (unlikely(rc == 0))
4265                 RETURN(-ENODATA);
4266
4267         if (rc < 0)
4268                 RETURN(rc);
4269
4270         if (unlikely(buf->lb_buf == NULL)) {
4271                 lu_buf_realloc(buf, rc);
4272                 if (buf->lb_buf == NULL)
4273                         RETURN(-ENOMEM);
4274
4275                 goto again;
4276         }
4277
4278         ldata.ld_buf = buf;
4279         rc = linkea_init(&ldata);
4280         if (rc == 0) {
4281                 linkea_first_entry(&ldata);
4282                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4283         }
4284
4285         RETURN(rc);
4286 }
4287
4288 /**
4289  * Calls ->lookup() to find dentry. From dentry get inode and
4290  * read inode's ea to get fid. This is required for  interoperability
4291  * mode (b11826)
4292  *
4293  * \retval   0, on success
4294  * \retval -ve, on error
4295  */
4296 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4297                              struct dt_rec *rec, const struct dt_key *key)
4298 {
4299         struct inode                    *dir    = obj->oo_inode;
4300         struct dentry                   *dentry;
4301         struct ldiskfs_dir_entry_2      *de;
4302         struct buffer_head              *bh;
4303         struct lu_fid                   *fid = (struct lu_fid *) rec;
4304         struct htree_lock               *hlock = NULL;
4305         int                             ino;
4306         int                             rc;
4307         ENTRY;
4308
4309         LASSERT(dir->i_op != NULL);
4310         LASSERT(dir->i_op->lookup != NULL);
4311
4312         dentry = osd_child_dentry_get(env, obj,
4313                                       (char *)key, strlen((char *)key));
4314
4315         if (obj->oo_hl_head != NULL) {
4316                 hlock = osd_oti_get(env)->oti_hlock;
4317                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4318                                    dir, LDISKFS_HLOCK_LOOKUP);
4319         } else {
4320                 down_read(&obj->oo_ext_idx_sem);
4321         }
4322
4323         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4324         if (bh) {
4325                 struct osd_thread_info *oti = osd_oti_get(env);
4326                 struct osd_inode_id *id = &oti->oti_id;
4327                 struct osd_idmap_cache *oic = &oti->oti_cache;
4328                 struct osd_device *dev = osd_obj2dev(obj);
4329                 struct osd_scrub *scrub = &dev->od_scrub;
4330                 struct scrub_file *sf = &scrub->os_file;
4331
4332                 ino = le32_to_cpu(de->inode);
4333                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4334                         brelse(bh);
4335                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4336                         GOTO(out, rc);
4337                 }
4338
4339                 rc = osd_get_fid_from_dentry(de, rec);
4340
4341                 /* done with de, release bh */
4342                 brelse(bh);
4343                 if (rc != 0) {
4344                         if (unlikely(ino == osd_remote_parent_ino(dev)))
4345                                 /* If the parent is on remote MDT, and there
4346                                  * is no FID-in-dirent, then we have to get
4347                                  * the parent FID from the linkEA.  */
4348                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
4349                         else
4350                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4351                 } else {
4352                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4353                 }
4354
4355                 if (rc != 0) {
4356                         fid_zero(&oic->oic_fid);
4357                         GOTO(out, rc);
4358                 }
4359
4360                 if (osd_remote_fid(env, dev, fid))
4361                         GOTO(out, rc = 0);
4362
4363                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4364                                       fid);
4365                 if (rc != 0)
4366                         GOTO(out, rc);
4367                 if ((scrub->os_pos_current <= ino) &&
4368                     ((sf->sf_flags & SF_INCONSISTENT) ||
4369                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4370                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4371                                       sf->sf_oi_bitmap)))
4372                         osd_consistency_check(oti, dev, oic);
4373         } else {
4374                 rc = -ENOENT;
4375         }
4376
4377         GOTO(out, rc);
4378
4379 out:
4380         if (hlock != NULL)
4381                 ldiskfs_htree_unlock(hlock);
4382         else
4383                 up_read(&obj->oo_ext_idx_sem);
4384         return rc;
4385 }
4386
4387 /**
4388  * Find the osd object for given fid.
4389  *
4390  * \param fid need to find the osd object having this fid
4391  *
4392  * \retval osd_object on success
4393  * \retval        -ve on error
4394  */
4395 struct osd_object *osd_object_find(const struct lu_env *env,
4396                                    struct dt_object *dt,
4397                                    const struct lu_fid *fid)
4398 {
4399         struct lu_device  *ludev = dt->do_lu.lo_dev;
4400         struct osd_object *child = NULL;
4401         struct lu_object  *luch;
4402         struct lu_object  *lo;
4403
4404         /*
4405          * at this point topdev might not exist yet
4406          * (i.e. MGS is preparing profiles). so we can
4407          * not rely on topdev and instead lookup with
4408          * our device passed as topdev. this can't work
4409          * if the object isn't cached yet (as osd doesn't
4410          * allocate lu_header). IOW, the object must be
4411          * in the cache, otherwise lu_object_alloc() crashes
4412          * -bzzz
4413          */
4414         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4415                                  ludev : ludev->ld_site->ls_top_dev,
4416                                  fid, NULL);
4417         if (!IS_ERR(luch)) {
4418                 if (lu_object_exists(luch)) {
4419                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4420                         if (lo != NULL)
4421                                 child = osd_obj(lo);
4422                         else
4423                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4424                                                 "lu_object can't be located"
4425                                                 DFID"\n", PFID(fid));
4426
4427                         if (child == NULL) {
4428                                 lu_object_put(env, luch);
4429                                 CERROR("Unable to get osd_object\n");
4430                                 child = ERR_PTR(-ENOENT);
4431                         }
4432                 } else {
4433                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4434                                         "lu_object does not exists "DFID"\n",
4435                                         PFID(fid));
4436                         lu_object_put(env, luch);
4437                         child = ERR_PTR(-ENOENT);
4438                 }
4439         } else {
4440                 child = ERR_CAST(luch);
4441         }
4442
4443         return child;
4444 }
4445
4446 /**
4447  * Put the osd object once done with it.
4448  *
4449  * \param obj osd object that needs to be put
4450  */
4451 static inline void osd_object_put(const struct lu_env *env,
4452                                   struct osd_object *obj)
4453 {
4454         lu_object_put(env, &obj->oo_dt.do_lu);
4455 }
4456
4457 static int osd_index_declare_ea_insert(const struct lu_env *env,
4458                                        struct dt_object *dt,
4459                                        const struct dt_rec *rec,
4460                                        const struct dt_key *key,
4461                                        struct thandle *handle)
4462 {
4463         struct osd_thandle      *oh;
4464         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4465         struct lu_fid           *fid = (struct lu_fid *)rec;
4466         int                     rc;
4467         ENTRY;
4468
4469         LASSERT(!dt_object_remote(dt));
4470         LASSERT(handle != NULL);
4471
4472         oh = container_of0(handle, struct osd_thandle, ot_super);
4473         LASSERT(oh->ot_handle == NULL);
4474
4475         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4476                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4477
4478         if (osd_dt_obj(dt)->oo_inode != NULL) {
4479                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4480
4481                 /* We ignore block quota on meta pool (MDTs), so needn't
4482                  * calculate how many blocks will be consumed by this index
4483                  * insert */
4484                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4485                                            i_gid_read(inode), 0, oh,
4486                                            osd_dt_obj(dt), true, NULL, false);
4487         }
4488
4489         if (fid == NULL)
4490                 RETURN(0);
4491
4492         rc = osd_remote_fid(env, osd, fid);
4493         if (rc <= 0)
4494                 RETURN(rc);
4495
4496         rc = 0;
4497
4498         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4499                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4500         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4501                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4502         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4503                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4504
4505         RETURN(rc);
4506 }
4507
4508 /**
4509  * Index add function for interoperability mode (b11826).
4510  * It will add the directory entry.This entry is needed to
4511  * maintain name->fid mapping.
4512  *
4513  * \param key it is key i.e. file entry to be inserted
4514  * \param rec it is value of given key i.e. fid
4515  *
4516  * \retval   0, on success
4517  * \retval -ve, on error
4518  */
4519 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4520                                const struct dt_rec *rec,
4521                                const struct dt_key *key, struct thandle *th,
4522                                struct lustre_capa *capa, int ignore_quota)
4523 {
4524         struct osd_object       *obj = osd_dt_obj(dt);
4525         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4526         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4527         const struct lu_fid     *fid    = rec1->rec_fid;
4528         const char              *name = (const char *)key;
4529         struct osd_thread_info  *oti   = osd_oti_get(env);
4530         struct osd_inode_id     *id    = &oti->oti_id;
4531         struct inode            *child_inode = NULL;
4532         struct osd_object       *child = NULL;
4533         int                     rc;
4534         ENTRY;
4535
4536         LASSERT(osd_invariant(obj));
4537         LASSERT(dt_object_exists(dt));
4538         LASSERT(!dt_object_remote(dt));
4539         LASSERT(th != NULL);
4540
4541         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4542
4543         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4544                 RETURN(-EACCES);
4545
4546         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4547
4548         rc = osd_remote_fid(env, osd, fid);
4549         if (rc < 0) {
4550                 CERROR("%s: Can not find object "DFID" rc %d\n",
4551                        osd_name(osd), PFID(fid), rc);
4552                 RETURN(rc);
4553         }
4554
4555         if (rc == 1) {
4556                 /* Insert remote entry */
4557                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4558                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4559                         struct osd_thandle      *oh;
4560
4561                         /* If parent on remote MDT, we need put this object
4562                          * under AGENT */
4563                         oh = container_of(th, typeof(*oh), ot_super);
4564                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4565                         if (rc != 0) {
4566                                 CERROR("%s: add "DFID" error: rc = %d\n",
4567                                        osd_name(osd),
4568                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4569                                 RETURN(rc);
4570                         }
4571
4572                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4573                 } else {
4574                         child_inode = osd_create_local_agent_inode(env, osd,
4575                                         obj, fid, rec1->rec_type & S_IFMT, th);
4576                         if (IS_ERR(child_inode))
4577                                 RETURN(PTR_ERR(child_inode));
4578                 }
4579         } else {
4580                 /* Insert local entry */
4581                 child = osd_object_find(env, dt, fid);
4582                 if (IS_ERR(child)) {
4583                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4584                                osd_name(osd), PFID(fid),
4585                                id->oii_ino, id->oii_gen,
4586                                (int)PTR_ERR(child));
4587                         RETURN(PTR_ERR(child));
4588                 }
4589                 child_inode = igrab(child->oo_inode);
4590         }
4591
4592         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4593
4594         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4595                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4596
4597         iput(child_inode);
4598         if (child != NULL)
4599                 osd_object_put(env, child);
4600         LASSERT(osd_invariant(obj));
4601         RETURN(rc);
4602 }
4603
4604 /**
4605  *  Initialize osd Iterator for given osd index object.
4606  *
4607  *  \param  dt      osd index object
4608  */
4609
4610 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4611                                      struct dt_object *dt,
4612                                      __u32 unused,
4613                                      struct lustre_capa *capa)
4614 {
4615         struct osd_it_iam      *it;
4616         struct osd_thread_info *oti = osd_oti_get(env);
4617         struct osd_object      *obj = osd_dt_obj(dt);
4618         struct lu_object       *lo  = &dt->do_lu;
4619         struct iam_path_descr  *ipd;
4620         struct iam_container   *bag = &obj->oo_dir->od_container;
4621
4622         LASSERT(lu_object_exists(lo));
4623
4624         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4625                 return ERR_PTR(-EACCES);
4626
4627         if (oti->oti_it_inline) {
4628                 OBD_ALLOC_PTR(it);
4629                 if (it == NULL)
4630                         return ERR_PTR(-ENOMEM);
4631         } else {
4632                 it = &oti->oti_it;
4633                 oti->oti_it_inline = 1;
4634         }
4635
4636         ipd = osd_it_ipd_get(env, bag);
4637         if (likely(ipd != NULL)) {
4638                 it->oi_obj = obj;
4639                 it->oi_ipd = ipd;
4640                 lu_object_get(lo);
4641                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4642                 return (struct dt_it *)it;
4643         } else {
4644                 if (it != &oti->oti_it)
4645                         OBD_FREE_PTR(it);
4646                 else
4647                         oti->oti_it_inline = 0;
4648
4649                 return ERR_PTR(-ENOMEM);
4650         }
4651 }
4652
4653 /**
4654  * free given Iterator.
4655  */
4656
4657 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4658 {
4659         struct osd_thread_info  *oti = osd_oti_get(env);
4660         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4661         struct osd_object       *obj = it->oi_obj;
4662
4663         iam_it_fini(&it->oi_it);
4664         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4665         lu_object_put(env, &obj->oo_dt.do_lu);
4666         if (it != &oti->oti_it)
4667                 OBD_FREE_PTR(it);
4668         else
4669                 oti->oti_it_inline = 0;
4670 }
4671
4672 /**
4673  *  Move Iterator to record specified by \a key
4674  *
4675  *  \param  di      osd iterator
4676  *  \param  key     key for index
4677  *
4678  *  \retval +ve  di points to record with least key not larger than key
4679  *  \retval  0   di points to exact matched key
4680  *  \retval -ve  failure
4681  */
4682
4683 static int osd_it_iam_get(const struct lu_env *env,
4684                           struct dt_it *di, const struct dt_key *key)
4685 {
4686         struct osd_thread_info  *oti = osd_oti_get(env);
4687         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4688
4689         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4690                 /* swab quota uid/gid */
4691                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4692                 key = (struct dt_key *)&oti->oti_quota_id;
4693         }
4694
4695         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4696 }
4697
4698 /**
4699  *  Release Iterator
4700  *
4701  *  \param  di      osd iterator
4702  */
4703 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4704 {
4705         struct osd_it_iam *it = (struct osd_it_iam *)di;
4706
4707         iam_it_put(&it->oi_it);
4708 }
4709
4710 /**
4711  *  Move iterator by one record
4712  *
4713  *  \param  di      osd iterator
4714  *
4715  *  \retval +1   end of container reached
4716  *  \retval  0   success
4717  *  \retval -ve  failure
4718  */
4719
4720 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4721 {
4722         struct osd_it_iam *it = (struct osd_it_iam *)di;
4723
4724         return iam_it_next(&it->oi_it);
4725 }
4726
4727 /**
4728  * Return pointer to the key under iterator.
4729  */
4730
4731 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4732                                  const struct dt_it *di)
4733 {
4734         struct osd_thread_info *oti = osd_oti_get(env);
4735         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4736         struct osd_object      *obj = it->oi_obj;
4737         struct dt_key          *key;
4738
4739         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4740
4741         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4742                 /* swab quota uid/gid */
4743                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4744                 key = (struct dt_key *)&oti->oti_quota_id;
4745         }
4746
4747         return key;
4748 }
4749
4750 /**
4751  * Return size of key under iterator (in bytes)
4752  */
4753
4754 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4755 {
4756         struct osd_it_iam *it = (struct osd_it_iam *)di;
4757
4758         return iam_it_key_size(&it->oi_it);
4759 }
4760
4761 static inline void
4762 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4763 {
4764         /* check if file type is required */
4765         if (ent->lde_attrs & LUDA_TYPE) {
4766                 struct luda_type *lt;
4767                 int align = sizeof(*lt) - 1;
4768
4769                 len = (len + align) & ~align;
4770                 lt = (struct luda_type *)(ent->lde_name + len);
4771                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4772         }
4773
4774         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4775 }
4776
4777 /**
4778  * build lu direct from backend fs dirent.
4779  */
4780
4781 static inline void
4782 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4783                    char *name, __u16 namelen, __u16 type, __u32 attr)
4784 {
4785         ent->lde_attrs = attr | LUDA_FID;
4786         fid_cpu_to_le(&ent->lde_fid, fid);
4787
4788         ent->lde_hash = cpu_to_le64(offset);
4789         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4790
4791         strncpy(ent->lde_name, name, namelen);
4792         ent->lde_name[namelen] = '\0';
4793         ent->lde_namelen = cpu_to_le16(namelen);
4794
4795         /* append lustre attributes */
4796         osd_it_append_attrs(ent, namelen, type);
4797 }
4798
4799 /**
4800  * Return pointer to the record under iterator.
4801  */
4802 static int osd_it_iam_rec(const struct lu_env *env,
4803                           const struct dt_it *di,
4804                           struct dt_rec *dtrec, __u32 attr)
4805 {
4806         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4807         struct osd_thread_info *info = osd_oti_get(env);
4808         ENTRY;
4809
4810         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4811                 const struct osd_fid_pack *rec;
4812                 struct lu_fid             *fid = &info->oti_fid;
4813                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4814                 char                      *name;
4815                 int                        namelen;
4816                 __u64                      hash;
4817                 int                        rc;
4818
4819                 name = (char *)iam_it_key_get(&it->oi_it);
4820                 if (IS_ERR(name))
4821                         RETURN(PTR_ERR(name));
4822
4823                 namelen = iam_it_key_size(&it->oi_it);
4824
4825                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4826                 if (IS_ERR(rec))
4827                         RETURN(PTR_ERR(rec));
4828
4829                 rc = osd_fid_unpack(fid, rec);
4830                 if (rc)
4831                         RETURN(rc);
4832
4833                 hash = iam_it_store(&it->oi_it);
4834
4835                 /* IAM does not store object type in IAM index (dir) */
4836                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4837                                    0, LUDA_FID);
4838         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4839                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4840                            (struct iam_rec *)dtrec);
4841                 osd_quota_unpack(it->oi_obj, dtrec);
4842         } else {
4843                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4844                            (struct iam_rec *)dtrec);
4845         }
4846
4847         RETURN(0);
4848 }
4849
4850 /**
4851  * Returns cookie for current Iterator position.
4852  */
4853 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4854 {
4855         struct osd_it_iam *it = (struct osd_it_iam *)di;
4856
4857         return iam_it_store(&it->oi_it);
4858 }
4859
4860 /**
4861  * Restore iterator from cookie.
4862  *
4863  * \param  di      osd iterator
4864  * \param  hash    Iterator location cookie
4865  *
4866  * \retval +ve  di points to record with least key not larger than key.
4867  * \retval  0   di points to exact matched key
4868  * \retval -ve  failure
4869  */
4870
4871 static int osd_it_iam_load(const struct lu_env *env,
4872                            const struct dt_it *di, __u64 hash)
4873 {
4874         struct osd_it_iam *it = (struct osd_it_iam *)di;
4875
4876         return iam_it_load(&it->oi_it, hash);
4877 }
4878
4879 static const struct dt_index_operations osd_index_iam_ops = {
4880         .dio_lookup         = osd_index_iam_lookup,
4881         .dio_declare_insert = osd_index_declare_iam_insert,
4882         .dio_insert         = osd_index_iam_insert,
4883         .dio_declare_delete = osd_index_declare_iam_delete,
4884         .dio_delete         = osd_index_iam_delete,
4885         .dio_it     = {
4886                 .init     = osd_it_iam_init,
4887                 .fini     = osd_it_iam_fini,
4888                 .get      = osd_it_iam_get,
4889                 .put      = osd_it_iam_put,
4890                 .next     = osd_it_iam_next,
4891                 .key      = osd_it_iam_key,
4892                 .key_size = osd_it_iam_key_size,
4893                 .rec      = osd_it_iam_rec,
4894                 .store    = osd_it_iam_store,
4895                 .load     = osd_it_iam_load
4896         }
4897 };
4898
4899
4900 /**
4901  * Creates or initializes iterator context.
4902  *
4903  * \retval struct osd_it_ea, iterator structure on success
4904  *
4905  */
4906 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4907                                     struct dt_object *dt,
4908                                     __u32 attr,
4909                                     struct lustre_capa *capa)
4910 {
4911         struct osd_object       *obj  = osd_dt_obj(dt);
4912         struct osd_thread_info  *info = osd_oti_get(env);
4913         struct osd_it_ea        *it;
4914         struct file             *file;
4915         struct lu_object        *lo   = &dt->do_lu;
4916         struct dentry           *obj_dentry = &info->oti_it_dentry;
4917         ENTRY;
4918
4919         LASSERT(lu_object_exists(lo));
4920
4921         if (info->oti_it_inline) {
4922                 OBD_ALLOC_PTR(it);
4923                 if (it == NULL)
4924                         RETURN(ERR_PTR(-ENOMEM));
4925         } else {
4926                 it = &info->oti_it_ea;
4927                 info->oti_it_inline = 1;
4928         }
4929
4930         obj_dentry->d_inode = obj->oo_inode;
4931         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4932         obj_dentry->d_name.hash = 0;
4933
4934         it->oie_rd_dirent       = 0;
4935         it->oie_it_dirent       = 0;
4936         it->oie_dirent          = NULL;
4937         it->oie_buf             = info->oti_it_ea_buf;
4938         it->oie_obj             = obj;
4939
4940         file = &it->oie_file;
4941         /* Reset the "file" totally to avoid to reuse any old value from
4942          * former readdir handling, the "file->f_pos" should be zero. */
4943         memset(file, 0, sizeof(*file));
4944         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4945         if (attr & LUDA_64BITHASH)
4946                 file->f_mode    = FMODE_64BITHASH;
4947         else
4948                 file->f_mode    = FMODE_32BITHASH;
4949         file->f_dentry          = obj_dentry;
4950         file->f_mapping         = obj->oo_inode->i_mapping;
4951         file->f_op              = obj->oo_inode->i_fop;
4952         set_file_inode(file, obj->oo_inode);
4953
4954         lu_object_get(lo);
4955         RETURN((struct dt_it *) it);
4956 }
4957
4958 /**
4959  * Destroy or finishes iterator context.
4960  *
4961  * \param di iterator structure to be destroyed
4962  */
4963 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4964 {
4965         struct osd_thread_info  *info   = osd_oti_get(env);
4966         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4967         struct osd_object       *obj    = it->oie_obj;
4968         struct inode            *inode  = obj->oo_inode;
4969
4970         ENTRY;
4971         it->oie_file.f_op->release(inode, &it->oie_file);
4972         lu_object_put(env, &obj->oo_dt.do_lu);
4973         if (it != &info->oti_it_ea)
4974                 OBD_FREE_PTR(it);
4975         else
4976                 info->oti_it_inline = 0;
4977         EXIT;
4978 }
4979
4980 /**
4981  * It position the iterator at given key, so that next lookup continues from
4982  * that key Or it is similar to dio_it->load() but based on a key,
4983  * rather than file position.
4984  *
4985  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4986  * to the beginning.
4987  *
4988  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4989  */
4990 static int osd_it_ea_get(const struct lu_env *env,
4991                          struct dt_it *di, const struct dt_key *key)
4992 {
4993         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4994
4995         ENTRY;
4996         LASSERT(((const char *)key)[0] == '\0');
4997         it->oie_file.f_pos      = 0;
4998         it->oie_rd_dirent       = 0;
4999         it->oie_it_dirent       = 0;
5000         it->oie_dirent          = NULL;
5001
5002         RETURN(+1);
5003 }
5004
5005 /**
5006  * Does nothing
5007  */
5008 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5009 {
5010 }
5011
5012 struct osd_filldir_cbs {
5013 #ifdef HAVE_DIR_CONTEXT
5014         struct dir_context ctx;
5015 #endif
5016         struct osd_it_ea  *it;
5017 };
5018 /**
5019  * It is called internally by ->readdir(). It fills the
5020  * iterator's in-memory data structure with required
5021  * information i.e. name, namelen, rec_size etc.
5022  *
5023  * \param buf in which information to be filled in.
5024  * \param name name of the file in given dir
5025  *
5026  * \retval 0 on success
5027  * \retval 1 on buffer full
5028  */
5029 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5030                                loff_t offset, __u64 ino,
5031                                unsigned d_type)
5032 {
5033         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5034         struct osd_object       *obj  = it->oie_obj;
5035         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5036         struct lu_fid           *fid  = &ent->oied_fid;
5037         struct osd_fid_pack     *rec;
5038         ENTRY;
5039
5040         /* this should never happen */
5041         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5042                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5043                 RETURN(-EIO);
5044         }
5045
5046         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5047             OSD_IT_EA_BUFSIZE)
5048                 RETURN(1);
5049
5050         /* "." is just the object itself. */
5051         if (namelen == 1 && name[0] == '.') {
5052                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5053         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5054                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5055                 if (osd_fid_unpack(fid, rec) != 0)
5056                         fid_zero(fid);
5057         } else {
5058                 fid_zero(fid);
5059         }
5060         d_type &= ~LDISKFS_DIRENT_LUFID;
5061
5062         /* NOT export local root. */
5063         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5064                 ino = obj->oo_inode->i_ino;
5065                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5066         }
5067
5068         ent->oied_ino     = ino;
5069         ent->oied_off     = offset;
5070         ent->oied_namelen = namelen;
5071         ent->oied_type    = d_type;
5072
5073         memcpy(ent->oied_name, name, namelen);
5074
5075         it->oie_rd_dirent++;
5076         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5077         RETURN(0);
5078 }
5079
5080 /**
5081  * Calls ->readdir() to load a directory entry at a time
5082  * and stored it in iterator's in-memory data structure.
5083  *
5084  * \param di iterator's in memory structure
5085  *
5086  * \retval   0 on success
5087  * \retval -ve on error
5088  * \retval +1 reach the end of entry
5089  */
5090 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5091                                const struct dt_it *di)
5092 {
5093         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5094         struct osd_object  *obj   = it->oie_obj;
5095         struct inode       *inode = obj->oo_inode;
5096         struct htree_lock  *hlock = NULL;
5097         struct file        *filp  = &it->oie_file;
5098         int                 rc = 0;
5099         struct osd_filldir_cbs buf = {
5100 #ifdef HAVE_DIR_CONTEXT
5101                 .ctx.actor = osd_ldiskfs_filldir,
5102 #endif
5103                 .it = it
5104         };
5105
5106         ENTRY;
5107         it->oie_dirent = it->oie_buf;
5108         it->oie_rd_dirent = 0;
5109
5110         if (obj->oo_hl_head != NULL) {
5111                 hlock = osd_oti_get(env)->oti_hlock;
5112                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5113                                    inode, LDISKFS_HLOCK_READDIR);
5114         } else {
5115                 down_read(&obj->oo_ext_idx_sem);
5116         }
5117
5118 #ifdef HAVE_DIR_CONTEXT
5119         buf.ctx.pos = filp->f_pos;
5120         rc = inode->i_fop->iterate(filp, &buf.ctx);
5121         filp->f_pos = buf.ctx.pos;
5122 #else
5123         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5124 #endif
5125
5126         if (hlock != NULL)
5127                 ldiskfs_htree_unlock(hlock);
5128         else
5129                 up_read(&obj->oo_ext_idx_sem);
5130
5131         if (it->oie_rd_dirent == 0) {
5132                 /*If it does not get any dirent, it means it has been reached
5133                  *to the end of the dir */
5134                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5135                 if (rc == 0)
5136                         rc = 1;
5137         } else {
5138                 it->oie_dirent = it->oie_buf;
5139                 it->oie_it_dirent = 1;
5140         }
5141
5142         RETURN(rc);
5143 }
5144
5145 /**
5146  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5147  * to load a directory entry at a time and stored it in
5148  * iterator's in-memory data structure.
5149  *
5150  * \param di iterator's in memory structure
5151  *
5152  * \retval +ve iterator reached to end
5153  * \retval   0 iterator not reached to end
5154  * \retval -ve on error
5155  */
5156 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5157 {
5158         struct osd_it_ea *it = (struct osd_it_ea *)di;
5159         int rc;
5160
5161         ENTRY;
5162
5163         if (it->oie_it_dirent < it->oie_rd_dirent) {
5164                 it->oie_dirent =
5165                         (void *) it->oie_dirent +
5166                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5167                                        it->oie_dirent->oied_namelen);
5168                 it->oie_it_dirent++;
5169                 RETURN(0);
5170         } else {
5171                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5172                         rc = +1;
5173                 else
5174                         rc = osd_ldiskfs_it_fill(env, di);
5175         }
5176
5177         RETURN(rc);
5178 }
5179
5180 /**
5181  * Returns the key at current position from iterator's in memory structure.
5182  *
5183  * \param di iterator's in memory structure
5184  *
5185  * \retval key i.e. struct dt_key on success
5186  */
5187 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5188                                     const struct dt_it *di)
5189 {
5190         struct osd_it_ea *it = (struct osd_it_ea *)di;
5191
5192         return (struct dt_key *)it->oie_dirent->oied_name;
5193 }
5194
5195 /**
5196  * Returns the key's size at current position from iterator's in memory structure.
5197  *
5198  * \param di iterator's in memory structure
5199  *
5200  * \retval key_size i.e. struct dt_key on success
5201  */
5202 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5203 {
5204         struct osd_it_ea *it = (struct osd_it_ea *)di;
5205
5206         return it->oie_dirent->oied_namelen;
5207 }
5208
5209 static int
5210 osd_dirent_update(handle_t *jh, struct super_block *sb,
5211                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5212                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5213 {
5214         struct osd_fid_pack *rec;
5215         int                  rc;
5216         ENTRY;
5217
5218         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5219         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5220
5221         rc = ldiskfs_journal_get_write_access(jh, bh);
5222         if (rc != 0)
5223                 RETURN(rc);
5224
5225         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5226         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5227         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5228
5229         RETURN(rc);
5230 }
5231
5232 static inline int
5233 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5234 {
5235         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5236             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5237                 return 1;
5238         else
5239                 return 0;
5240 }
5241
5242 static inline int
5243 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5244 {
5245         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5246                  "dot_dotdot = %d\n", dot_dotdot);
5247
5248         if (LDISKFS_DIR_REC_LEN(de) >=
5249             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5250                 return 1;
5251         else
5252                 return 0;
5253 }
5254
5255 static int
5256 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5257                     struct inode *dir, struct inode *inode,
5258                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5259                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5260                     struct htree_lock *hlock)
5261 {
5262         struct dentry               *dentry;
5263         struct osd_fid_pack         *rec;
5264         struct ldiskfs_dentry_param *ldp;
5265         int                          rc;
5266         ENTRY;
5267
5268         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5269                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5270                 RETURN(0);
5271
5272         /* There is enough space to hold the FID-in-dirent. */
5273         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5274                                  dir->i_sb->s_blocksize)) {
5275                 rc = ldiskfs_journal_get_write_access(jh, bh);
5276                 if (rc != 0)
5277                         RETURN(rc);
5278
5279                 de->name[de->name_len] = 0;
5280                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5281                 rec->fp_len = sizeof(struct lu_fid) + 1;
5282                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5283                 de->file_type |= LDISKFS_DIRENT_LUFID;
5284
5285                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5286
5287                 RETURN(rc);
5288         }
5289
5290         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5291         if (rc != 0)
5292                 RETURN(rc);
5293
5294         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5295                                            ent->oied_namelen);
5296         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5297         osd_get_ldiskfs_dirent_param(ldp, fid);
5298         dentry->d_fsdata = (void *)ldp;
5299         ll_vfs_dq_init(dir);
5300         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5301         /* It is too bad, we cannot reinsert the name entry back.
5302          * That means we lose it! */
5303         if (rc != 0)
5304                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5305                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5306                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5307                        dir->i_ino, dir->i_generation,
5308                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5309
5310         RETURN(rc);
5311 }
5312
5313 static int
5314 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5315                         struct osd_it_ea *it, struct lu_fid *fid,
5316                         struct osd_inode_id *id, __u32 *attr)
5317 {
5318         struct osd_thread_info     *info        = osd_oti_get(env);
5319         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5320         struct osd_device          *dev         = osd_obj2dev(obj);
5321         struct super_block         *sb          = osd_sb(dev);
5322         const char                 *devname     =
5323                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5324         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5325         struct inode               *dir         = obj->oo_inode;
5326         struct htree_lock          *hlock       = NULL;
5327         struct buffer_head         *bh          = NULL;
5328         handle_t                   *jh          = NULL;
5329         struct ldiskfs_dir_entry_2 *de;
5330         struct dentry              *dentry;
5331         struct inode               *inode;
5332         int                         credits;
5333         int                         rc;
5334         int                         dot_dotdot  = 0;
5335         bool                        dirty       = false;
5336         ENTRY;
5337
5338         if (ent->oied_name[0] == '.') {
5339                 if (ent->oied_namelen == 1)
5340                         dot_dotdot = 1;
5341                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5342                         dot_dotdot = 2;
5343         }
5344
5345         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5346                                       ent->oied_namelen);
5347
5348         /* We need to ensure that the name entry is still valid.
5349          * Because it may be removed or renamed by other already.
5350          *
5351          * The unlink or rename operation will start journal before PDO lock,
5352          * so to avoid deadlock, here we need to start journal handle before
5353          * related PDO lock also. But because we do not know whether there
5354          * will be something to be repaired before PDO lock, we just start
5355          * journal without conditions.
5356          *
5357          * We may need to remove the name entry firstly, then insert back.
5358          * One credit is for user quota file update.
5359          * One credit is for group quota file update.
5360          * Two credits are for dirty inode. */
5361         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5362                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5363
5364 again:
5365         if (dev->od_dirent_journal) {
5366                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5367                 if (IS_ERR(jh)) {
5368                         rc = PTR_ERR(jh);
5369                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5370                                "check_repair, dir = %lu/%u, credits = %d, "
5371                                "name = %.*s: rc = %d\n",
5372                                devname, dir->i_ino, dir->i_generation, credits,
5373                                ent->oied_namelen, ent->oied_name, rc);
5374                         RETURN(rc);
5375                 }
5376
5377                 if (obj->oo_hl_head != NULL) {
5378                         hlock = osd_oti_get(env)->oti_hlock;
5379                         /* "0" means exclusive lock for the whole directory.
5380                          * We need to prevent others access such name entry
5381                          * during the delete + insert. Neither HLOCK_ADD nor
5382                          * HLOCK_DEL cannot guarantee the atomicity. */
5383                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5384                 } else {
5385                         down_write(&obj->oo_ext_idx_sem);
5386                 }
5387         } else {
5388                 if (obj->oo_hl_head != NULL) {
5389                         hlock = osd_oti_get(env)->oti_hlock;
5390                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5391                                            LDISKFS_HLOCK_LOOKUP);
5392                 } else {
5393                         down_read(&obj->oo_ext_idx_sem);
5394                 }
5395         }
5396
5397         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5398         /* For dot/dotdot entry, if there is not enough space to hold the
5399          * FID-in-dirent, just keep them there. It only happens when the
5400          * device upgraded from 1.8 or restored from MDT file-level backup.
5401          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5402          * and needs to get FID from LMA when readdir, it will not affect the
5403          * performance much. */
5404         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5405             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5406                 *attr |= LUDA_IGNORE;
5407                 GOTO(out_journal, rc = 0);
5408         }
5409
5410         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5411         inode = osd_iget(info, dev, id);
5412         if (IS_ERR(inode)) {
5413                 rc = PTR_ERR(inode);
5414                 if (rc == -ENOENT || rc == -ESTALE)
5415                         rc = 1;
5416                 else
5417                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5418                                "check_repair, dir = %lu/%u, name = %.*s: "
5419                                "rc = %d\n",
5420                                devname, dir->i_ino, dir->i_generation,
5421                                ent->oied_namelen, ent->oied_name, rc);
5422
5423                 GOTO(out_journal, rc);
5424         }
5425
5426         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5427         if (rc == 0) {
5428                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5429
5430                 if (fid_is_sane(fid)) {
5431                         /* FID-in-dirent is valid. */
5432                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5433                                 GOTO(out_inode, rc = 0);
5434
5435                         /* Do not repair under dryrun mode. */
5436                         if (*attr & LUDA_VERIFY_DRYRUN) {
5437                                 *attr |= LUDA_REPAIR;
5438                                 GOTO(out_inode, rc = 0);
5439                         }
5440
5441                         if (!dev->od_dirent_journal) {
5442                                 iput(inode);
5443                                 brelse(bh);
5444                                 if (hlock != NULL)
5445                                         ldiskfs_htree_unlock(hlock);
5446                                 else
5447                                         up_read(&obj->oo_ext_idx_sem);
5448                                 dev->od_dirent_journal = 1;
5449                                 goto again;
5450                         }
5451
5452                         *fid = lma->lma_self_fid;
5453                         dirty = true;
5454                         /* Update the FID-in-dirent. */
5455                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5456                         if (rc == 0)
5457                                 *attr |= LUDA_REPAIR;
5458                         else
5459                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5460                                        "in the dirent, dir = %lu/%u, "
5461                                        "name = %.*s, "DFID": rc = %d\n",
5462                                        devname, dir->i_ino, dir->i_generation,
5463                                        ent->oied_namelen, ent->oied_name,
5464                                        PFID(fid), rc);
5465                 } else {
5466                         /* Do not repair under dryrun mode. */
5467                         if (*attr & LUDA_VERIFY_DRYRUN) {
5468                                 *fid = lma->lma_self_fid;
5469                                 *attr |= LUDA_REPAIR;
5470                                 GOTO(out_inode, rc = 0);
5471                         }
5472
5473                         if (!dev->od_dirent_journal) {
5474                                 iput(inode);
5475                                 brelse(bh);
5476                                 if (hlock != NULL)
5477                                         ldiskfs_htree_unlock(hlock);
5478                                 else
5479                                         up_read(&obj->oo_ext_idx_sem);
5480                                 dev->od_dirent_journal = 1;
5481                                 goto again;
5482                         }
5483
5484                         *fid = lma->lma_self_fid;
5485                         dirty = true;
5486                         /* Append the FID-in-dirent. */
5487                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5488                                                  fid, bh, de, hlock);
5489                         if (rc == 0)
5490                                 *attr |= LUDA_REPAIR;
5491                         else
5492                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5493                                        "after the dirent, dir = %lu/%u, "
5494                                        "name = %.*s, "DFID": rc = %d\n",
5495                                        devname, dir->i_ino, dir->i_generation,
5496                                        ent->oied_namelen, ent->oied_name,
5497                                        PFID(fid), rc);
5498                 }
5499         } else if (rc == -ENODATA) {
5500                 /* Do not repair under dryrun mode. */
5501                 if (*attr & LUDA_VERIFY_DRYRUN) {
5502                         if (fid_is_sane(fid)) {
5503                                 *attr |= LUDA_REPAIR;
5504                         } else {
5505                                 lu_igif_build(fid, inode->i_ino,
5506                                               inode->i_generation);
5507                                 *attr |= LUDA_UPGRADE;
5508                         }
5509                         GOTO(out_inode, rc = 0);
5510                 }
5511
5512                 if (!dev->od_dirent_journal) {
5513                         iput(inode);
5514                         brelse(bh);
5515                         if (hlock != NULL)
5516                                 ldiskfs_htree_unlock(hlock);
5517                         else
5518                                 up_read(&obj->oo_ext_idx_sem);
5519                         dev->od_dirent_journal = 1;
5520                         goto again;
5521                 }
5522
5523                 dirty = true;
5524                 if (unlikely(fid_is_sane(fid))) {
5525                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5526                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5527                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5528                         if (rc == 0)
5529                                 *attr |= LUDA_REPAIR;
5530                         else
5531                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5532                                        "update dirent, dir = %lu/%u, "
5533                                        "name = %.*s, "DFID": rc = %d\n",
5534                                        devname, dir->i_ino, dir->i_generation,
5535                                        ent->oied_namelen, ent->oied_name,
5536                                        PFID(fid), rc);
5537                 } else {
5538                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5539                         /* It is probably IGIF object. Only aappend the
5540                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5541                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5542                                                  fid, bh, de, hlock);
5543                         if (rc == 0)
5544                                 *attr |= LUDA_UPGRADE;
5545                         else
5546                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5547                                        "after the dirent, dir = %lu/%u, "
5548                                        "name = %.*s, "DFID": rc = %d\n",
5549                                        devname, dir->i_ino, dir->i_generation,
5550                                        ent->oied_namelen, ent->oied_name,
5551                                        PFID(fid), rc);
5552                 }
5553         }
5554
5555         GOTO(out_inode, rc);
5556
5557 out_inode:
5558         iput(inode);
5559
5560 out_journal:
5561         brelse(bh);
5562         if (hlock != NULL) {
5563                 ldiskfs_htree_unlock(hlock);
5564         } else {
5565                 if (dev->od_dirent_journal)
5566                         up_write(&obj->oo_ext_idx_sem);
5567                 else
5568                         up_read(&obj->oo_ext_idx_sem);
5569         }
5570         if (jh != NULL)
5571                 ldiskfs_journal_stop(jh);
5572         if (rc >= 0 && !dirty)
5573                 dev->od_dirent_journal = 0;
5574         return rc;
5575 }
5576
5577 /**
5578  * Returns the value at current position from iterator's in memory structure.
5579  *
5580  * \param di struct osd_it_ea, iterator's in memory structure
5581  * \param attr attr requested for dirent.
5582  * \param lde lustre dirent
5583  *
5584  * \retval   0 no error and \param lde has correct lustre dirent.
5585  * \retval -ve on error
5586  */
5587 static inline int osd_it_ea_rec(const struct lu_env *env,
5588                                 const struct dt_it *di,
5589                                 struct dt_rec *dtrec, __u32 attr)
5590 {
5591         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5592         struct osd_object      *obj   = it->oie_obj;
5593         struct osd_device      *dev   = osd_obj2dev(obj);
5594         struct osd_scrub       *scrub = &dev->od_scrub;
5595         struct scrub_file      *sf    = &scrub->os_file;
5596         struct osd_thread_info *oti   = osd_oti_get(env);
5597         struct osd_inode_id    *id    = &oti->oti_id;
5598         struct osd_idmap_cache *oic   = &oti->oti_cache;
5599         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5600         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5601         __u32                   ino   = it->oie_dirent->oied_ino;
5602         int                     rc    = 0;
5603         ENTRY;
5604
5605         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5606
5607         if (attr & LUDA_VERIFY) {
5608                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5609                         attr |= LUDA_IGNORE;
5610                         /* If the parent is on remote MDT, and there
5611                          * is no FID-in-dirent, then we have to get
5612                          * the parent FID from the linkEA.  */
5613                         osd_get_pfid_from_linkea(env, obj, fid);
5614                 } else {
5615                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5616                                                      &attr);
5617                 }
5618         } else {
5619                 attr &= ~LU_DIRENT_ATTRS_MASK;
5620                 if (!fid_is_sane(fid)) {
5621                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5622                             likely(it->oie_dirent->oied_namelen != 2 ||
5623                                    it->oie_dirent->oied_name[0] != '.' ||
5624                                    it->oie_dirent->oied_name[1] != '.'))
5625                                 RETURN(-ENOENT);
5626
5627                         if (unlikely(ino == osd_remote_parent_ino(dev)))
5628                                 /* If the parent is on remote MDT, and there
5629                                  * is no FID-in-dirent, then we have to get
5630                                  * the parent FID from the linkEA.  */
5631                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5632                         else
5633                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5634                 } else {
5635                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5636                 }
5637         }
5638
5639         /* Pack the entry anyway, at least the offset is right. */
5640         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5641                            it->oie_dirent->oied_name,
5642                            it->oie_dirent->oied_namelen,
5643                            it->oie_dirent->oied_type, attr);
5644
5645         if (rc < 0)
5646                 RETURN(rc);
5647
5648         if (osd_remote_fid(env, dev, fid))
5649                 RETURN(0);
5650
5651         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5652                 rc = osd_add_oi_cache(oti, dev, id, fid);
5653
5654         if (!(attr & LUDA_VERIFY) &&
5655             (scrub->os_pos_current <= ino) &&
5656             ((sf->sf_flags & SF_INCONSISTENT) ||
5657              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5658              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5659                 osd_consistency_check(oti, dev, oic);
5660
5661         RETURN(rc > 0 ? 0 : rc);
5662 }
5663
5664 /**
5665  * Returns the record size size at current position.
5666  *
5667  * This function will return record(lu_dirent) size in bytes.
5668  *
5669  * \param[in] env       execution environment
5670  * \param[in] di        iterator's in memory structure
5671  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5672  *                      calculate the lu_dirent size.
5673  *
5674  * \retval      record size(in bytes & in memory) of the current lu_dirent
5675  *              entry.
5676  */
5677 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5678                               __u32 attr)
5679 {
5680         struct osd_it_ea *it = (struct osd_it_ea *)di;
5681
5682         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5683 }
5684
5685 /**
5686  * Returns a cookie for current position of the iterator head, so that
5687  * user can use this cookie to load/start the iterator next time.
5688  *
5689  * \param di iterator's in memory structure
5690  *
5691  * \retval cookie for current position, on success
5692  */
5693 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5694 {
5695         struct osd_it_ea *it = (struct osd_it_ea *)di;
5696
5697         return it->oie_dirent->oied_off;
5698 }
5699
5700 /**
5701  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5702  * to load a directory entry at a time and stored it i inn,
5703  * in iterator's in-memory data structure.
5704  *
5705  * \param di struct osd_it_ea, iterator's in memory structure
5706  *
5707  * \retval +ve on success
5708  * \retval -ve on error
5709  */
5710 static int osd_it_ea_load(const struct lu_env *env,
5711                           const struct dt_it *di, __u64 hash)
5712 {
5713         struct osd_it_ea *it = (struct osd_it_ea *)di;
5714         int rc;
5715
5716         ENTRY;
5717         it->oie_file.f_pos = hash;
5718
5719         rc =  osd_ldiskfs_it_fill(env, di);
5720         if (rc > 0)
5721                 rc = -ENODATA;
5722
5723         if (rc == 0)
5724                 rc = +1;
5725
5726         RETURN(rc);
5727 }
5728
5729 /**
5730  * Index lookup function for interoperability mode (b11826).
5731  *
5732  * \param key,  key i.e. file name to be searched
5733  *
5734  * \retval +ve, on success
5735  * \retval -ve, on error
5736  */
5737 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5738                                struct dt_rec *rec, const struct dt_key *key,
5739                                struct lustre_capa *capa)
5740 {
5741         struct osd_object *obj = osd_dt_obj(dt);
5742         int rc = 0;
5743
5744         ENTRY;
5745
5746         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5747         LINVRNT(osd_invariant(obj));
5748
5749         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5750                 return -EACCES;
5751
5752         rc = osd_ea_lookup_rec(env, obj, rec, key);
5753         if (rc == 0)
5754                 rc = +1;
5755         RETURN(rc);
5756 }
5757
5758 /**
5759  * Index and Iterator operations for interoperability
5760  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5761  */
5762 static const struct dt_index_operations osd_index_ea_ops = {
5763         .dio_lookup         = osd_index_ea_lookup,
5764         .dio_declare_insert = osd_index_declare_ea_insert,
5765         .dio_insert         = osd_index_ea_insert,
5766         .dio_declare_delete = osd_index_declare_ea_delete,
5767         .dio_delete         = osd_index_ea_delete,
5768         .dio_it     = {
5769                 .init     = osd_it_ea_init,
5770                 .fini     = osd_it_ea_fini,
5771                 .get      = osd_it_ea_get,
5772                 .put      = osd_it_ea_put,
5773                 .next     = osd_it_ea_next,
5774                 .key      = osd_it_ea_key,
5775                 .key_size = osd_it_ea_key_size,
5776                 .rec      = osd_it_ea_rec,
5777                 .rec_size = osd_it_ea_rec_size,
5778                 .store    = osd_it_ea_store,
5779                 .load     = osd_it_ea_load
5780         }
5781 };
5782
5783 static void *osd_key_init(const struct lu_context *ctx,
5784                           struct lu_context_key *key)
5785 {
5786         struct osd_thread_info *info;
5787
5788         OBD_ALLOC_PTR(info);
5789         if (info == NULL)
5790                 return ERR_PTR(-ENOMEM);
5791
5792         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5793         if (info->oti_it_ea_buf == NULL)
5794                 goto out_free_info;
5795
5796         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5797
5798         info->oti_hlock = ldiskfs_htree_lock_alloc();
5799         if (info->oti_hlock == NULL)
5800                 goto out_free_ea;
5801
5802         return info;
5803
5804  out_free_ea:
5805         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5806  out_free_info:
5807         OBD_FREE_PTR(info);
5808         return ERR_PTR(-ENOMEM);
5809 }
5810
5811 static void osd_key_fini(const struct lu_context *ctx,
5812                          struct lu_context_key *key, void* data)
5813 {
5814         struct osd_thread_info *info = data;
5815
5816         if (info->oti_hlock != NULL)
5817                 ldiskfs_htree_lock_free(info->oti_hlock);
5818         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5819         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5820         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5821         lu_buf_free(&info->oti_big_buf);
5822         OBD_FREE_PTR(info);
5823 }
5824
5825 static void osd_key_exit(const struct lu_context *ctx,
5826                          struct lu_context_key *key, void *data)
5827 {
5828         struct osd_thread_info *info = data;
5829
5830         LASSERT(info->oti_r_locks == 0);
5831         LASSERT(info->oti_w_locks == 0);
5832         LASSERT(info->oti_txns    == 0);
5833 }
5834
5835 /* type constructor/destructor: osd_type_init, osd_type_fini */
5836 LU_TYPE_INIT_FINI(osd, &osd_key);
5837
5838 struct lu_context_key osd_key = {
5839         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5840         .lct_init = osd_key_init,
5841         .lct_fini = osd_key_fini,
5842         .lct_exit = osd_key_exit
5843 };
5844
5845
5846 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5847                            const char *name, struct lu_device *next)
5848 {
5849         struct osd_device *osd = osd_dev(d);
5850
5851         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5852             >= sizeof(osd->od_svname))
5853                 return -E2BIG;
5854         return osd_procfs_init(osd, name);
5855 }
5856
5857 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5858 {
5859         struct seq_server_site  *ss = osd_seq_site(osd);
5860         int                     rc;
5861         ENTRY;
5862
5863         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5864                 RETURN(0);
5865
5866         if (unlikely(ss == NULL))
5867                 RETURN(-ENODEV);
5868
5869         OBD_ALLOC_PTR(osd->od_cl_seq);
5870         if (osd->od_cl_seq == NULL)
5871                 RETURN(-ENOMEM);
5872
5873         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5874                              osd->od_svname, ss->ss_server_seq);
5875
5876         if (rc != 0) {
5877                 OBD_FREE_PTR(osd->od_cl_seq);
5878                 osd->od_cl_seq = NULL;
5879         }
5880
5881         RETURN(rc);
5882 }
5883
5884 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5885 {
5886         if (osd->od_cl_seq == NULL)
5887                 return;
5888
5889         seq_client_fini(osd->od_cl_seq);
5890         OBD_FREE_PTR(osd->od_cl_seq);
5891         osd->od_cl_seq = NULL;
5892 }
5893
5894 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5895 {
5896         ENTRY;
5897
5898         /* shutdown quota slave instance associated with the device */
5899         if (o->od_quota_slave != NULL) {
5900                 qsd_fini(env, o->od_quota_slave);
5901                 o->od_quota_slave = NULL;
5902         }
5903
5904         osd_fid_fini(env, o);
5905
5906         RETURN(0);
5907 }
5908
5909 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5910 {
5911         ENTRY;
5912
5913         if (o->od_mnt != NULL) {
5914                 shrink_dcache_sb(osd_sb(o));
5915                 osd_sync(env, &o->od_dt_dev);
5916
5917                 mntput(o->od_mnt);
5918                 o->od_mnt = NULL;
5919         }
5920
5921         EXIT;
5922 }
5923
5924 static int osd_mount(const struct lu_env *env,
5925                      struct osd_device *o, struct lustre_cfg *cfg)
5926 {
5927         const char              *name  = lustre_cfg_string(cfg, 0);
5928         const char              *dev  = lustre_cfg_string(cfg, 1);
5929         const char              *opts;
5930         unsigned long            page, s_flags, lmd_flags = 0;
5931         struct page             *__page;
5932         struct file_system_type *type;
5933         char                    *options = NULL;
5934         char                    *str;
5935         struct osd_thread_info  *info = osd_oti_get(env);
5936         struct lu_fid           *fid = &info->oti_fid;
5937         struct inode            *inode;
5938         int                      rc = 0;
5939         ENTRY;
5940
5941         if (o->od_mnt != NULL)
5942                 RETURN(0);
5943
5944         if (strlen(dev) >= sizeof(o->od_mntdev))
5945                 RETURN(-E2BIG);
5946         strcpy(o->od_mntdev, dev);
5947
5948         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5949         if (__page == NULL)
5950                 GOTO(out, rc = -ENOMEM);
5951
5952         str = lustre_cfg_string(cfg, 2);
5953         s_flags = simple_strtoul(str, NULL, 0);
5954         str = strstr(str, ":");
5955         if (str)
5956                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5957         opts = lustre_cfg_string(cfg, 3);
5958         page = (unsigned long)page_address(__page);
5959         options = (char *)page;
5960         *options = '\0';
5961         if (opts == NULL)
5962                 strcat(options, "user_xattr,acl");
5963         else
5964                 strcat(options, opts);
5965
5966         /* Glom up mount options */
5967         if (*options != '\0')
5968                 strcat(options, ",");
5969         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5970
5971         type = get_fs_type("ldiskfs");
5972         if (!type) {
5973                 CERROR("%s: cannot find ldiskfs module\n", name);
5974                 GOTO(out, rc = -ENODEV);
5975         }
5976
5977         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5978         module_put(type->owner);
5979
5980         if (IS_ERR(o->od_mnt)) {
5981                 rc = PTR_ERR(o->od_mnt);
5982                 o->od_mnt = NULL;
5983                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5984                 GOTO(out, rc);
5985         }
5986
5987 #ifdef HAVE_DEV_SET_RDONLY
5988         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5989                 CERROR("%s: underlying device %s is marked as read-only. "
5990                        "Setup failed\n", name, dev);
5991                 GOTO(out_mnt, rc = -EROFS);
5992         }
5993 #endif
5994
5995         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5996                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5997                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5998                 GOTO(out_mnt, rc = -EINVAL);
5999         }
6000
6001 #ifdef LDISKFS_MOUNT_DIRDATA
6002         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6003                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6004                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6005         else if (!o->od_is_ost)
6006                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6007                       "enabling the dirdata feature. If you do not want to "
6008                       "downgrade to Lustre-1.x again, you can enable it via "
6009                       "'tune2fs -O dirdata device'\n", name, dev);
6010 #endif
6011         inode = osd_sb(o)->s_root->d_inode;
6012         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6013         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6014         if (rc != 0) {
6015                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6016                 GOTO(out_mnt, rc);
6017         }
6018
6019         if (lmd_flags & LMD_FLG_NOSCRUB)
6020                 o->od_noscrub = 1;
6021
6022         GOTO(out, rc = 0);
6023
6024 out_mnt:
6025         mntput(o->od_mnt);
6026         o->od_mnt = NULL;
6027
6028 out:
6029         if (__page)
6030                 OBD_PAGE_FREE(__page);
6031
6032         return rc;
6033 }
6034
6035 static struct lu_device *osd_device_fini(const struct lu_env *env,
6036                                          struct lu_device *d)
6037 {
6038         struct osd_device *o = osd_dev(d);
6039         ENTRY;
6040
6041         osd_shutdown(env, o);
6042         osd_procfs_fini(o);
6043         osd_scrub_cleanup(env, o);
6044         osd_obj_map_fini(o);
6045         osd_umount(env, o);
6046
6047         RETURN(NULL);
6048 }
6049
6050 static int osd_device_init0(const struct lu_env *env,
6051                             struct osd_device *o,
6052                             struct lustre_cfg *cfg)
6053 {
6054         struct lu_device        *l = osd2lu_dev(o);
6055         struct osd_thread_info *info;
6056         int                     rc;
6057         int                     cplen = 0;
6058
6059         /* if the module was re-loaded, env can loose its keys */
6060         rc = lu_env_refill((struct lu_env *) env);
6061         if (rc)
6062                 GOTO(out, rc);
6063         info = osd_oti_get(env);
6064         LASSERT(info);
6065
6066         l->ld_ops = &osd_lu_ops;
6067         o->od_dt_dev.dd_ops = &osd_dt_ops;
6068
6069         spin_lock_init(&o->od_osfs_lock);
6070         mutex_init(&o->od_otable_mutex);
6071
6072         o->od_capa_hash = init_capa_hash();
6073         if (o->od_capa_hash == NULL)
6074                 GOTO(out, rc = -ENOMEM);
6075
6076         o->od_read_cache = 1;
6077         o->od_writethrough_cache = 1;
6078         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6079
6080         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6081                         sizeof(o->od_svname));
6082         if (cplen >= sizeof(o->od_svname)) {
6083                 rc = -E2BIG;
6084                 GOTO(out_capa, rc);
6085         }
6086
6087         if (server_name_is_ost(o->od_svname))
6088                 o->od_is_ost = 1;
6089
6090         rc = osd_mount(env, o, cfg);
6091         if (rc != 0)
6092                 GOTO(out_capa, rc);
6093
6094         rc = osd_obj_map_init(env, o);
6095         if (rc != 0)
6096                 GOTO(out_mnt, rc);
6097
6098         rc = lu_site_init(&o->od_site, l);
6099         if (rc != 0)
6100                 GOTO(out_compat, rc);
6101         o->od_site.ls_bottom_dev = l;
6102
6103         rc = lu_site_init_finish(&o->od_site);
6104         if (rc != 0)
6105                 GOTO(out_site, rc);
6106
6107         /* self-repair LMA by default */
6108         o->od_lma_self_repair = 1;
6109
6110         INIT_LIST_HEAD(&o->od_ios_list);
6111         /* setup scrub, including OI files initialization */
6112         rc = osd_scrub_setup(env, o);
6113         if (rc < 0)
6114                 GOTO(out_site, rc);
6115
6116         rc = osd_procfs_init(o, o->od_svname);
6117         if (rc != 0) {
6118                 CERROR("%s: can't initialize procfs: rc = %d\n",
6119                        o->od_svname, rc);
6120                 GOTO(out_scrub, rc);
6121         }
6122
6123         LASSERT(l->ld_site->ls_linkage.next != NULL);
6124         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6125
6126         /* initialize quota slave instance */
6127         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6128                                      o->od_proc_entry);
6129         if (IS_ERR(o->od_quota_slave)) {
6130                 rc = PTR_ERR(o->od_quota_slave);
6131                 o->od_quota_slave = NULL;
6132                 GOTO(out_procfs, rc);
6133         }
6134
6135         RETURN(0);
6136
6137 out_procfs:
6138         osd_procfs_fini(o);
6139 out_scrub:
6140         osd_scrub_cleanup(env, o);
6141 out_site:
6142         lu_site_fini(&o->od_site);
6143 out_compat:
6144         osd_obj_map_fini(o);
6145 out_mnt:
6146         osd_umount(env, o);
6147 out_capa:
6148         cleanup_capa_hash(o->od_capa_hash);
6149 out:
6150         return rc;
6151 }
6152
6153 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6154                                           struct lu_device_type *t,
6155                                           struct lustre_cfg *cfg)
6156 {
6157         struct osd_device *o;
6158         int                rc;
6159
6160         OBD_ALLOC_PTR(o);
6161         if (o == NULL)
6162                 return ERR_PTR(-ENOMEM);
6163
6164         rc = dt_device_init(&o->od_dt_dev, t);
6165         if (rc == 0) {
6166                 /* Because the ctx might be revived in dt_device_init,
6167                  * refill the env here */
6168                 lu_env_refill((struct lu_env *)env);
6169                 rc = osd_device_init0(env, o, cfg);
6170                 if (rc)
6171                         dt_device_fini(&o->od_dt_dev);
6172         }
6173
6174         if (unlikely(rc != 0))
6175                 OBD_FREE_PTR(o);
6176
6177         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6178 }
6179
6180 static struct lu_device *osd_device_free(const struct lu_env *env,
6181                                          struct lu_device *d)
6182 {
6183         struct osd_device *o = osd_dev(d);
6184         ENTRY;
6185
6186         cleanup_capa_hash(o->od_capa_hash);
6187         /* XXX: make osd top device in order to release reference */
6188         d->ld_site->ls_top_dev = d;
6189         lu_site_purge(env, d->ld_site, -1);
6190         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6191                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6192                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6193         }
6194         lu_site_fini(&o->od_site);
6195         dt_device_fini(&o->od_dt_dev);
6196         OBD_FREE_PTR(o);
6197         RETURN(NULL);
6198 }
6199
6200 static int osd_process_config(const struct lu_env *env,
6201                               struct lu_device *d, struct lustre_cfg *cfg)
6202 {
6203         struct osd_device               *o = osd_dev(d);
6204         int                             rc;
6205         ENTRY;
6206
6207         switch (cfg->lcfg_command) {
6208         case LCFG_SETUP:
6209                 rc = osd_mount(env, o, cfg);
6210                 break;
6211         case LCFG_CLEANUP:
6212                 lu_dev_del_linkage(d->ld_site, d);
6213                 rc = osd_shutdown(env, o);
6214                 break;
6215         case LCFG_PARAM:
6216                 LASSERT(&o->od_dt_dev);
6217                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6218                                               cfg, &o->od_dt_dev);
6219                 if (rc > 0 || rc == -ENOSYS)
6220                         rc = class_process_proc_param(PARAM_OST,
6221                                                       lprocfs_osd_obd_vars,
6222                                                       cfg, &o->od_dt_dev);
6223                 break;
6224         default:
6225                 rc = -ENOSYS;
6226         }
6227
6228         RETURN(rc);
6229 }
6230
6231 static int osd_recovery_complete(const struct lu_env *env,
6232                                  struct lu_device *d)
6233 {
6234         struct osd_device       *osd = osd_dev(d);
6235         int                      rc = 0;
6236         ENTRY;
6237
6238         if (osd->od_quota_slave == NULL)
6239                 RETURN(0);
6240
6241         /* start qsd instance on recovery completion, this notifies the quota
6242          * slave code that we are about to process new requests now */
6243         rc = qsd_start(env, osd->od_quota_slave);
6244         RETURN(rc);
6245 }
6246
6247 /*
6248  * we use exports to track all osd users
6249  */
6250 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6251                            struct obd_device *obd, struct obd_uuid *cluuid,
6252                            struct obd_connect_data *data, void *localdata)
6253 {
6254         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6255         struct lustre_handle  conn;
6256         int                   rc;
6257         ENTRY;
6258
6259         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6260
6261         rc = class_connect(&conn, obd, cluuid);
6262         if (rc)
6263                 RETURN(rc);
6264
6265         *exp = class_conn2export(&conn);
6266
6267         spin_lock(&osd->od_osfs_lock);
6268         osd->od_connects++;
6269         spin_unlock(&osd->od_osfs_lock);
6270
6271         RETURN(0);
6272 }
6273
6274 /*
6275  * once last export (we don't count self-export) disappeared
6276  * osd can be released
6277  */
6278 static int osd_obd_disconnect(struct obd_export *exp)
6279 {
6280         struct obd_device *obd = exp->exp_obd;
6281         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6282         int                rc, release = 0;
6283         ENTRY;
6284
6285         /* Only disconnect the underlying layers on the final disconnect. */
6286         spin_lock(&osd->od_osfs_lock);
6287         osd->od_connects--;
6288         if (osd->od_connects == 0)
6289                 release = 1;
6290         spin_unlock(&osd->od_osfs_lock);
6291
6292         rc = class_disconnect(exp); /* bz 9811 */
6293
6294         if (rc == 0 && release)
6295                 class_manual_cleanup(obd);
6296         RETURN(rc);
6297 }
6298
6299 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6300                        struct lu_device *dev)
6301 {
6302         struct osd_device *osd = osd_dev(dev);
6303         int                result = 0;
6304         ENTRY;
6305
6306         if (osd->od_quota_slave != NULL) {
6307                 /* set up quota slave objects */
6308                 result = qsd_prepare(env, osd->od_quota_slave);
6309                 if (result != 0)
6310                         RETURN(result);
6311         }
6312
6313         result = osd_fid_init(env, osd);
6314
6315         RETURN(result);
6316 }
6317
6318 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6319                   struct lu_fid *fid, struct md_op_data *op_data)
6320 {
6321         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6322
6323         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6324 }
6325
6326 static const struct lu_object_operations osd_lu_obj_ops = {
6327         .loo_object_init      = osd_object_init,
6328         .loo_object_delete    = osd_object_delete,
6329         .loo_object_release   = osd_object_release,
6330         .loo_object_free      = osd_object_free,
6331         .loo_object_print     = osd_object_print,
6332         .loo_object_invariant = osd_object_invariant
6333 };
6334
6335 const struct lu_device_operations osd_lu_ops = {
6336         .ldo_object_alloc      = osd_object_alloc,
6337         .ldo_process_config    = osd_process_config,
6338         .ldo_recovery_complete = osd_recovery_complete,
6339         .ldo_prepare           = osd_prepare,
6340 };
6341
6342 static const struct lu_device_type_operations osd_device_type_ops = {
6343         .ldto_init = osd_type_init,
6344         .ldto_fini = osd_type_fini,
6345
6346         .ldto_start = osd_type_start,
6347         .ldto_stop  = osd_type_stop,
6348
6349         .ldto_device_alloc = osd_device_alloc,
6350         .ldto_device_free  = osd_device_free,
6351
6352         .ldto_device_init    = osd_device_init,
6353         .ldto_device_fini    = osd_device_fini
6354 };
6355
6356 struct lu_device_type osd_device_type = {
6357         .ldt_tags     = LU_DEVICE_DT,
6358         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6359         .ldt_ops      = &osd_device_type_ops,
6360         .ldt_ctx_tags = LCT_LOCAL,
6361 };
6362
6363 /*
6364  * lprocfs legacy support.
6365  */
6366 static struct obd_ops osd_obd_device_ops = {
6367         .o_owner = THIS_MODULE,
6368         .o_connect      = osd_obd_connect,
6369         .o_disconnect   = osd_obd_disconnect,
6370         .o_fid_alloc    = osd_fid_alloc,
6371 };
6372
6373 static int __init osd_mod_init(void)
6374 {
6375         int rc;
6376
6377         osd_oi_mod_init();
6378
6379         rc = lu_kmem_init(ldiskfs_caches);
6380         if (rc)
6381                 return rc;
6382
6383         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6384                                  lprocfs_osd_module_vars,
6385                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6386         if (rc)
6387                 lu_kmem_fini(ldiskfs_caches);
6388         return rc;
6389 }
6390
6391 static void __exit osd_mod_exit(void)
6392 {
6393         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6394         lu_kmem_fini(ldiskfs_caches);
6395 }
6396
6397 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6398 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6399 MODULE_LICENSE("GPL");
6400
6401 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);