lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76 #include <lustre_linkea.h>
  77
  78 int ldiskfs_pdo = 1;
  79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  80                 "ldiskfs with parallel directory operations");
  81
  82 int ldiskfs_track_declares_assert;
  83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  84                 "LBUG during tracking of declares");
  85
  86 /* Slab to allocate dynlocks */
  87 struct kmem_cache *dynlock_cachep;
  88
  89 static struct lu_kmem_descr ldiskfs_caches[] = {
  90         {
  91                 .ckd_cache = &dynlock_cachep,
  92                 .ckd_name  = "dynlock_cache",
  93                 .ckd_size  = sizeof(struct dynlock_handle)
  94         },
  95         {
  96                 .ckd_cache = NULL
  97         }
  98 };
  99
 100 static const char dot[] = ".";
 101 static const char dotdot[] = "..";
 102 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 103
 104 static const struct lu_object_operations      osd_lu_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ops;
 106 static const struct dt_object_operations      osd_obj_ea_ops;
 107 static const struct dt_object_operations      osd_obj_otable_it_ops;
 108 static const struct dt_index_operations       osd_index_iam_ops;
 109 static const struct dt_index_operations       osd_index_ea_ops;
 110
 111 int osd_trans_declare_op2rb[] = {
 112         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 113         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 114         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 115         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 116         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 117         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 118         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 119         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 120         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 121         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 122         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 123         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 124 };
 125
 126 static int osd_has_index(const struct osd_object *obj)
 127 {
 128         return obj->oo_dt.do_index_ops != NULL;
 129 }
 130
 131 static int osd_object_invariant(const struct lu_object *l)
 132 {
 133         return osd_invariant(osd_obj(l));
 134 }
 135
 136 /*
 137  * Concurrency: doesn't matter
 138  */
 139
 140 /*
 141  * Concurrency: doesn't matter
 142  */
 143 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 144 {
 145         struct osd_thread_info *oti = osd_oti_get(env);
 146         return oti->oti_w_locks > 0 && o->oo_owner == env;
 147 }
 148
 149 /*
 150  * Concurrency: doesn't access mutable data
 151  */
 152 static int osd_root_get(const struct lu_env *env,
 153                         struct dt_device *dev, struct lu_fid *f)
 154 {
 155         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 156         return 0;
 157 }
 158
 159 /*
 160  * OSD object methods.
 161  */
 162
 163 /*
 164  * Concurrency: no concurrent access is possible that early in object
 165  * life-cycle.
 166  */
 167 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 168                                           const struct lu_object_header *hdr,
 169                                           struct lu_device *d)
 170 {
 171         struct osd_object *mo;
 172
 173         OBD_ALLOC_PTR(mo);
 174         if (mo != NULL) {
 175                 struct lu_object *l;
 176
 177                 l = &mo->oo_dt.do_lu;
 178                 dt_object_init(&mo->oo_dt, NULL, d);
 179                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 180                 l->lo_ops = &osd_lu_obj_ops;
 181                 init_rwsem(&mo->oo_sem);
 182                 init_rwsem(&mo->oo_ext_idx_sem);
 183                 spin_lock_init(&mo->oo_guard);
 184                 return l;
 185         } else {
 186                 return NULL;
 187         }
 188 }
 189
 190 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 191                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 192 {
 193         int rc;
 194
 195         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 196         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 197                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 198         if (rc > 0) {
 199                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 200                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 201                 rc = 0;
 202                 lustre_lma_swab(lma);
 203                 /* Check LMA compatibility */
 204                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 205                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 206                               "for fid = "DFID", ino = %lu\n",
 207                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 208                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 209                               PFID(&lma->lma_self_fid), inode->i_ino);
 210                         rc = -EOPNOTSUPP;
 211                 }
 212         } else if (rc == 0) {
 213                 rc = -ENODATA;
 214         }
 215
 216         return rc;
 217 }
 218
 219 /*
 220  * retrieve object from backend ext fs.
 221  **/
 222 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 223                        struct osd_inode_id *id)
 224 {
 225         struct inode *inode = NULL;
 226
 227         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 228         if (IS_ERR(inode)) {
 229                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 230                        id->oii_ino, PTR_ERR(inode));
 231         } else if (id->oii_gen != OSD_OII_NOGEN &&
 232                    inode->i_generation != id->oii_gen) {
 233                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 234                        "i_generation = %u\n",
 235                        id->oii_ino, id->oii_gen, inode->i_generation);
 236                 iput(inode);
 237                 inode = ERR_PTR(-ESTALE);
 238         } else if (inode->i_nlink == 0) {
 239                 /* due to parallel readdir and unlink,
 240                 * we can have dead inode here. */
 241                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 242                 iput(inode);
 243                 inode = ERR_PTR(-ESTALE);
 244         } else if (is_bad_inode(inode)) {
 245                 CWARN("%.16s: bad inode: ino = %u\n",
 246                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 247                 iput(inode);
 248                 inode = ERR_PTR(-ENOENT);
 249         } else {
 250                 if (id->oii_gen == OSD_OII_NOGEN)
 251                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 252
 253                 /* Do not update file c/mtime in ldiskfs.
 254                  * NB: we don't have any lock to protect this because we don't
 255                  * have reference on osd_object now, but contention with
 256                  * another lookup + attr_set can't happen in the tiny window
 257                  * between if (...) and set S_NOCMTIME. */
 258                 if (!(inode->i_flags & S_NOCMTIME))
 259                         inode->i_flags |= S_NOCMTIME;
 260         }
 261         return inode;
 262 }
 263
 264 static struct inode *
 265 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 266              struct osd_inode_id *id, struct lu_fid *fid)
 267 {
 268         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 269         struct inode            *inode;
 270         int                      rc;
 271
 272         inode = osd_iget(info, dev, id);
 273         if (IS_ERR(inode))
 274                 return inode;
 275
 276         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 277         if (rc == 0) {
 278                 *fid = lma->lma_self_fid;
 279         } else if (rc == -ENODATA) {
 280                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 281                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 282                 else
 283                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 284         } else {
 285                 iput(inode);
 286                 inode = ERR_PTR(rc);
 287         }
 288         return inode;
 289 }
 290
 291 static struct inode *osd_iget_check(struct osd_thread_info *info,
 292                                     struct osd_device *dev,
 293                                     const struct lu_fid *fid,
 294                                     struct osd_inode_id *id,
 295                                     bool in_oi)
 296 {
 297         struct inode    *inode;
 298         int              rc     = 0;
 299         ENTRY;
 300
 301         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 302         if (IS_ERR(inode)) {
 303                 rc = PTR_ERR(inode);
 304                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 305                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 306                                id->oii_ino, rc);
 307
 308                         GOTO(put, rc);
 309                 }
 310
 311                 goto check_oi;
 312         }
 313
 314         if (is_bad_inode(inode)) {
 315                 rc = -ENOENT;
 316                 if (!in_oi) {
 317                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 318
 319                         GOTO(put, rc);
 320                 }
 321
 322                 goto check_oi;
 323         }
 324
 325         if (id->oii_gen != OSD_OII_NOGEN &&
 326             inode->i_generation != id->oii_gen) {
 327                 rc = -ESTALE;
 328                 if (!in_oi) {
 329                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 330                                "oii_gen = %u, i_generation = %u\n",
 331                                id->oii_ino, id->oii_gen, inode->i_generation);
 332
 333                         GOTO(put, rc);
 334                 }
 335
 336                 goto check_oi;
 337         }
 338
 339         if (inode->i_nlink == 0) {
 340                 rc = -ENOENT;
 341                 if (!in_oi) {
 342                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 343
 344                         GOTO(put, rc);
 345                 }
 346
 347                 goto check_oi;
 348         }
 349
 350 check_oi:
 351         if (rc != 0) {
 352                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 353
 354                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 355                 /* XXX: There are some possible cases:
 356                  *      1. rc = 0.
 357                  *         Backup/restore caused the OI invalid.
 358                  *      2. rc = 0.
 359                  *         Someone unlinked the object but NOT removed
 360                  *         the OI mapping, such as mount target device
 361                  *         as ldiskfs, and modify something directly.
 362                  *      3. rc = -ENOENT.
 363                  *         Someone just removed the object between the
 364                  *         former oi_lookup and the iget. It is normal.
 365                  *      4. Other failure cases.
 366                  *
 367                  *      Generally, when the device is mounted, it will
 368                  *      auto check whether the system is restored from
 369                  *      file-level backup or not. We trust such detect
 370                  *      to distinguish the 1st case from the 2nd case. */
 371                 if (rc == 0) {
 372                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 373                             inode->i_generation == id->oii_gen)
 374                                 rc = -ENOENT;
 375                         else
 376                                 rc = -EREMCHG;
 377                 }
 378         } else {
 379                 if (id->oii_gen == OSD_OII_NOGEN)
 380                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 381
 382                 /* Do not update file c/mtime in ldiskfs.
 383                  * NB: we don't have any lock to protect this because we don't
 384                  * have reference on osd_object now, but contention with
 385                  * another lookup + attr_set can't happen in the tiny window
 386                  * between if (...) and set S_NOCMTIME. */
 387                 if (!(inode->i_flags & S_NOCMTIME))
 388                         inode->i_flags |= S_NOCMTIME;
 389         }
 390
 391         GOTO(put, rc);
 392
 393 put:
 394         if (rc != 0) {
 395                 if (!IS_ERR(inode))
 396                         iput(inode);
 397
 398                 inode = ERR_PTR(rc);
 399         }
 400
 401         return inode;
 402 }
 403
 404 /**
 405  * \retval +v: new filter_fid, does not contain self-fid
 406  * \retval 0:  filter_fid_old, contains self-fid
 407  * \retval -v: other failure cases
 408  */
 409 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 410                  struct dentry *dentry, struct lu_fid *fid)
 411 {
 412         struct filter_fid_old   *ff     = &info->oti_ff;
 413         struct ost_id           *ostid  = &info->oti_ostid;
 414         int                      rc;
 415
 416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 417         if (rc == sizeof(*ff)) {
 418                 rc = 0;
 419                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 420                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 421                 /* XXX: should use real OST index in the future. LU-3569 */
 422                 ostid_to_fid(fid, ostid, 0);
 423         } else if (rc == sizeof(struct filter_fid)) {
 424                 rc = 1;
 425         } else if (rc >= 0) {
 426                 rc = -EINVAL;
 427         }
 428
 429         return rc;
 430 }
 431
 432 static int osd_lma_self_repair(struct osd_thread_info *info,
 433                                struct osd_device *osd, struct inode *inode,
 434                                const struct lu_fid *fid, __u32 compat)
 435 {
 436         handle_t *jh;
 437         int       rc;
 438
 439         LASSERT(current->journal_info == NULL);
 440
 441         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 442                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 443         if (IS_ERR(jh)) {
 444                 rc = PTR_ERR(jh);
 445                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 446                       osd_name(osd), rc);
 447                 return rc;
 448         }
 449
 450         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 451         if (rc != 0)
 452                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 453                       osd_name(osd), rc);
 454         ldiskfs_journal_stop(jh);
 455         return rc;
 456 }
 457
 458 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 459 {
 460         struct osd_thread_info  *info   = osd_oti_get(env);
 461         struct osd_device       *osd    = osd_obj2dev(obj);
 462         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 463         struct inode            *inode  = obj->oo_inode;
 464         struct dentry           *dentry = &info->oti_obj_dentry;
 465         struct lu_fid           *fid    = NULL;
 466         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 467         int                      rc;
 468         ENTRY;
 469
 470         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 471         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 472                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 473         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 474                 fid = &lma->lma_self_fid;
 475                 rc = osd_get_idif(info, inode, dentry, fid);
 476                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 477                         /* For the given OST-object, if it has neither LMA nor
 478                          * FID in XATTR_NAME_FID, then the given FID (which is
 479                          * contained in the @obj, from client RPC for locating
 480                          * the OST-object) is trusted. We use it to generate
 481                          * the LMA. */
 482                         osd_lma_self_repair(info, osd, inode, rfid,
 483                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 484                                 LMAC_FID_ON_OST : 0);
 485                         RETURN(0);
 486                 }
 487         }
 488
 489         if (unlikely(rc == -ENODATA))
 490                 RETURN(0);
 491
 492         if (rc < 0)
 493                 RETURN(rc);
 494
 495         if (rc > 0) {
 496                 rc = 0;
 497                 lustre_lma_swab(lma);
 498                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 499                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 500                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 501                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 502                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 503                               PFID(rfid), inode->i_ino);
 504                         rc = -EOPNOTSUPP;
 505                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 506                         fid = &lma->lma_self_fid;
 507                 }
 508         }
 509
 510         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 511                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 512                         struct ost_id   *oi   = &info->oti_ostid;
 513                         struct lu_fid   *fid1 = &info->oti_fid3;
 514                         __u32            idx  = fid_idif_ost_idx(rfid);
 515
 516                         /* For old IDIF, the OST index is not part of the IDIF,
 517                          * Means that different OSTs may have the same IDIFs.
 518                          * Under such case, we need to make some compatible
 519                          * check to make sure to trigger OI scrub properly. */
 520                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 521                                 /* Given @rfid is new, LMA is old. */
 522                                 fid_to_ostid(fid, oi);
 523                                 ostid_to_fid(fid1, oi, idx);
 524                                 if (lu_fid_eq(fid1, rfid)) {
 525                                         if (osd->od_lma_self_repair)
 526                                                 osd_lma_self_repair(info, osd,
 527                                                         inode, rfid,
 528                                                         LMAC_FID_ON_OST);
 529                                         RETURN(0);
 530                                 }
 531                         }
 532                 }
 533
 534                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 535                        osd_name(osd), PFID(rfid), PFID(fid));
 536                 rc = -EREMCHG;
 537         }
 538
 539         RETURN(rc);
 540 }
 541
 542 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 543                           const struct lu_fid *fid,
 544                           const struct lu_object_conf *conf)
 545 {
 546         struct osd_thread_info *info;
 547         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 548         struct osd_device      *dev;
 549         struct osd_idmap_cache *oic;
 550         struct osd_inode_id    *id;
 551         struct inode           *inode;
 552         struct osd_scrub       *scrub;
 553         struct scrub_file      *sf;
 554         int                     result;
 555         int                     saved  = 0;
 556         bool                    in_oi  = false;
 557         bool                    triggered = false;
 558         ENTRY;
 559
 560         LINVRNT(osd_invariant(obj));
 561         LASSERT(obj->oo_inode == NULL);
 562         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 563
 564         dev = osd_dev(ldev);
 565         scrub = &dev->od_scrub;
 566         sf = &scrub->os_file;
 567         info = osd_oti_get(env);
 568         LASSERT(info);
 569         oic = &info->oti_cache;
 570
 571         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 572                 RETURN(-ENOENT);
 573
 574         /* For the object is created as locking anchor, or for the object to
 575          * be created on disk. No need to osd_oi_lookup() at here because FID
 576          * shouldn't never be re-used, if it's really a duplicate FID from
 577          * unexpected reason, we should be able to detect it later by calling
 578          * do_create->osd_oi_insert(). */
 579         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 580                 GOTO(out, result = 0);
 581
 582         /* Search order: 1. per-thread cache. */
 583         if (lu_fid_eq(fid, &oic->oic_fid) &&
 584             likely(oic->oic_dev == dev)) {
 585                 id = &oic->oic_lid;
 586                 goto iget;
 587         }
 588
 589         id = &info->oti_id;
 590         if (!list_empty(&scrub->os_inconsistent_items)) {
 591                 /* Search order: 2. OI scrub pending list. */
 592                 result = osd_oii_lookup(dev, fid, id);
 593                 if (result == 0)
 594                         goto iget;
 595         }
 596
 597         /* Search order: 3. OI files. */
 598         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 599         if (result == -ENOENT) {
 600                 if (!fid_is_norm(fid) ||
 601                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 602                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 603                                       sf->sf_oi_bitmap))
 604                         GOTO(out, result = 0);
 605
 606                 goto trigger;
 607         }
 608
 609         if (result != 0)
 610                 GOTO(out, result);
 611
 612         in_oi = true;
 613
 614 iget:
 615         inode = osd_iget_check(info, dev, fid, id, in_oi);
 616         if (IS_ERR(inode)) {
 617                 result = PTR_ERR(inode);
 618                 if (result == -ENOENT || result == -ESTALE) {
 619                         if (!in_oi)
 620                                 fid_zero(&oic->oic_fid);
 621
 622                         GOTO(out, result = -ENOENT);
 623                 } else if (result == -EREMCHG) {
 624
 625 trigger:
 626                         if (!in_oi)
 627                                 fid_zero(&oic->oic_fid);
 628
 629                         if (unlikely(triggered))
 630                                 GOTO(out, result = saved);
 631
 632                         triggered = true;
 633                         if (thread_is_running(&scrub->os_thread)) {
 634                                 result = -EINPROGRESS;
 635                         } else if (!dev->od_noscrub) {
 636                                 result = osd_scrub_start(dev);
 637                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 638                                               "for "DFID", rc = %d [1]\n",
 639                                               osd_name(dev), PFID(fid),result);
 640                                 if (result == 0 || result == -EALREADY)
 641                                         result = -EINPROGRESS;
 642                                 else
 643                                         result = -EREMCHG;
 644                         }
 645
 646                         /* We still have chance to get the valid inode: for the
 647                          * object which is referenced by remote name entry, the
 648                          * object on the local MDT will be linked under the dir
 649                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 650                          *
 651                          * We do not know whether the object for the given FID
 652                          * is referenced by some remote name entry or not, and
 653                          * especially for DNE II, a multiple-linked object may
 654                          * have many name entries reside on many MDTs.
 655                          *
 656                          * To simplify the operation, OSD will not distinguish
 657                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 658                          * only happened for the RPC from other MDT during the
 659                          * OI scrub, or for the client side RPC with FID only,
 660                          * such as FID to path, or from old connected client. */
 661                         saved = result;
 662                         result = osd_lookup_in_remote_parent(info, dev,
 663                                                              fid, id);
 664                         if (result == 0) {
 665                                 in_oi = false;
 666                                 goto iget;
 667                         }
 668
 669                         result = saved;
 670                 }
 671
 672                 GOTO(out, result);
 673         }
 674
 675         obj->oo_inode = inode;
 676         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 677
 678         result = osd_check_lma(env, obj);
 679         if (result != 0) {
 680                 iput(inode);
 681                 obj->oo_inode = NULL;
 682                 if (result == -EREMCHG) {
 683                         if (!in_oi) {
 684                                 result = osd_oi_lookup(info, dev, fid, id,
 685                                                        OI_CHECK_FLD);
 686                                 if (result != 0) {
 687                                         fid_zero(&oic->oic_fid);
 688                                         GOTO(out, result);
 689                                 }
 690                         }
 691
 692                         goto trigger;
 693                 }
 694
 695                 GOTO(out, result);
 696         }
 697
 698         obj->oo_compat_dot_created = 1;
 699         obj->oo_compat_dotdot_created = 1;
 700
 701         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 702                 GOTO(out, result = 0);
 703
 704         LASSERT(obj->oo_hl_head == NULL);
 705         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 706         if (obj->oo_hl_head == NULL) {
 707                 obj->oo_inode = NULL;
 708                 iput(inode);
 709                 GOTO(out, result = -ENOMEM);
 710         }
 711         GOTO(out, result = 0);
 712
 713 out:
 714         LINVRNT(osd_invariant(obj));
 715         return result;
 716 }
 717
 718 /*
 719  * Concurrency: shouldn't matter.
 720  */
 721 static void osd_object_init0(struct osd_object *obj)
 722 {
 723         LASSERT(obj->oo_inode != NULL);
 724         obj->oo_dt.do_body_ops = &osd_body_ops;
 725         obj->oo_dt.do_lu.lo_header->loh_attr |=
 726                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 727 }
 728
 729 /*
 730  * Concurrency: no concurrent access is possible that early in object
 731  * life-cycle.
 732  */
 733 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 734                            const struct lu_object_conf *conf)
 735 {
 736         struct osd_object *obj = osd_obj(l);
 737         int result;
 738
 739         LINVRNT(osd_invariant(obj));
 740
 741         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 742                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 743                 l->lo_header->loh_attr |= LOHA_EXISTS;
 744                 return 0;
 745         }
 746
 747         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 748         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 749         if (result == 0 && obj->oo_inode != NULL)
 750                 osd_object_init0(obj);
 751
 752         LINVRNT(osd_invariant(obj));
 753         return result;
 754 }
 755
 756 /*
 757  * Concurrency: no concurrent access is possible that late in object
 758  * life-cycle.
 759  */
 760 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 761 {
 762         struct osd_object *obj = osd_obj(l);
 763
 764         LINVRNT(osd_invariant(obj));
 765
 766         dt_object_fini(&obj->oo_dt);
 767         if (obj->oo_hl_head != NULL)
 768                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 769         OBD_FREE_PTR(obj);
 770 }
 771
 772 /*
 773  * Concurrency: no concurrent access is possible that late in object
 774  * life-cycle.
 775  */
 776 static void osd_index_fini(struct osd_object *o)
 777 {
 778         struct iam_container *bag;
 779
 780         if (o->oo_dir != NULL) {
 781                 bag = &o->oo_dir->od_container;
 782                 if (o->oo_inode != NULL) {
 783                         if (bag->ic_object == o->oo_inode)
 784                                 iam_container_fini(bag);
 785                 }
 786                 OBD_FREE_PTR(o->oo_dir);
 787                 o->oo_dir = NULL;
 788         }
 789 }
 790
 791 /*
 792  * Concurrency: no concurrent access is possible that late in object
 793  * life-cycle (for all existing callers, that is. New callers have to provide
 794  * their own locking.)
 795  */
 796 static int osd_inode_unlinked(const struct inode *inode)
 797 {
 798         return inode->i_nlink == 0;
 799 }
 800
 801 enum {
 802         OSD_TXN_OI_DELETE_CREDITS    = 20,
 803         OSD_TXN_INODE_DELETE_CREDITS = 20
 804 };
 805
 806 /*
 807  * Journal
 808  */
 809
 810 #if OSD_THANDLE_STATS
 811 /**
 812  * Set time when the handle is allocated
 813  */
 814 static void osd_th_alloced(struct osd_thandle *oth)
 815 {
 816         oth->oth_alloced = cfs_time_current();
 817 }
 818
 819 /**
 820  * Set time when the handle started
 821  */
 822 static void osd_th_started(struct osd_thandle *oth)
 823 {
 824         oth->oth_started = cfs_time_current();
 825 }
 826
 827 /**
 828  * Helper function to convert time interval to microseconds packed in
 829  * long int.
 830  */
 831 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 832 {
 833         struct timeval val;
 834
 835         cfs_duration_usec(cfs_time_sub(end, start), &val);
 836         return val.tv_sec * 1000000 + val.tv_usec;
 837 }
 838
 839 /**
 840  * Check whether the we deal with this handle for too long.
 841  */
 842 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 843                                 cfs_time_t alloced, cfs_time_t started,
 844                                 cfs_time_t closed)
 845 {
 846         cfs_time_t now = cfs_time_current();
 847
 848         LASSERT(dev != NULL);
 849
 850         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 851                             interval_to_usec(alloced, started));
 852         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 853                             interval_to_usec(started, closed));
 854         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 855                             interval_to_usec(closed, now));
 856
 857         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 858                 CWARN("transaction handle %p was open for too long: "
 859                       "now "CFS_TIME_T" ,"
 860                       "alloced "CFS_TIME_T" ,"
 861                       "started "CFS_TIME_T" ,"
 862                       "closed "CFS_TIME_T"\n",
 863                       oth, now, alloced, started, closed);
 864                 libcfs_debug_dumpstack(NULL);
 865         }
 866 }
 867
 868 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 869 {                                                                       \
 870         cfs_time_t __closed = cfs_time_current();                       \
 871         cfs_time_t __alloced = oth->oth_alloced;                        \
 872         cfs_time_t __started = oth->oth_started;                        \
 873                                                                         \
 874         expr;                                                           \
 875         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 876 }
 877
 878 #else /* OSD_THANDLE_STATS */
 879
 880 #define osd_th_alloced(h)                  do {} while(0)
 881 #define osd_th_started(h)                  do {} while(0)
 882 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 883
 884 #endif /* OSD_THANDLE_STATS */
 885
 886 /*
 887  * Concurrency: doesn't access mutable data.
 888  */
 889 static int osd_param_is_not_sane(const struct osd_device *dev,
 890                                  const struct thandle *th)
 891 {
 892         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 893
 894         return oh->ot_credits > osd_transaction_size(dev);
 895 }
 896
 897 /*
 898  * Concurrency: shouldn't matter.
 899  */
 900 static void osd_trans_commit_cb(struct super_block *sb,
 901                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 902 {
 903         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 904         struct thandle     *th  = &oh->ot_super;
 905         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 906         struct dt_txn_commit_cb *dcb, *tmp;
 907
 908         LASSERT(oh->ot_handle == NULL);
 909
 910         if (error)
 911                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 912
 913         dt_txn_hook_commit(th);
 914
 915         /* call per-transaction callbacks if any */
 916         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 917                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 918                          "commit callback entry: magic=%x name='%s'\n",
 919                          dcb->dcb_magic, dcb->dcb_name);
 920                 list_del_init(&dcb->dcb_linkage);
 921                 dcb->dcb_func(NULL, th, dcb, error);
 922         }
 923
 924         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 925         lu_device_put(lud);
 926         th->th_dev = NULL;
 927
 928         lu_context_exit(&th->th_ctx);
 929         lu_context_fini(&th->th_ctx);
 930         thandle_put(th);
 931 }
 932
 933 static struct thandle *osd_trans_create(const struct lu_env *env,
 934                                         struct dt_device *d)
 935 {
 936         struct osd_thread_info  *oti = osd_oti_get(env);
 937         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 938         struct osd_thandle      *oh;
 939         struct thandle          *th;
 940         ENTRY;
 941
 942         /* on pending IO in this thread should left from prev. request */
 943         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 944
 945         th = ERR_PTR(-ENOMEM);
 946         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 947         if (oh != NULL) {
 948                 oh->ot_quota_trans = &oti->oti_quota_trans;
 949                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 950                 th = &oh->ot_super;
 951                 th->th_dev = d;
 952                 th->th_result = 0;
 953                 th->th_tags = LCT_TX_HANDLE;
 954                 oh->ot_credits = 0;
 955                 atomic_set(&th->th_refc, 1);
 956                 th->th_alloc_size = sizeof(*oh);
 957                 oti->oti_dev = osd_dt_dev(d);
 958                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 959                 osd_th_alloced(oh);
 960
 961                 memset(oti->oti_declare_ops, 0,
 962                                         sizeof(oti->oti_declare_ops));
 963                 memset(oti->oti_declare_ops_rb, 0,
 964                                         sizeof(oti->oti_declare_ops_rb));
 965                 memset(oti->oti_declare_ops_cred, 0,
 966                                         sizeof(oti->oti_declare_ops_cred));
 967                 oti->oti_rollback = false;
 968         }
 969         RETURN(th);
 970 }
 971
 972 /*
 973  * Concurrency: shouldn't matter.
 974  */
 975 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 976                     struct thandle *th)
 977 {
 978         struct osd_thread_info *oti = osd_oti_get(env);
 979         struct osd_device  *dev = osd_dt_dev(d);
 980         handle_t           *jh;
 981         struct osd_thandle *oh;
 982         int rc;
 983
 984         ENTRY;
 985
 986         LASSERT(current->journal_info == NULL);
 987
 988         oh = container_of0(th, struct osd_thandle, ot_super);
 989         LASSERT(oh != NULL);
 990         LASSERT(oh->ot_handle == NULL);
 991
 992         rc = dt_txn_hook_start(env, d, th);
 993         if (rc != 0)
 994                 GOTO(out, rc);
 995
 996         if (unlikely(osd_param_is_not_sane(dev, th))) {
 997                 static unsigned long last_printed;
 998                 static int last_credits;
 999
1000                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1001                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1002                       oh->ot_credits,
1003                       osd_journal(dev)->j_max_transaction_buffers);
1004                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1005                       oti->oti_declare_ops[OSD_OT_CREATE],
1006                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1007                       oti->oti_declare_ops[OSD_OT_DESTROY],
1008                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1009                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1010                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1012                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1013                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1014                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1015                       oti->oti_declare_ops[OSD_OT_WRITE],
1016                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1017                       oti->oti_declare_ops[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1019                       oti->oti_declare_ops[OSD_OT_QUOTA],
1020                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1021                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1022                       oti->oti_declare_ops[OSD_OT_INSERT],
1023                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1024                       oti->oti_declare_ops[OSD_OT_DELETE],
1025                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1026                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1027                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1029                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1030                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1031
1032                 if (last_credits != oh->ot_credits &&
1033                     time_after(jiffies, last_printed +
1034                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1035                         libcfs_debug_dumpstack(NULL);
1036                         last_credits = oh->ot_credits;
1037                         last_printed = jiffies;
1038                 }
1039                 /* XXX Limit the credits to 'max_transaction_buffers', and
1040                  *     let the underlying filesystem to catch the error if
1041                  *     we really need so many credits.
1042                  *
1043                  *     This should be removed when we can calculate the
1044                  *     credits precisely. */
1045                 oh->ot_credits = osd_transaction_size(dev);
1046         }
1047
1048         /*
1049          * XXX temporary stuff. Some abstraction layer should
1050          * be used.
1051          */
1052         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1053         osd_th_started(oh);
1054         if (!IS_ERR(jh)) {
1055                 oh->ot_handle = jh;
1056                 LASSERT(oti->oti_txns == 0);
1057                 lu_context_init(&th->th_ctx, th->th_tags);
1058                 lu_context_enter(&th->th_ctx);
1059
1060                 lu_device_get(&d->dd_lu_dev);
1061                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1062                               "osd-tx", th);
1063                 oti->oti_txns++;
1064                 rc = 0;
1065         } else {
1066                 rc = PTR_ERR(jh);
1067         }
1068 out:
1069         RETURN(rc);
1070 }
1071
1072 static int osd_seq_exists(const struct lu_env *env,
1073                               struct osd_device *osd, obd_seq seq)
1074 {
1075         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1076         struct seq_server_site  *ss = osd_seq_site(osd);
1077         int                     rc;
1078         ENTRY;
1079
1080         LASSERT(ss != NULL);
1081         LASSERT(ss->ss_server_fld != NULL);
1082
1083         rc = osd_fld_lookup(env, osd, seq, range);
1084         if (rc != 0) {
1085                 if (rc != -ENOENT)
1086                         CERROR("%s: can't lookup FLD sequence "LPX64
1087                                ": rc = %d\n", osd_name(osd), seq, rc);
1088                 RETURN(0);
1089         }
1090
1091         RETURN(ss->ss_node_id == range->lsr_index);
1092 }
1093
1094 /*
1095  * Concurrency: shouldn't matter.
1096  */
1097 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1098                           struct thandle *th)
1099 {
1100         int                     rc = 0;
1101         struct osd_thandle     *oh;
1102         struct osd_thread_info *oti = osd_oti_get(env);
1103         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1104         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1105         struct lquota_trans    *qtrans;
1106         ENTRY;
1107
1108         oh = container_of0(th, struct osd_thandle, ot_super);
1109
1110         qtrans = oh->ot_quota_trans;
1111         oh->ot_quota_trans = NULL;
1112
1113         if (oh->ot_handle != NULL) {
1114                 handle_t *hdl = oh->ot_handle;
1115
1116                 /*
1117                  * add commit callback
1118                  * notice we don't do this in osd_trans_start()
1119                  * as underlying transaction can change during truncate
1120                  */
1121                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1122                                          &oh->ot_jcb);
1123
1124                 LASSERT(oti->oti_txns == 1);
1125                 oti->oti_txns--;
1126                 rc = dt_txn_hook_stop(env, th);
1127                 if (rc != 0)
1128                         CERROR("Failure in transaction hook: %d\n", rc);
1129
1130                 /* hook functions might modify th_sync */
1131                 hdl->h_sync = th->th_sync;
1132
1133                 oh->ot_handle = NULL;
1134                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1135                                   rc = ldiskfs_journal_stop(hdl));
1136                 if (rc != 0)
1137                         CERROR("Failure to stop transaction: %d\n", rc);
1138         } else {
1139                 thandle_put(&oh->ot_super);
1140         }
1141
1142         /* inform the quota slave device that the transaction is stopping */
1143         qsd_op_end(env, qsd, qtrans);
1144
1145         /* as we want IO to journal and data IO be concurrent, we don't block
1146          * awaiting data IO completion in osd_do_bio(), instead we wait here
1147          * once transaction is submitted to the journal. all reqular requests
1148          * don't do direct IO (except read/write), thus this wait_event becomes
1149          * no-op for them.
1150          *
1151          * IMPORTANT: we have to wait till any IO submited by the thread is
1152          * completed otherwise iobuf may be corrupted by different request
1153          */
1154         wait_event(iobuf->dr_wait,
1155                        atomic_read(&iobuf->dr_numreqs) == 0);
1156         osd_fini_iobuf(oti->oti_dev, iobuf);
1157         if (!rc)
1158                 rc = iobuf->dr_error;
1159
1160         RETURN(rc);
1161 }
1162
1163 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1164 {
1165         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1166                                                ot_super);
1167
1168         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1169         LASSERT(&dcb->dcb_func != NULL);
1170         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1171
1172         return 0;
1173 }
1174
1175 /*
1176  * Called just before object is freed. Releases all resources except for
1177  * object itself (that is released by osd_object_free()).
1178  *
1179  * Concurrency: no concurrent access is possible that late in object
1180  * life-cycle.
1181  */
1182 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1183 {
1184         struct osd_object *obj   = osd_obj(l);
1185         struct inode      *inode = obj->oo_inode;
1186
1187         LINVRNT(osd_invariant(obj));
1188
1189         /*
1190          * If object is unlinked remove fid->ino mapping from object index.
1191          */
1192
1193         osd_index_fini(obj);
1194         if (inode != NULL) {
1195                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1196                 qid_t                    uid = i_uid_read(inode);
1197                 qid_t                    gid = i_gid_read(inode);
1198
1199                 iput(inode);
1200                 obj->oo_inode = NULL;
1201
1202                 if (qsd != NULL) {
1203                         struct osd_thread_info  *info = osd_oti_get(env);
1204                         struct lquota_id_info   *qi = &info->oti_qi;
1205
1206                         /* Release granted quota to master if necessary */
1207                         qi->lqi_id.qid_uid = uid;
1208                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1209
1210                         qi->lqi_id.qid_uid = gid;
1211                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1212                 }
1213         }
1214 }
1215
1216 /*
1217  * Concurrency: ->loo_object_release() is called under site spin-lock.
1218  */
1219 static void osd_object_release(const struct lu_env *env,
1220                                struct lu_object *l)
1221 {
1222 }
1223
1224 /*
1225  * Concurrency: shouldn't matter.
1226  */
1227 static int osd_object_print(const struct lu_env *env, void *cookie,
1228                             lu_printer_t p, const struct lu_object *l)
1229 {
1230         struct osd_object *o = osd_obj(l);
1231         struct iam_descr  *d;
1232
1233         if (o->oo_dir != NULL)
1234                 d = o->oo_dir->od_container.ic_descr;
1235         else
1236                 d = NULL;
1237         return (*p)(env, cookie,
1238                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1239                     o, o->oo_inode,
1240                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1241                     o->oo_inode ? o->oo_inode->i_generation : 0,
1242                     d ? d->id_ops->id_name : "plain");
1243 }
1244
1245 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1246
1247 /*
1248  * Concurrency: shouldn't matter.
1249  */
1250 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1251                struct obd_statfs *sfs)
1252 {
1253         struct osd_device  *osd = osd_dt_dev(d);
1254         struct super_block *sb = osd_sb(osd);
1255         struct kstatfs     *ksfs;
1256         int result = 0;
1257
1258         if (unlikely(osd->od_mnt == NULL))
1259                 return -EINPROGRESS;
1260
1261         /* osd_lproc.c call this without env, allocate ksfs for that case */
1262         if (unlikely(env == NULL)) {
1263                 OBD_ALLOC_PTR(ksfs);
1264                 if (ksfs == NULL)
1265                         return -ENOMEM;
1266         } else {
1267                 ksfs = &osd_oti_get(env)->oti_ksfs;
1268         }
1269
1270         spin_lock(&osd->od_osfs_lock);
1271         result = sb->s_op->statfs(sb->s_root, ksfs);
1272         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1273                 statfs_pack(sfs, ksfs);
1274                 if (sb->s_flags & MS_RDONLY)
1275                         sfs->os_state = OS_STATE_READONLY;
1276         }
1277
1278         spin_unlock(&osd->od_osfs_lock);
1279
1280         if (unlikely(env == NULL))
1281                 OBD_FREE_PTR(ksfs);
1282
1283         /* Reserve a small amount of space for local objects like last_rcvd,
1284          * llog, quota files, ... */
1285         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1286                 sfs->os_bavail = 0;
1287         } else {
1288                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1289                 /** Take out metadata overhead for indirect blocks */
1290                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1291         }
1292
1293         return result;
1294 }
1295
1296 /**
1297  * Estimate space needed for file creations. We assume the largest filename
1298  * which is 2^64 - 1, hence a filename of 20 chars.
1299  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1300  */
1301 #ifdef __LDISKFS_DIR_REC_LEN
1302 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1303 #else
1304 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1305 #endif
1306
1307 /*
1308  * Concurrency: doesn't access mutable data.
1309  */
1310 static void osd_conf_get(const struct lu_env *env,
1311                          const struct dt_device *dev,
1312                          struct dt_device_param *param)
1313 {
1314         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1315         int                ea_overhead;
1316
1317         /*
1318          * XXX should be taken from not-yet-existing fs abstraction layer.
1319          */
1320         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1321         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1322         param->ddp_block_shift  = sb->s_blocksize_bits;
1323         param->ddp_mount_type     = LDD_MT_LDISKFS;
1324         param->ddp_maxbytes       = sb->s_maxbytes;
1325         /* Overhead estimate should be fairly accurate, so we really take a tiny
1326          * error margin which also avoids fragmenting the filesystem too much */
1327         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1328         /* inode are statically allocated, so per-inode space consumption
1329          * is the space consumed by the directory entry */
1330         param->ddp_inodespace     = PER_OBJ_USAGE;
1331         /* per-fragment overhead to be used by the client code */
1332         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1333         param->ddp_mntopts      = 0;
1334         if (test_opt(sb, XATTR_USER))
1335                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1336         if (test_opt(sb, POSIX_ACL))
1337                 param->ddp_mntopts |= MNTOPT_ACL;
1338
1339         /* LOD might calculate the max stripe count based on max_ea_size,
1340          * so we need take account in the overhead as well,
1341          * xattr_header + magic + xattr_entry_head */
1342         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1343                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1344
1345 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1346         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1347                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1348                                                                 ea_overhead;
1349         else
1350 #endif
1351                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1352 }
1353
1354 /*
1355  * Concurrency: shouldn't matter.
1356  */
1357 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1358 {
1359         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1360         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1361 }
1362
1363 /**
1364  * Start commit for OSD device.
1365  *
1366  * An implementation of dt_commit_async method for OSD device.
1367  * Asychronously starts underlayng fs sync and thereby a transaction
1368  * commit.
1369  *
1370  * \param env environment
1371  * \param d dt device
1372  *
1373  * \see dt_device_operations
1374  */
1375 static int osd_commit_async(const struct lu_env *env,
1376                             struct dt_device *d)
1377 {
1378         struct super_block *s = osd_sb(osd_dt_dev(d));
1379         ENTRY;
1380
1381         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1382         RETURN(s->s_op->sync_fs(s, 0));
1383 }
1384
1385 /*
1386  * Concurrency: shouldn't matter.
1387  */
1388
1389 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1390 {
1391         struct super_block *sb = osd_sb(osd_dt_dev(d));
1392         struct block_device *dev = sb->s_bdev;
1393 #ifdef HAVE_DEV_SET_RDONLY
1394         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1395         int rc = 0;
1396 #else
1397         int rc = -EOPNOTSUPP;
1398 #endif
1399         ENTRY;
1400
1401 #ifdef HAVE_DEV_SET_RDONLY
1402         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1403
1404         if (jdev && (jdev != dev)) {
1405                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1406                        (long)jdev);
1407                 dev_set_rdonly(jdev);
1408         }
1409         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1410         dev_set_rdonly(dev);
1411 #else
1412         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1413                osd_dt_dev(d)->od_svname, (long)dev, rc);
1414 #endif
1415         RETURN(rc);
1416 }
1417
1418 /*
1419  * Concurrency: serialization provided by callers.
1420  */
1421 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1422                               int mode, unsigned long timeout, __u32 alg,
1423                               struct lustre_capa_key *keys)
1424 {
1425         struct osd_device *dev = osd_dt_dev(d);
1426         ENTRY;
1427
1428         dev->od_fl_capa = mode;
1429         dev->od_capa_timeout = timeout;
1430         dev->od_capa_alg = alg;
1431         dev->od_capa_keys = keys;
1432         RETURN(0);
1433 }
1434
1435 /**
1436  * Note: we do not count into QUOTA here.
1437  * If we mount with --data_journal we may need more.
1438  */
1439 const int osd_dto_credits_noquota[DTO_NR] = {
1440         /**
1441          * Insert.
1442          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1443          * SINGLEDATA_TRANS_BLOCKS(8)
1444          * XXX Note: maybe iam need more, since iam have more level than
1445          *           EXT3 htree.
1446          */
1447         [DTO_INDEX_INSERT]  = 16,
1448         /**
1449          * Delete
1450          * just modify a single entry, probably merge few within a block
1451          */
1452         [DTO_INDEX_DELETE]  = 1,
1453         /**
1454          * Used for OI scrub
1455          */
1456         [DTO_INDEX_UPDATE]  = 16,
1457         /**
1458          * 4(inode, inode bits, groups, GDT)
1459          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1460          */
1461         [DTO_OBJECT_CREATE] = 4,
1462         /**
1463          * 4(inode, inode bits, groups, GDT)
1464          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1465          */
1466         [DTO_OBJECT_DELETE] = 4,
1467         /**
1468          * Attr set credits (inode)
1469          */
1470         [DTO_ATTR_SET_BASE] = 1,
1471         /**
1472          * Xattr set. The same as xattr of EXT3.
1473          * DATA_TRANS_BLOCKS(14)
1474          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1475          * are also counted in. Do not know why?
1476          */
1477         [DTO_XATTR_SET]     = 14,
1478         /**
1479          * credits for inode change during write.
1480          */
1481         [DTO_WRITE_BASE]    = 3,
1482         /**
1483          * credits for single block write.
1484          */
1485         [DTO_WRITE_BLOCK]   = 14,
1486         /**
1487          * Attr set credits for chown.
1488          * This is extra credits for setattr, and it is null without quota
1489          */
1490         [DTO_ATTR_SET_CHOWN] = 0
1491 };
1492
1493 static const struct dt_device_operations osd_dt_ops = {
1494         .dt_root_get       = osd_root_get,
1495         .dt_statfs         = osd_statfs,
1496         .dt_trans_create   = osd_trans_create,
1497         .dt_trans_start    = osd_trans_start,
1498         .dt_trans_stop     = osd_trans_stop,
1499         .dt_trans_cb_add   = osd_trans_cb_add,
1500         .dt_conf_get       = osd_conf_get,
1501         .dt_sync           = osd_sync,
1502         .dt_ro             = osd_ro,
1503         .dt_commit_async   = osd_commit_async,
1504         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1505 };
1506
1507 static void osd_object_read_lock(const struct lu_env *env,
1508                                  struct dt_object *dt, unsigned role)
1509 {
1510         struct osd_object *obj = osd_dt_obj(dt);
1511         struct osd_thread_info *oti = osd_oti_get(env);
1512
1513         LINVRNT(osd_invariant(obj));
1514
1515         LASSERT(obj->oo_owner != env);
1516         down_read_nested(&obj->oo_sem, role);
1517
1518         LASSERT(obj->oo_owner == NULL);
1519         oti->oti_r_locks++;
1520 }
1521
1522 static void osd_object_write_lock(const struct lu_env *env,
1523                                   struct dt_object *dt, unsigned role)
1524 {
1525         struct osd_object *obj = osd_dt_obj(dt);
1526         struct osd_thread_info *oti = osd_oti_get(env);
1527
1528         LINVRNT(osd_invariant(obj));
1529
1530         LASSERT(obj->oo_owner != env);
1531         down_write_nested(&obj->oo_sem, role);
1532
1533         LASSERT(obj->oo_owner == NULL);
1534         obj->oo_owner = env;
1535         oti->oti_w_locks++;
1536 }
1537
1538 static void osd_object_read_unlock(const struct lu_env *env,
1539                                    struct dt_object *dt)
1540 {
1541         struct osd_object *obj = osd_dt_obj(dt);
1542         struct osd_thread_info *oti = osd_oti_get(env);
1543
1544         LINVRNT(osd_invariant(obj));
1545
1546         LASSERT(oti->oti_r_locks > 0);
1547         oti->oti_r_locks--;
1548         up_read(&obj->oo_sem);
1549 }
1550
1551 static void osd_object_write_unlock(const struct lu_env *env,
1552                                     struct dt_object *dt)
1553 {
1554         struct osd_object *obj = osd_dt_obj(dt);
1555         struct osd_thread_info *oti = osd_oti_get(env);
1556
1557         LINVRNT(osd_invariant(obj));
1558
1559         LASSERT(obj->oo_owner == env);
1560         LASSERT(oti->oti_w_locks > 0);
1561         oti->oti_w_locks--;
1562         obj->oo_owner = NULL;
1563         up_write(&obj->oo_sem);
1564 }
1565
1566 static int osd_object_write_locked(const struct lu_env *env,
1567                                    struct dt_object *dt)
1568 {
1569         struct osd_object *obj = osd_dt_obj(dt);
1570
1571         LINVRNT(osd_invariant(obj));
1572
1573         return obj->oo_owner == env;
1574 }
1575
1576 static int capa_is_sane(const struct lu_env *env,
1577                         struct osd_device *dev,
1578                         struct lustre_capa *capa,
1579                         struct lustre_capa_key *keys)
1580 {
1581         struct osd_thread_info *oti = osd_oti_get(env);
1582         struct lustre_capa *tcapa = &oti->oti_capa;
1583         struct obd_capa *oc;
1584         int i, rc = 0;
1585         ENTRY;
1586
1587         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1588         if (oc) {
1589                 if (capa_is_expired(oc)) {
1590                         DEBUG_CAPA(D_ERROR, capa, "expired");
1591                         rc = -ESTALE;
1592                 }
1593                 capa_put(oc);
1594                 RETURN(rc);
1595         }
1596
1597         if (capa_is_expired_sec(capa)) {
1598                 DEBUG_CAPA(D_ERROR, capa, "expired");
1599                 RETURN(-ESTALE);
1600         }
1601
1602         spin_lock(&capa_lock);
1603         for (i = 0; i < 2; i++) {
1604                 if (keys[i].lk_keyid == capa->lc_keyid) {
1605                         oti->oti_capa_key = keys[i];
1606                         break;
1607                 }
1608         }
1609         spin_unlock(&capa_lock);
1610
1611         if (i == 2) {
1612                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1613                 RETURN(-ESTALE);
1614         }
1615
1616         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1617         if (rc)
1618                 RETURN(rc);
1619
1620         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1621                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1622                 RETURN(-EACCES);
1623         }
1624
1625         oc = capa_add(dev->od_capa_hash, capa);
1626         capa_put(oc);
1627
1628         RETURN(0);
1629 }
1630
1631 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1632                     struct lustre_capa *capa, __u64 opc)
1633 {
1634         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1635         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1636         struct lu_capainfo *lci;
1637         int rc;
1638
1639         if (!osd->od_fl_capa)
1640                 return 0;
1641
1642         if (capa == BYPASS_CAPA)
1643                 return 0;
1644
1645         lci = lu_capainfo_get(env);
1646         if (unlikely(lci == NULL))
1647                 return 0;
1648
1649         if (lci->lci_auth == LC_ID_NONE)
1650                 return 0;
1651
1652         if (capa == NULL) {
1653                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1654                        osd_name(osd), PFID(fid), -EACCES);
1655                 return -EACCES;
1656         }
1657
1658         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1659                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1660                            PFID(fid));
1661                 return -EACCES;
1662         }
1663
1664         if (!capa_opc_supported(capa, opc)) {
1665                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1666                 return -EACCES;
1667         }
1668
1669         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1670         if (rc != 0) {
1671                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1672                 return -EACCES;
1673         }
1674
1675         return 0;
1676 }
1677
1678 static struct timespec *osd_inode_time(const struct lu_env *env,
1679                                        struct inode *inode, __u64 seconds)
1680 {
1681         struct osd_thread_info  *oti = osd_oti_get(env);
1682         struct timespec         *t   = &oti->oti_time;
1683
1684         t->tv_sec = seconds;
1685         t->tv_nsec = 0;
1686         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1687         return t;
1688 }
1689
1690
1691 static void osd_inode_getattr(const struct lu_env *env,
1692                               struct inode *inode, struct lu_attr *attr)
1693 {
1694         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1695                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1696                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1697                            LA_TYPE;
1698
1699         attr->la_atime   = LTIME_S(inode->i_atime);
1700         attr->la_mtime   = LTIME_S(inode->i_mtime);
1701         attr->la_ctime   = LTIME_S(inode->i_ctime);
1702         attr->la_mode    = inode->i_mode;
1703         attr->la_size    = i_size_read(inode);
1704         attr->la_blocks  = inode->i_blocks;
1705         attr->la_uid     = i_uid_read(inode);
1706         attr->la_gid     = i_gid_read(inode);
1707         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1708         attr->la_nlink   = inode->i_nlink;
1709         attr->la_rdev    = inode->i_rdev;
1710         attr->la_blksize = 1 << inode->i_blkbits;
1711         attr->la_blkbits = inode->i_blkbits;
1712 }
1713
1714 static int osd_attr_get(const struct lu_env *env,
1715                         struct dt_object *dt,
1716                         struct lu_attr *attr,
1717                         struct lustre_capa *capa)
1718 {
1719         struct osd_object *obj = osd_dt_obj(dt);
1720
1721         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1722         LINVRNT(osd_invariant(obj));
1723
1724         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1725                 return -EACCES;
1726
1727         spin_lock(&obj->oo_guard);
1728         osd_inode_getattr(env, obj->oo_inode, attr);
1729         spin_unlock(&obj->oo_guard);
1730         return 0;
1731 }
1732
1733 static int osd_declare_attr_set(const struct lu_env *env,
1734                                 struct dt_object *dt,
1735                                 const struct lu_attr *attr,
1736                                 struct thandle *handle)
1737 {
1738         struct osd_thandle     *oh;
1739         struct osd_object      *obj;
1740         struct osd_thread_info *info = osd_oti_get(env);
1741         struct lquota_id_info  *qi = &info->oti_qi;
1742         qid_t                   uid;
1743         qid_t                   gid;
1744         long long               bspace;
1745         int                     rc = 0;
1746         bool                    enforce;
1747         ENTRY;
1748
1749         LASSERT(dt != NULL);
1750         LASSERT(handle != NULL);
1751
1752         obj = osd_dt_obj(dt);
1753         LASSERT(osd_invariant(obj));
1754
1755         oh = container_of0(handle, struct osd_thandle, ot_super);
1756         LASSERT(oh->ot_handle == NULL);
1757
1758         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1759                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1760
1761         if (attr == NULL || obj->oo_inode == NULL)
1762                 RETURN(rc);
1763
1764         bspace   = obj->oo_inode->i_blocks;
1765         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1766         bspace   = toqb(bspace);
1767
1768         /* Changing ownership is always preformed by super user, it should not
1769          * fail with EDQUOT.
1770          *
1771          * We still need to call the osd_declare_qid() to calculate the journal
1772          * credits for updating quota accounting files and to trigger quota
1773          * space adjustment once the operation is completed.*/
1774         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1775                 /* USERQUOTA */
1776                 uid = i_uid_read(obj->oo_inode);
1777                 qi->lqi_type = USRQUOTA;
1778                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1779                 /* inode accounting */
1780                 qi->lqi_is_blk = false;
1781
1782                 /* one more inode for the new uid ... */
1783                 qi->lqi_id.qid_uid = attr->la_uid;
1784                 qi->lqi_space      = 1;
1785                 /* Reserve credits for the new uid */
1786                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1787                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1788                         rc = 0;
1789                 if (rc)
1790                         RETURN(rc);
1791
1792                 /* and one less inode for the current uid */
1793                 qi->lqi_id.qid_uid = uid;
1794                 qi->lqi_space      = -1;
1795                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1796                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1797                         rc = 0;
1798                 if (rc)
1799                         RETURN(rc);
1800
1801                 /* block accounting */
1802                 qi->lqi_is_blk = true;
1803
1804                 /* more blocks for the new uid ... */
1805                 qi->lqi_id.qid_uid = attr->la_uid;
1806                 qi->lqi_space      = bspace;
1807                 /*
1808                  * Credits for the new uid has been reserved, re-use "obj"
1809                  * to save credit reservation.
1810                  */
1811                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1812                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1813                         rc = 0;
1814                 if (rc)
1815                         RETURN(rc);
1816
1817                 /* and finally less blocks for the current uid */
1818                 qi->lqi_id.qid_uid = uid;
1819                 qi->lqi_space      = -bspace;
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                 /* GROUP QUOTA */
1827                 gid = i_gid_read(obj->oo_inode);
1828                 qi->lqi_type = GRPQUOTA;
1829                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1830
1831                 /* inode accounting */
1832                 qi->lqi_is_blk = false;
1833
1834                 /* one more inode for the new gid ... */
1835                 qi->lqi_id.qid_gid = attr->la_gid;
1836                 qi->lqi_space      = 1;
1837                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1838                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1839                         rc = 0;
1840                 if (rc)
1841                         RETURN(rc);
1842
1843                 /* and one less inode for the current gid */
1844                 qi->lqi_id.qid_gid = gid;
1845                 qi->lqi_space      = -1;
1846                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1847                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1848                         rc = 0;
1849                 if (rc)
1850                         RETURN(rc);
1851
1852                 /* block accounting */
1853                 qi->lqi_is_blk = true;
1854
1855                 /* more blocks for the new gid ... */
1856                 qi->lqi_id.qid_gid = attr->la_gid;
1857                 qi->lqi_space      = bspace;
1858                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1859                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1860                         rc = 0;
1861                 if (rc)
1862                         RETURN(rc);
1863
1864                 /* and finally less blocks for the current gid */
1865                 qi->lqi_id.qid_gid = 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
1874         RETURN(rc);
1875 }
1876
1877 static int osd_inode_setattr(const struct lu_env *env,
1878                              struct inode *inode, const struct lu_attr *attr)
1879 {
1880         __u64 bits = attr->la_valid;
1881
1882         /* Only allow set size for regular file */
1883         if (!S_ISREG(inode->i_mode))
1884                 bits &= ~(LA_SIZE | LA_BLOCKS);
1885
1886         if (bits == 0)
1887                 return 0;
1888
1889         if (bits & LA_ATIME)
1890                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1891         if (bits & LA_CTIME)
1892                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1893         if (bits & LA_MTIME)
1894                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1895         if (bits & LA_SIZE) {
1896                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1897                 i_size_write(inode, attr->la_size);
1898         }
1899
1900 #if 0
1901         /* OSD should not change "i_blocks" which is used by quota.
1902          * "i_blocks" should be changed by ldiskfs only. */
1903         if (bits & LA_BLOCKS)
1904                 inode->i_blocks = attr->la_blocks;
1905 #endif
1906         if (bits & LA_MODE)
1907                 inode->i_mode = (inode->i_mode & S_IFMT) |
1908                                 (attr->la_mode & ~S_IFMT);
1909         if (bits & LA_UID)
1910                 i_uid_write(inode, attr->la_uid);
1911         if (bits & LA_GID)
1912                 i_gid_write(inode, attr->la_gid);
1913         if (bits & LA_NLINK)
1914                 set_nlink(inode, attr->la_nlink);
1915         if (bits & LA_RDEV)
1916                 inode->i_rdev = attr->la_rdev;
1917
1918         if (bits & LA_FLAGS) {
1919                 /* always keep S_NOCMTIME */
1920                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1921                                  S_NOCMTIME;
1922         }
1923         return 0;
1924 }
1925
1926 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1927 {
1928         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1929             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1930                 struct iattr    iattr;
1931                 int             rc;
1932
1933                 ll_vfs_dq_init(inode);
1934                 iattr.ia_valid = 0;
1935                 if (attr->la_valid & LA_UID)
1936                         iattr.ia_valid |= ATTR_UID;
1937                 if (attr->la_valid & LA_GID)
1938                         iattr.ia_valid |= ATTR_GID;
1939                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1940                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1941
1942                 rc = ll_vfs_dq_transfer(inode, &iattr);
1943                 if (rc) {
1944                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1945                                "enforcement enabled on the ldiskfs "
1946                                "filesystem?\n", inode->i_sb->s_id, rc);
1947                         return rc;
1948                 }
1949         }
1950         return 0;
1951 }
1952
1953 static int osd_attr_set(const struct lu_env *env,
1954                         struct dt_object *dt,
1955                         const struct lu_attr *attr,
1956                         struct thandle *handle,
1957                         struct lustre_capa *capa)
1958 {
1959         struct osd_object *obj = osd_dt_obj(dt);
1960         struct inode      *inode;
1961         int rc;
1962
1963         LASSERT(handle != NULL);
1964         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1965         LASSERT(osd_invariant(obj));
1966
1967         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1968                 return -EACCES;
1969
1970         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1971
1972         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1973                 struct osd_thread_info  *oti  = osd_oti_get(env);
1974                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1975                 struct lu_fid           *fid1 = &oti->oti_fid;
1976                 struct osd_inode_id     *id   = &oti->oti_id;
1977                 struct iam_path_descr   *ipd;
1978                 struct iam_container    *bag;
1979                 struct osd_thandle      *oh;
1980                 int                      rc;
1981
1982                 fid_cpu_to_be(fid1, fid0);
1983                 memset(id, 1, sizeof(*id));
1984                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1985                                   fid0)->oi_dir.od_container;
1986                 ipd = osd_idx_ipd_get(env, bag);
1987                 if (unlikely(ipd == NULL))
1988                         RETURN(-ENOMEM);
1989
1990                 oh = container_of0(handle, struct osd_thandle, ot_super);
1991                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1992                                 (const struct iam_rec *)id, ipd);
1993                 osd_ipd_put(env, bag, ipd);
1994                 return(rc > 0 ? 0 : rc);
1995         }
1996
1997         inode = obj->oo_inode;
1998
1999         rc = osd_quota_transfer(inode, attr);
2000         if (rc)
2001                 return rc;
2002
2003         spin_lock(&obj->oo_guard);
2004         rc = osd_inode_setattr(env, inode, attr);
2005         spin_unlock(&obj->oo_guard);
2006
2007         if (!rc)
2008                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2009         return rc;
2010 }
2011
2012 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2013                                     struct osd_object *obj,
2014                                     const char *name, const int namelen)
2015 {
2016         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2017 }
2018
2019 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2020                       umode_t mode, struct dt_allocation_hint *hint,
2021                       struct thandle *th)
2022 {
2023         int result;
2024         struct osd_device  *osd = osd_obj2dev(obj);
2025         struct osd_thandle *oth;
2026         struct dt_object   *parent = NULL;
2027         struct inode       *inode;
2028
2029         LINVRNT(osd_invariant(obj));
2030         LASSERT(obj->oo_inode == NULL);
2031         LASSERT(obj->oo_hl_head == NULL);
2032
2033         if (S_ISDIR(mode) && ldiskfs_pdo) {
2034                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2035                 if (obj->oo_hl_head == NULL)
2036                         return -ENOMEM;
2037         }
2038
2039         oth = container_of(th, struct osd_thandle, ot_super);
2040         LASSERT(oth->ot_handle->h_transaction != NULL);
2041
2042         if (hint && hint->dah_parent)
2043                 parent = hint->dah_parent;
2044
2045         inode = ldiskfs_create_inode(oth->ot_handle,
2046                                      parent ? osd_dt_obj(parent)->oo_inode :
2047                                               osd_sb(osd)->s_root->d_inode,
2048                                      mode);
2049         if (!IS_ERR(inode)) {
2050                 /* Do not update file c/mtime in ldiskfs.
2051                  * NB: don't need any lock because no contention at this
2052                  * early stage */
2053                 inode->i_flags |= S_NOCMTIME;
2054
2055                 /* For new created object, it must be consistent,
2056                  * and it is unnecessary to scrub against it. */
2057                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2058                 obj->oo_inode = inode;
2059                 result = 0;
2060         } else {
2061                 if (obj->oo_hl_head != NULL) {
2062                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2063                         obj->oo_hl_head = NULL;
2064                 }
2065                 result = PTR_ERR(inode);
2066         }
2067         LINVRNT(osd_invariant(obj));
2068         return result;
2069 }
2070
2071 enum {
2072         OSD_NAME_LEN = 255
2073 };
2074
2075 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2076                      struct lu_attr *attr,
2077                      struct dt_allocation_hint *hint,
2078                      struct dt_object_format *dof,
2079                      struct thandle *th)
2080 {
2081         int result;
2082         struct osd_thandle *oth;
2083         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2084
2085         LASSERT(S_ISDIR(attr->la_mode));
2086
2087         oth = container_of(th, struct osd_thandle, ot_super);
2088         LASSERT(oth->ot_handle->h_transaction != NULL);
2089         result = osd_mkfile(info, obj, mode, hint, th);
2090
2091         return result;
2092 }
2093
2094 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2095                         struct lu_attr *attr,
2096                         struct dt_allocation_hint *hint,
2097                         struct dt_object_format *dof,
2098                         struct thandle *th)
2099 {
2100         int result;
2101         struct osd_thandle *oth;
2102         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2103
2104         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2105
2106         LASSERT(S_ISREG(attr->la_mode));
2107
2108         oth = container_of(th, struct osd_thandle, ot_super);
2109         LASSERT(oth->ot_handle->h_transaction != NULL);
2110
2111         result = osd_mkfile(info, obj, mode, hint, th);
2112         if (result == 0) {
2113                 LASSERT(obj->oo_inode != NULL);
2114                 if (feat->dif_flags & DT_IND_VARKEY)
2115                         result = iam_lvar_create(obj->oo_inode,
2116                                                  feat->dif_keysize_max,
2117                                                  feat->dif_ptrsize,
2118                                                  feat->dif_recsize_max,
2119                                                  oth->ot_handle);
2120                 else
2121                         result = iam_lfix_create(obj->oo_inode,
2122                                                  feat->dif_keysize_max,
2123                                                  feat->dif_ptrsize,
2124                                                  feat->dif_recsize_max,
2125                                                  oth->ot_handle);
2126
2127         }
2128         return result;
2129 }
2130
2131 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2132                      struct lu_attr *attr,
2133                      struct dt_allocation_hint *hint,
2134                      struct dt_object_format *dof,
2135                      struct thandle *th)
2136 {
2137         LASSERT(S_ISREG(attr->la_mode));
2138         return osd_mkfile(info, obj, (attr->la_mode &
2139                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2140 }
2141
2142 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2143                      struct lu_attr *attr,
2144                      struct dt_allocation_hint *hint,
2145                      struct dt_object_format *dof,
2146                      struct thandle *th)
2147 {
2148         LASSERT(S_ISLNK(attr->la_mode));
2149         return osd_mkfile(info, obj, (attr->la_mode &
2150                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2151 }
2152
2153 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2154                      struct lu_attr *attr,
2155                      struct dt_allocation_hint *hint,
2156                      struct dt_object_format *dof,
2157                      struct thandle *th)
2158 {
2159         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2160         int result;
2161
2162         LINVRNT(osd_invariant(obj));
2163         LASSERT(obj->oo_inode == NULL);
2164         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2165                 S_ISFIFO(mode) || S_ISSOCK(mode));
2166
2167         result = osd_mkfile(info, obj, mode, hint, th);
2168         if (result == 0) {
2169                 LASSERT(obj->oo_inode != NULL);
2170                 /*
2171                  * This inode should be marked dirty for i_rdev.  Currently
2172                  * that is done in the osd_attr_init().
2173                  */
2174                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2175                                    attr->la_rdev);
2176         }
2177         LINVRNT(osd_invariant(obj));
2178         return result;
2179 }
2180
2181 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2182                               struct lu_attr *,
2183                               struct dt_allocation_hint *hint,
2184                               struct dt_object_format *dof,
2185                               struct thandle *);
2186
2187 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2188 {
2189         osd_obj_type_f result;
2190
2191         switch (type) {
2192         case DFT_DIR:
2193                 result = osd_mkdir;
2194                 break;
2195         case DFT_REGULAR:
2196                 result = osd_mkreg;
2197                 break;
2198         case DFT_SYM:
2199                 result = osd_mksym;
2200                 break;
2201         case DFT_NODE:
2202                 result = osd_mknod;
2203                 break;
2204         case DFT_INDEX:
2205                 result = osd_mk_index;
2206                 break;
2207
2208         default:
2209                 LBUG();
2210                 break;
2211         }
2212         return result;
2213 }
2214
2215
2216 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2217                         struct dt_object *parent, struct dt_object *child,
2218                         umode_t child_mode)
2219 {
2220         LASSERT(ah);
2221
2222         ah->dah_parent = parent;
2223         ah->dah_mode = child_mode;
2224 }
2225
2226 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2227                           struct lu_attr *attr, struct dt_object_format *dof)
2228 {
2229         struct inode   *inode = obj->oo_inode;
2230         __u64           valid = attr->la_valid;
2231         int             result;
2232
2233         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2234
2235         if (dof->dof_type != DFT_NODE)
2236                 attr->la_valid &= ~LA_RDEV;
2237         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2238                 attr->la_valid &= ~LA_ATIME;
2239         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2240                 attr->la_valid &= ~LA_CTIME;
2241         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2242                 attr->la_valid &= ~LA_MTIME;
2243
2244         result = osd_quota_transfer(inode, attr);
2245         if (result)
2246                 return;
2247
2248         if (attr->la_valid != 0) {
2249                 result = osd_inode_setattr(info->oti_env, inode, attr);
2250                 /*
2251                  * The osd_inode_setattr() should always succeed here.  The
2252                  * only error that could be returned is EDQUOT when we are
2253                  * trying to change the UID or GID of the inode. However, this
2254                  * should not happen since quota enforcement is no longer
2255                  * enabled on ldiskfs (lquota takes care of it).
2256                  */
2257                 LASSERTF(result == 0, "%d\n", result);
2258                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2259         }
2260
2261         attr->la_valid = valid;
2262 }
2263
2264 /**
2265  * Helper function for osd_object_create()
2266  *
2267  * \retval 0, on success
2268  */
2269 static int __osd_object_create(struct osd_thread_info *info,
2270                                struct osd_object *obj, struct lu_attr *attr,
2271                                struct dt_allocation_hint *hint,
2272                                struct dt_object_format *dof,
2273                                struct thandle *th)
2274 {
2275         int     result;
2276         __u32   umask;
2277
2278         /* we drop umask so that permissions we pass are not affected */
2279         umask = current->fs->umask;
2280         current->fs->umask = 0;
2281
2282         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2283                                                   th);
2284         if (result == 0) {
2285                 osd_attr_init(info, obj, attr, dof);
2286                 osd_object_init0(obj);
2287                 /* bz 24037 */
2288                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2289                         unlock_new_inode(obj->oo_inode);
2290         }
2291
2292         /* restore previous umask value */
2293         current->fs->umask = umask;
2294
2295         return result;
2296 }
2297
2298 /**
2299  * Helper function for osd_object_create()
2300  *
2301  * \retval 0, on success
2302  */
2303 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2304                            const struct lu_fid *fid, struct thandle *th)
2305 {
2306         struct osd_thread_info *info = osd_oti_get(env);
2307         struct osd_inode_id    *id   = &info->oti_id;
2308         struct osd_device      *osd  = osd_obj2dev(obj);
2309         struct osd_thandle     *oh;
2310
2311         LASSERT(obj->oo_inode != NULL);
2312
2313         oh = container_of0(th, struct osd_thandle, ot_super);
2314         LASSERT(oh->ot_handle);
2315
2316         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2317         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2318 }
2319
2320 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2321                    obd_seq seq, struct lu_seq_range *range)
2322 {
2323         struct seq_server_site  *ss = osd_seq_site(osd);
2324
2325         if (fid_seq_is_idif(seq)) {
2326                 fld_range_set_ost(range);
2327                 range->lsr_index = idif_ost_idx(seq);
2328                 return 0;
2329         }
2330
2331         if (!fid_seq_in_fldb(seq)) {
2332                 fld_range_set_mdt(range);
2333                 if (ss != NULL)
2334                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2335                          * at all */
2336                         range->lsr_index = ss->ss_node_id;
2337                 return 0;
2338         }
2339
2340         LASSERT(ss != NULL);
2341         fld_range_set_any(range);
2342         /* OSD will only do local fld lookup */
2343         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2344 }
2345
2346 /*
2347  * Concurrency: no external locking is necessary.
2348  */
2349 static int osd_declare_object_create(const struct lu_env *env,
2350                                      struct dt_object *dt,
2351                                      struct lu_attr *attr,
2352                                      struct dt_allocation_hint *hint,
2353                                      struct dt_object_format *dof,
2354                                      struct thandle *handle)
2355 {
2356         struct osd_thandle      *oh;
2357         int                      rc;
2358         ENTRY;
2359
2360         LASSERT(handle != NULL);
2361
2362         oh = container_of0(handle, struct osd_thandle, ot_super);
2363         LASSERT(oh->ot_handle == NULL);
2364
2365         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2366                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2367         /* Reuse idle OI block may cause additional one OI block
2368          * to be changed. */
2369         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2370                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2371
2372         /* If this is directory, then we expect . and .. to be inserted as
2373          * well. The one directory block always needs to be created for the
2374          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2375          * block), there is no danger of needing a tree for the first block.
2376          */
2377         if (attr && S_ISDIR(attr->la_mode)) {
2378                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2379                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2380                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2381         }
2382
2383         if (!attr)
2384                 RETURN(0);
2385
2386         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2387                                    osd_dt_obj(dt), false, NULL, false);
2388         if (rc != 0)
2389                 RETURN(rc);
2390
2391         RETURN(rc);
2392 }
2393
2394 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2395                              struct lu_attr *attr,
2396                              struct dt_allocation_hint *hint,
2397                              struct dt_object_format *dof,
2398                              struct thandle *th)
2399 {
2400         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2401         struct osd_object      *obj    = osd_dt_obj(dt);
2402         struct osd_thread_info *info   = osd_oti_get(env);
2403         int result;
2404
2405         ENTRY;
2406
2407         LINVRNT(osd_invariant(obj));
2408         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2409         LASSERT(osd_write_locked(env, obj));
2410         LASSERT(th != NULL);
2411
2412         if (unlikely(fid_is_acct(fid)))
2413                 /* Quota files can't be created from the kernel any more,
2414                  * 'tune2fs -O quota' will take care of creating them */
2415                 RETURN(-EPERM);
2416
2417         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2418         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2419
2420         result = __osd_object_create(info, obj, attr, hint, dof, th);
2421         if (result == 0)
2422                 result = __osd_oi_insert(env, obj, fid, th);
2423
2424         LASSERT(ergo(result == 0,
2425                      dt_object_exists(dt) && !dt_object_remote(dt)));
2426
2427         LASSERT(osd_invariant(obj));
2428         RETURN(result);
2429 }
2430
2431 /**
2432  * Called to destroy on-disk representation of the object
2433  *
2434  * Concurrency: must be locked
2435  */
2436 static int osd_declare_object_destroy(const struct lu_env *env,
2437                                       struct dt_object *dt,
2438                                       struct thandle *th)
2439 {
2440         struct osd_object  *obj = osd_dt_obj(dt);
2441         struct inode       *inode = obj->oo_inode;
2442         struct osd_thandle *oh;
2443         int                 rc;
2444         ENTRY;
2445
2446         oh = container_of0(th, struct osd_thandle, ot_super);
2447         LASSERT(oh->ot_handle == NULL);
2448         LASSERT(inode);
2449
2450         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2451                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2452         /* Recycle idle OI leaf may cause additional three OI blocks
2453          * to be changed. */
2454         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2455                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2456         /* one less inode */
2457         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2458                                    -1, oh, obj, false, NULL, false);
2459         if (rc)
2460                 RETURN(rc);
2461         /* data to be truncated */
2462         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2463                                    0, oh, obj, true, NULL, false);
2464         RETURN(rc);
2465 }
2466
2467 static int osd_object_destroy(const struct lu_env *env,
2468                               struct dt_object *dt,
2469                               struct thandle *th)
2470 {
2471         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2472         struct osd_object      *obj = osd_dt_obj(dt);
2473         struct inode           *inode = obj->oo_inode;
2474         struct osd_device      *osd = osd_obj2dev(obj);
2475         struct osd_thandle     *oh;
2476         int                     result;
2477         ENTRY;
2478
2479         oh = container_of0(th, struct osd_thandle, ot_super);
2480         LASSERT(oh->ot_handle);
2481         LASSERT(inode);
2482         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2483
2484         if (unlikely(fid_is_acct(fid)))
2485                 RETURN(-EPERM);
2486
2487         if (S_ISDIR(inode->i_mode)) {
2488                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2489                         inode->i_nlink == 2);
2490                 /* it will check/delete the inode from remote parent,
2491                  * how to optimize it? unlink performance impaction XXX */
2492                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2493                 if (result != 0 && result != -ENOENT) {
2494                         CERROR("%s: delete inode "DFID": rc = %d\n",
2495                                osd_name(osd), PFID(fid), result);
2496                 }
2497                 spin_lock(&obj->oo_guard);
2498                 clear_nlink(inode);
2499                 spin_unlock(&obj->oo_guard);
2500                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2501         }
2502
2503         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2504
2505         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2506                                OI_CHECK_FLD);
2507
2508         /* XXX: add to ext3 orphan list */
2509         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2510
2511         /* not needed in the cache anymore */
2512         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2513
2514         RETURN(0);
2515 }
2516
2517 /**
2518  * Put the fid into lustre_mdt_attrs, and then place the structure
2519  * inode's ea. This fid should not be altered during the life time
2520  * of the inode.
2521  *
2522  * \retval +ve, on success
2523  * \retval -ve, on error
2524  *
2525  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2526  */
2527 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2528                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2529 {
2530         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2531         int                      rc;
2532         ENTRY;
2533
2534         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2535                 RETURN(0);
2536
2537         lustre_lma_init(lma, fid, compat, incompat);
2538         lustre_lma_swab(lma);
2539
2540         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2541                              XATTR_CREATE);
2542         /* LMA may already exist, but we need to check that all the
2543          * desired compat/incompat flags have been added. */
2544         if (unlikely(rc == -EEXIST)) {
2545                 if (compat == 0 && incompat == 0)
2546                         RETURN(0);
2547
2548                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2549                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2550                                      LMA_OLD_SIZE);
2551                 if (rc <= 0)
2552                         RETURN(-EINVAL);
2553
2554                 lustre_lma_swab(lma);
2555                 if (!(~lma->lma_compat & compat) &&
2556                     !(~lma->lma_incompat & incompat))
2557                         RETURN(0);
2558
2559                 lma->lma_compat |= compat;
2560                 lma->lma_incompat |= incompat;
2561                 lustre_lma_swab(lma);
2562                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2563                                      sizeof(*lma), XATTR_REPLACE);
2564         }
2565
2566         RETURN(rc);
2567 }
2568
2569 /**
2570  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2571  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2572  * To have compatilibility with 1.8 ldiskfs driver we need to have
2573  * magic number at start of fid data.
2574  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2575  * its inmemory API.
2576  */
2577 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2578                                   const struct lu_fid *fid)
2579 {
2580         if (!fid_is_namespace_visible(fid) ||
2581             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2582                 param->edp_magic = 0;
2583                 return;
2584         }
2585
2586         param->edp_magic = LDISKFS_LUFID_MAGIC;
2587         param->edp_len =  sizeof(struct lu_fid) + 1;
2588         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2589 }
2590
2591 /**
2592  * Try to read the fid from inode ea into dt_rec.
2593  *
2594  * \param fid object fid.
2595  *
2596  * \retval 0 on success
2597  */
2598 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2599                           __u32 ino, struct lu_fid *fid,
2600                           struct osd_inode_id *id)
2601 {
2602         struct osd_thread_info *info  = osd_oti_get(env);
2603         struct inode           *inode;
2604         ENTRY;
2605
2606         osd_id_gen(id, ino, OSD_OII_NOGEN);
2607         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2608         if (IS_ERR(inode))
2609                 RETURN(PTR_ERR(inode));
2610
2611         iput(inode);
2612         RETURN(0);
2613 }
2614
2615 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2616                                         struct inode *dir,
2617                                         struct inode *parent_dir,
2618                                         const struct lu_fid *dot_fid,
2619                                         const struct lu_fid *dot_dot_fid,
2620                                         struct osd_thandle *oth)
2621 {
2622         struct ldiskfs_dentry_param *dot_ldp;
2623         struct ldiskfs_dentry_param *dot_dot_ldp;
2624
2625         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2626         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2627
2628         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2629         dot_ldp->edp_magic = 0;
2630         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2631                                         dir, dot_ldp, dot_dot_ldp);
2632 }
2633
2634 /**
2635  * Create an local agent inode for remote entry
2636  */
2637 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2638                                                   struct osd_device *osd,
2639                                                   struct osd_object *pobj,
2640                                                   const struct lu_fid *fid,
2641                                                   __u32 type,
2642                                                   struct thandle *th)
2643 {
2644         struct osd_thread_info  *info = osd_oti_get(env);
2645         struct inode            *local;
2646         struct osd_thandle      *oh;
2647         int                     rc;
2648         ENTRY;
2649
2650         LASSERT(th);
2651         oh = container_of(th, struct osd_thandle, ot_super);
2652         LASSERT(oh->ot_handle->h_transaction != NULL);
2653
2654         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2655         if (IS_ERR(local)) {
2656                 CERROR("%s: create local error %d\n", osd_name(osd),
2657                        (int)PTR_ERR(local));
2658                 RETURN(local);
2659         }
2660
2661         /* Set special LMA flag for local agent inode */
2662         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2663         if (rc != 0) {
2664                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2665                        osd_name(osd), PFID(fid), rc);
2666                 RETURN(ERR_PTR(rc));
2667         }
2668
2669         if (!S_ISDIR(type))
2670                 RETURN(local);
2671
2672         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2673                                          lu_object_fid(&pobj->oo_dt.do_lu),
2674                                          fid, oh);
2675         if (rc != 0) {
2676                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2677                         osd_name(osd), PFID(fid), rc);
2678                 RETURN(ERR_PTR(rc));
2679         }
2680
2681         RETURN(local);
2682 }
2683
2684 /**
2685  * Delete local agent inode for remote entry
2686  */
2687 static int osd_delete_local_agent_inode(const struct lu_env *env,
2688                                         struct osd_device *osd,
2689                                         const struct lu_fid *fid,
2690                                         __u32 ino, struct osd_thandle *oh)
2691 {
2692         struct osd_thread_info  *oti = osd_oti_get(env);
2693         struct osd_inode_id     *id = &oti->oti_id;
2694         struct inode            *inode;
2695         ENTRY;
2696
2697         id->oii_ino = le32_to_cpu(ino);
2698         id->oii_gen = OSD_OII_NOGEN;
2699         inode = osd_iget(oti, osd, id);
2700         if (IS_ERR(inode)) {
2701                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2702                        PFID(fid), id->oii_ino, id->oii_gen);
2703                 RETURN(PTR_ERR(inode));
2704         }
2705
2706         clear_nlink(inode);
2707         mark_inode_dirty(inode);
2708         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2709                 osd_name(osd), PFID(fid), inode->i_ino);
2710         iput(inode);
2711         RETURN(0);
2712 }
2713
2714 /**
2715  * OSD layer object create function for interoperability mode (b11826).
2716  * This is mostly similar to osd_object_create(). Only difference being, fid is
2717  * inserted into inode ea here.
2718  *
2719  * \retval   0, on success
2720  * \retval -ve, on error
2721  */
2722 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2723                                 struct lu_attr *attr,
2724                                 struct dt_allocation_hint *hint,
2725                                 struct dt_object_format *dof,
2726                                 struct thandle *th)
2727 {
2728         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2729         struct osd_object      *obj    = osd_dt_obj(dt);
2730         struct osd_thread_info *info   = osd_oti_get(env);
2731         int                     result;
2732
2733         ENTRY;
2734
2735         LASSERT(osd_invariant(obj));
2736         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2737         LASSERT(osd_write_locked(env, obj));
2738         LASSERT(th != NULL);
2739
2740         if (unlikely(fid_is_acct(fid)))
2741                 /* Quota files can't be created from the kernel any more,
2742                  * 'tune2fs -O quota' will take care of creating them */
2743                 RETURN(-EPERM);
2744
2745         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2746         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2747
2748         result = __osd_object_create(info, obj, attr, hint, dof, th);
2749         if (result == 0)
2750                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2751                                 fid_is_on_ost(info, osd_obj2dev(obj),
2752                                               fid, OI_CHECK_FLD) ?
2753                                 LMAC_FID_ON_OST : 0, 0);
2754
2755         if (result == 0)
2756                 result = __osd_oi_insert(env, obj, fid, th);
2757
2758         LASSERT(ergo(result == 0,
2759                      dt_object_exists(dt) && !dt_object_remote(dt)));
2760         LINVRNT(osd_invariant(obj));
2761         RETURN(result);
2762 }
2763
2764 static int osd_declare_object_ref_add(const struct lu_env *env,
2765                                       struct dt_object *dt,
2766                                       struct thandle *handle)
2767 {
2768         struct osd_thandle       *oh;
2769
2770         /* it's possible that object doesn't exist yet */
2771         LASSERT(handle != NULL);
2772
2773         oh = container_of0(handle, struct osd_thandle, ot_super);
2774         LASSERT(oh->ot_handle == NULL);
2775
2776         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2777                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2778
2779         return 0;
2780 }
2781
2782 /*
2783  * Concurrency: @dt is write locked.
2784  */
2785 static int osd_object_ref_add(const struct lu_env *env,
2786                               struct dt_object *dt, struct thandle *th)
2787 {
2788         struct osd_object  *obj = osd_dt_obj(dt);
2789         struct inode       *inode = obj->oo_inode;
2790         struct osd_thandle *oh;
2791         int                 rc = 0;
2792
2793         LINVRNT(osd_invariant(obj));
2794         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2795         LASSERT(osd_write_locked(env, obj));
2796         LASSERT(th != NULL);
2797
2798         oh = container_of0(th, struct osd_thandle, ot_super);
2799         LASSERT(oh->ot_handle != NULL);
2800
2801         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2802
2803         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2804                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2805         /*
2806          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2807          * (65000) subdirectories by storing "1" in i_nlink if the link count
2808          * would otherwise overflow. Directory tranversal tools understand
2809          * that (st_nlink == 1) indicates that the filesystem dose not track
2810          * hard links count on the directory, and will not abort subdirectory
2811          * scanning early once (st_nlink - 2) subdirs have been found.
2812          *
2813          * This also has to properly handle the case of inodes with nlink == 0
2814          * in case they are being linked into the PENDING directory
2815          */
2816         spin_lock(&obj->oo_guard);
2817         if (unlikely(inode->i_nlink == 0))
2818                 /* inc_nlink from 0 may cause WARN_ON */
2819                 set_nlink(inode, 1);
2820         else {
2821                 ldiskfs_inc_count(oh->ot_handle, inode);
2822                 if (!S_ISDIR(inode->i_mode))
2823                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2824         }
2825         spin_unlock(&obj->oo_guard);
2826
2827         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2828         LINVRNT(osd_invariant(obj));
2829
2830         return rc;
2831 }
2832
2833 static int osd_declare_object_ref_del(const struct lu_env *env,
2834                                       struct dt_object *dt,
2835                                       struct thandle *handle)
2836 {
2837         struct osd_thandle *oh;
2838
2839         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2840         LASSERT(handle != NULL);
2841
2842         oh = container_of0(handle, struct osd_thandle, ot_super);
2843         LASSERT(oh->ot_handle == NULL);
2844
2845         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2846                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2847
2848         return 0;
2849 }
2850
2851 /*
2852  * Concurrency: @dt is write locked.
2853  */
2854 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2855                               struct thandle *th)
2856 {
2857         struct osd_object       *obj = osd_dt_obj(dt);
2858         struct inode            *inode = obj->oo_inode;
2859         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2860         struct osd_thandle      *oh;
2861
2862         LINVRNT(osd_invariant(obj));
2863         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2864         LASSERT(osd_write_locked(env, obj));
2865         LASSERT(th != NULL);
2866
2867         oh = container_of0(th, struct osd_thandle, ot_super);
2868         LASSERT(oh->ot_handle != NULL);
2869
2870         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2871
2872         spin_lock(&obj->oo_guard);
2873         /* That can be result of upgrade from old Lustre version and
2874          * applied only to local files.  Just skip this ref_del call.
2875          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2876         if (inode->i_nlink == 0) {
2877                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2878                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2879                              ", maybe an upgraded file? (LU-3915)\n",
2880                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2881                 spin_unlock(&obj->oo_guard);
2882                 return 0;
2883         }
2884
2885         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2886                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2887
2888         ldiskfs_dec_count(oh->ot_handle, inode);
2889         spin_unlock(&obj->oo_guard);
2890
2891         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2892         LINVRNT(osd_invariant(obj));
2893
2894         return 0;
2895 }
2896
2897 /*
2898  * Get the 64-bit version for an inode.
2899  */
2900 static int osd_object_version_get(const struct lu_env *env,
2901                                   struct dt_object *dt, dt_obj_version_t *ver)
2902 {
2903         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2904
2905         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2906                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2907         *ver = LDISKFS_I(inode)->i_fs_version;
2908         return 0;
2909 }
2910
2911 /*
2912  * Concurrency: @dt is read locked.
2913  */
2914 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2915                          struct lu_buf *buf, const char *name,
2916                          struct lustre_capa *capa)
2917 {
2918         struct osd_object      *obj    = osd_dt_obj(dt);
2919         struct inode           *inode  = obj->oo_inode;
2920         struct osd_thread_info *info   = osd_oti_get(env);
2921         struct dentry          *dentry = &info->oti_obj_dentry;
2922
2923         /* version get is not real XATTR but uses xattr API */
2924         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2925                 /* for version we are just using xattr API but change inode
2926                  * field instead */
2927                 if (buf->lb_len == 0)
2928                         return sizeof(dt_obj_version_t);
2929
2930                 if (buf->lb_len < sizeof(dt_obj_version_t))
2931                         return -ERANGE;
2932
2933                 osd_object_version_get(env, dt, buf->lb_buf);
2934
2935                 return sizeof(dt_obj_version_t);
2936         }
2937
2938         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2939         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2940
2941         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2942                 return -EACCES;
2943
2944         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2945 }
2946
2947
2948 static int osd_declare_xattr_set(const struct lu_env *env,
2949                                  struct dt_object *dt,
2950                                  const struct lu_buf *buf, const char *name,
2951                                  int fl, struct thandle *handle)
2952 {
2953         struct osd_thandle *oh;
2954         int credits;
2955         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2956
2957         LASSERT(handle != NULL);
2958
2959         oh = container_of0(handle, struct osd_thandle, ot_super);
2960         LASSERT(oh->ot_handle == NULL);
2961
2962         /* optimistic optimization: LMA is set first and usually fit inode */
2963         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2964                 if (dt_object_exists(dt))
2965                         credits = 0;
2966                 else
2967                         credits = 1;
2968         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2969                 credits = 1;
2970         } else {
2971                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2972                 if (buf && buf->lb_len > sb->s_blocksize) {
2973                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2974                                         sb->s_blocksize_bits;
2975                 }
2976                 /*
2977                  * xattr set may involve inode quota change, reserve credits for
2978                  * dquot_initialize()
2979                  */
2980                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2981         }
2982
2983         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2984
2985         return 0;
2986 }
2987
2988 /*
2989  * Set the 64-bit version for object
2990  */
2991 static void osd_object_version_set(const struct lu_env *env,
2992                                    struct dt_object *dt,
2993                                    dt_obj_version_t *new_version)
2994 {
2995         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2996
2997         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2998                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2999
3000         LDISKFS_I(inode)->i_fs_version = *new_version;
3001         /** Version is set after all inode operations are finished,
3002          *  so we should mark it dirty here */
3003         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3004 }
3005
3006 /*
3007  * Concurrency: @dt is write locked.
3008  */
3009 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3010                          const struct lu_buf *buf, const char *name, int fl,
3011                          struct thandle *handle, struct lustre_capa *capa)
3012 {
3013         struct osd_object      *obj      = osd_dt_obj(dt);
3014         struct inode           *inode    = obj->oo_inode;
3015         struct osd_thread_info *info     = osd_oti_get(env);
3016         int                     fs_flags = 0;
3017         ENTRY;
3018
3019         LASSERT(handle != NULL);
3020
3021         /* version set is not real XATTR */
3022         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3023                 /* for version we are just using xattr API but change inode
3024                  * field instead */
3025                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3026                 osd_object_version_set(env, dt, buf->lb_buf);
3027                 return sizeof(dt_obj_version_t);
3028         }
3029
3030         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3031                 return -EACCES;
3032
3033         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3034                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3035
3036         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3037         if (fl & LU_XATTR_REPLACE)
3038                 fs_flags |= XATTR_REPLACE;
3039
3040         if (fl & LU_XATTR_CREATE)
3041                 fs_flags |= XATTR_CREATE;
3042
3043         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3044                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3045                 int                      rc;
3046
3047                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3048                 if (rc != 0)
3049                         RETURN(rc);
3050
3051                 lma->lma_incompat |= LMAI_STRIPED;
3052                 lustre_lma_swab(lma);
3053                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3054                                      sizeof(*lma), XATTR_REPLACE);
3055                 if (rc != 0)
3056                         RETURN(rc);
3057         }
3058
3059         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3060             strcmp(name, XATTR_NAME_LINK) == 0)
3061                 return -ENOSPC;
3062
3063         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3064                                fs_flags);
3065 }
3066
3067 /*
3068  * Concurrency: @dt is read locked.
3069  */
3070 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3071                           struct lu_buf *buf, struct lustre_capa *capa)
3072 {
3073         struct osd_object      *obj    = osd_dt_obj(dt);
3074         struct inode           *inode  = obj->oo_inode;
3075         struct osd_thread_info *info   = osd_oti_get(env);
3076         struct dentry          *dentry = &info->oti_obj_dentry;
3077
3078         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3079         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3080
3081         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3082                 return -EACCES;
3083
3084         dentry->d_inode = inode;
3085         dentry->d_sb = inode->i_sb;
3086         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3087 }
3088
3089 static int osd_declare_xattr_del(const struct lu_env *env,
3090                                  struct dt_object *dt, const char *name,
3091                                  struct thandle *handle)
3092 {
3093         struct osd_thandle *oh;
3094         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3095
3096         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3097         LASSERT(handle != NULL);
3098
3099         oh = container_of0(handle, struct osd_thandle, ot_super);
3100         LASSERT(oh->ot_handle == NULL);
3101
3102         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3103                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3104         /*
3105          * xattr del may involve inode quota change, reserve credits for
3106          * dquot_initialize()
3107          */
3108         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3109
3110         return 0;
3111 }
3112
3113 /*
3114  * Concurrency: @dt is write locked.
3115  */
3116 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3117                          const char *name, struct thandle *handle,
3118                          struct lustre_capa *capa)
3119 {
3120         struct osd_object      *obj    = osd_dt_obj(dt);
3121         struct inode           *inode  = obj->oo_inode;
3122         struct osd_thread_info *info   = osd_oti_get(env);
3123         struct dentry          *dentry = &info->oti_obj_dentry;
3124         int                     rc;
3125
3126         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3127         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3128         LASSERT(handle != NULL);
3129
3130         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3131                 return -EACCES;
3132
3133         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3134
3135         ll_vfs_dq_init(inode);
3136         dentry->d_inode = inode;
3137         dentry->d_sb = inode->i_sb;
3138         rc = inode->i_op->removexattr(dentry, name);
3139         return rc;
3140 }
3141
3142 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3143                                      struct dt_object *dt,
3144                                      struct lustre_capa *old, __u64 opc)
3145 {
3146         struct osd_thread_info *info = osd_oti_get(env);
3147         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3148         struct osd_object *obj = osd_dt_obj(dt);
3149         struct osd_device *osd = osd_obj2dev(obj);
3150         struct lustre_capa_key *key = &info->oti_capa_key;
3151         struct lustre_capa *capa = &info->oti_capa;
3152         struct obd_capa *oc;
3153         struct lu_capainfo *lci;
3154         int rc;
3155         ENTRY;
3156
3157         if (!osd->od_fl_capa)
3158                 RETURN(ERR_PTR(-ENOENT));
3159
3160         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3161         LINVRNT(osd_invariant(obj));
3162
3163         /* renewal sanity check */
3164         if (old && osd_object_auth(env, dt, old, opc))
3165                 RETURN(ERR_PTR(-EACCES));
3166
3167         lci = lu_capainfo_get(env);
3168         if (unlikely(lci == NULL))
3169                 RETURN(ERR_PTR(-ENOENT));
3170
3171         switch (lci->lci_auth) {
3172         case LC_ID_NONE:
3173                 RETURN(NULL);
3174         case LC_ID_PLAIN:
3175                 capa->lc_uid = i_uid_read(obj->oo_inode);
3176                 capa->lc_gid = i_gid_read(obj->oo_inode);
3177                 capa->lc_flags = LC_ID_PLAIN;
3178                 break;
3179         case LC_ID_CONVERT: {
3180                 __u32 d[4], s[4];
3181
3182                 s[0] = i_uid_read(obj->oo_inode);
3183                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3184                 s[2] = i_uid_read(obj->oo_inode);
3185                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3186                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3187                 if (unlikely(rc))
3188                         RETURN(ERR_PTR(rc));
3189
3190                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3191                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3192                 capa->lc_flags = LC_ID_CONVERT;
3193                 break;
3194         }
3195         default:
3196                 RETURN(ERR_PTR(-EINVAL));
3197         }
3198
3199         capa->lc_fid = *fid;
3200         capa->lc_opc = opc;
3201         capa->lc_flags |= osd->od_capa_alg << 24;
3202         capa->lc_timeout = osd->od_capa_timeout;
3203         capa->lc_expiry = 0;
3204
3205         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3206         if (oc) {
3207                 LASSERT(!capa_is_expired(oc));
3208                 RETURN(oc);
3209         }
3210
3211         spin_lock(&capa_lock);
3212         *key = osd->od_capa_keys[1];
3213         spin_unlock(&capa_lock);
3214
3215         capa->lc_keyid = key->lk_keyid;
3216         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3217
3218         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3219         if (rc) {
3220                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3221                 RETURN(ERR_PTR(rc));
3222         }
3223
3224         oc = capa_add(osd->od_capa_hash, capa);
3225         RETURN(oc);
3226 }
3227
3228 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3229                            __u64 start, __u64 end)
3230 {
3231         struct osd_object       *obj    = osd_dt_obj(dt);
3232         struct inode            *inode  = obj->oo_inode;
3233         struct osd_thread_info  *info   = osd_oti_get(env);
3234         struct dentry           *dentry = &info->oti_obj_dentry;
3235         struct file             *file   = &info->oti_file;
3236         int                     rc;
3237
3238         ENTRY;
3239
3240         dentry->d_inode = inode;
3241         dentry->d_sb = inode->i_sb;
3242         file->f_dentry = dentry;
3243         file->f_mapping = inode->i_mapping;
3244         file->f_op = inode->i_fop;
3245         set_file_inode(file, inode);
3246
3247 #ifdef HAVE_FILE_FSYNC_4ARGS
3248         rc = file->f_op->fsync(file, start, end, 0);
3249 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3250         mutex_lock(&inode->i_mutex);
3251         rc = file->f_op->fsync(file, 0);
3252         mutex_unlock(&inode->i_mutex);
3253 #else
3254         mutex_lock(&inode->i_mutex);
3255         rc = file->f_op->fsync(file, dentry, 0);
3256         mutex_unlock(&inode->i_mutex);
3257 #endif
3258
3259         RETURN(rc);
3260 }
3261
3262 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3263                         void **data)
3264 {
3265         struct osd_object *obj = osd_dt_obj(dt);
3266         ENTRY;
3267
3268         *data = (void *)obj->oo_inode;
3269         RETURN(0);
3270 }
3271
3272 /*
3273  * Index operations.
3274  */
3275
3276 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3277                            const struct dt_index_features *feat)
3278 {
3279         struct iam_descr *descr;
3280
3281         if (osd_object_is_root(o))
3282                 return feat == &dt_directory_features;
3283
3284         LASSERT(o->oo_dir != NULL);
3285
3286         descr = o->oo_dir->od_container.ic_descr;
3287         if (feat == &dt_directory_features) {
3288                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3289                         return 1;
3290                 else
3291                         return 0;
3292         } else {
3293                 return
3294                         feat->dif_keysize_min <= descr->id_key_size &&
3295                         descr->id_key_size <= feat->dif_keysize_max &&
3296                         feat->dif_recsize_min <= descr->id_rec_size &&
3297                         descr->id_rec_size <= feat->dif_recsize_max &&
3298                         !(feat->dif_flags & (DT_IND_VARKEY |
3299                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3300                         ergo(feat->dif_flags & DT_IND_UPDATE,
3301                              1 /* XXX check that object (and file system) is
3302                                 * writable */);
3303         }
3304 }
3305
3306 static int osd_iam_container_init(const struct lu_env *env,
3307                                   struct osd_object *obj,
3308                                   struct osd_directory *dir)
3309 {
3310         struct iam_container *bag = &dir->od_container;
3311         int result;
3312
3313         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3314         if (result != 0)
3315                 return result;
3316
3317         result = iam_container_setup(bag);
3318         if (result == 0)
3319                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3320         else
3321                 iam_container_fini(bag);
3322
3323         return result;
3324 }
3325
3326
3327 /*
3328  * Concurrency: no external locking is necessary.
3329  */
3330 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3331                          const struct dt_index_features *feat)
3332 {
3333         int                      result;
3334         int                      skip_iam = 0;
3335         struct osd_object       *obj = osd_dt_obj(dt);
3336
3337         LINVRNT(osd_invariant(obj));
3338
3339         if (osd_object_is_root(obj)) {
3340                 dt->do_index_ops = &osd_index_ea_ops;
3341                 result = 0;
3342         } else if (feat == &dt_directory_features) {
3343                 dt->do_index_ops = &osd_index_ea_ops;
3344                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3345                         result = 0;
3346                 else
3347                         result = -ENOTDIR;
3348                 skip_iam = 1;
3349         } else if (unlikely(feat == &dt_otable_features)) {
3350                 dt->do_index_ops = &osd_otable_ops;
3351                 return 0;
3352         } else if (unlikely(feat == &dt_acct_features)) {
3353                 dt->do_index_ops = &osd_acct_index_ops;
3354                 result = 0;
3355                 skip_iam = 1;
3356         } else if (!osd_has_index(obj)) {
3357                 struct osd_directory *dir;
3358
3359                 OBD_ALLOC_PTR(dir);
3360                 if (dir != NULL) {
3361
3362                         spin_lock(&obj->oo_guard);
3363                         if (obj->oo_dir == NULL)
3364                                 obj->oo_dir = dir;
3365                         else
3366                                 /*
3367                                  * Concurrent thread allocated container data.
3368                                  */
3369                                 OBD_FREE_PTR(dir);
3370                         spin_unlock(&obj->oo_guard);
3371                         /*
3372                          * Now, that we have container data, serialize its
3373                          * initialization.
3374                          */
3375                         down_write(&obj->oo_ext_idx_sem);
3376                         /*
3377                          * recheck under lock.
3378                          */
3379                         if (!osd_has_index(obj))
3380                                 result = osd_iam_container_init(env, obj,
3381                                                                 obj->oo_dir);
3382                         else
3383                                 result = 0;
3384                         up_write(&obj->oo_ext_idx_sem);
3385                 } else {
3386                         result = -ENOMEM;
3387                 }
3388         } else {
3389                 result = 0;
3390         }
3391
3392         if (result == 0 && skip_iam == 0) {
3393                 if (!osd_iam_index_probe(env, obj, feat))
3394                         result = -ENOTDIR;
3395         }
3396         LINVRNT(osd_invariant(obj));
3397
3398         if (result == 0 && is_quota_glb_feat(feat) &&
3399             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3400                 result = osd_quota_migration(env, dt, feat);
3401
3402         return result;
3403 }
3404
3405 static int osd_otable_it_attr_get(const struct lu_env *env,
3406                                  struct dt_object *dt,
3407                                  struct lu_attr *attr,
3408                                  struct lustre_capa *capa)
3409 {
3410         attr->la_valid = 0;
3411         return 0;
3412 }
3413
3414 static const struct dt_object_operations osd_obj_ops = {
3415         .do_read_lock         = osd_object_read_lock,
3416         .do_write_lock        = osd_object_write_lock,
3417         .do_read_unlock       = osd_object_read_unlock,
3418         .do_write_unlock      = osd_object_write_unlock,
3419         .do_write_locked      = osd_object_write_locked,
3420         .do_attr_get          = osd_attr_get,
3421         .do_declare_attr_set  = osd_declare_attr_set,
3422         .do_attr_set          = osd_attr_set,
3423         .do_ah_init           = osd_ah_init,
3424         .do_declare_create    = osd_declare_object_create,
3425         .do_create            = osd_object_create,
3426         .do_declare_destroy   = osd_declare_object_destroy,
3427         .do_destroy           = osd_object_destroy,
3428         .do_index_try         = osd_index_try,
3429         .do_declare_ref_add   = osd_declare_object_ref_add,
3430         .do_ref_add           = osd_object_ref_add,
3431         .do_declare_ref_del   = osd_declare_object_ref_del,
3432         .do_ref_del           = osd_object_ref_del,
3433         .do_xattr_get         = osd_xattr_get,
3434         .do_declare_xattr_set = osd_declare_xattr_set,
3435         .do_xattr_set         = osd_xattr_set,
3436         .do_declare_xattr_del = osd_declare_xattr_del,
3437         .do_xattr_del         = osd_xattr_del,
3438         .do_xattr_list        = osd_xattr_list,
3439         .do_capa_get          = osd_capa_get,
3440         .do_object_sync       = osd_object_sync,
3441         .do_data_get          = osd_data_get,
3442 };
3443
3444 /**
3445  * dt_object_operations for interoperability mode
3446  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3447  */
3448 static const struct dt_object_operations osd_obj_ea_ops = {
3449         .do_read_lock         = osd_object_read_lock,
3450         .do_write_lock        = osd_object_write_lock,
3451         .do_read_unlock       = osd_object_read_unlock,
3452         .do_write_unlock      = osd_object_write_unlock,
3453         .do_write_locked      = osd_object_write_locked,
3454         .do_attr_get          = osd_attr_get,
3455         .do_declare_attr_set  = osd_declare_attr_set,
3456         .do_attr_set          = osd_attr_set,
3457         .do_ah_init           = osd_ah_init,
3458         .do_declare_create    = osd_declare_object_create,
3459         .do_create            = osd_object_ea_create,
3460         .do_declare_destroy   = osd_declare_object_destroy,
3461         .do_destroy           = osd_object_destroy,
3462         .do_index_try         = osd_index_try,
3463         .do_declare_ref_add   = osd_declare_object_ref_add,
3464         .do_ref_add           = osd_object_ref_add,
3465         .do_declare_ref_del   = osd_declare_object_ref_del,
3466         .do_ref_del           = osd_object_ref_del,
3467         .do_xattr_get         = osd_xattr_get,
3468         .do_declare_xattr_set = osd_declare_xattr_set,
3469         .do_xattr_set         = osd_xattr_set,
3470         .do_declare_xattr_del = osd_declare_xattr_del,
3471         .do_xattr_del         = osd_xattr_del,
3472         .do_xattr_list        = osd_xattr_list,
3473         .do_capa_get          = osd_capa_get,
3474         .do_object_sync       = osd_object_sync,
3475         .do_data_get          = osd_data_get,
3476 };
3477
3478 static const struct dt_object_operations osd_obj_otable_it_ops = {
3479         .do_attr_get    = osd_otable_it_attr_get,
3480         .do_index_try   = osd_index_try,
3481 };
3482
3483 static int osd_index_declare_iam_delete(const struct lu_env *env,
3484                                         struct dt_object *dt,
3485                                         const struct dt_key *key,
3486                                         struct thandle *handle)
3487 {
3488         struct osd_thandle    *oh;
3489
3490         oh = container_of0(handle, struct osd_thandle, ot_super);
3491         LASSERT(oh->ot_handle == NULL);
3492
3493         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3494                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3495
3496         return 0;
3497 }
3498
3499 /**
3500  *      delete a (key, value) pair from index \a dt specified by \a key
3501  *
3502  *      \param  dt      osd index object
3503  *      \param  key     key for index
3504  *      \param  rec     record reference
3505  *      \param  handle  transaction handler
3506  *
3507  *      \retval  0  success
3508  *      \retval -ve   failure
3509  */
3510
3511 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3512                                 const struct dt_key *key,
3513                                 struct thandle *handle,
3514                                 struct lustre_capa *capa)
3515 {
3516         struct osd_thread_info *oti = osd_oti_get(env);
3517         struct osd_object      *obj = osd_dt_obj(dt);
3518         struct osd_thandle     *oh;
3519         struct iam_path_descr  *ipd;
3520         struct iam_container   *bag = &obj->oo_dir->od_container;
3521         int                     rc;
3522
3523         ENTRY;
3524
3525         LINVRNT(osd_invariant(obj));
3526         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3527         LASSERT(bag->ic_object == obj->oo_inode);
3528         LASSERT(handle != NULL);
3529
3530         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3531                 RETURN(-EACCES);
3532
3533         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3534
3535         ipd = osd_idx_ipd_get(env, bag);
3536         if (unlikely(ipd == NULL))
3537                 RETURN(-ENOMEM);
3538
3539         oh = container_of0(handle, struct osd_thandle, ot_super);
3540         LASSERT(oh->ot_handle != NULL);
3541         LASSERT(oh->ot_handle->h_transaction != NULL);
3542
3543         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3544                 /* swab quota uid/gid provided by caller */
3545                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3546                 key = (const struct dt_key *)&oti->oti_quota_id;
3547         }
3548
3549         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3550         osd_ipd_put(env, bag, ipd);
3551         LINVRNT(osd_invariant(obj));
3552         RETURN(rc);
3553 }
3554
3555 static int osd_index_declare_ea_delete(const struct lu_env *env,
3556                                        struct dt_object *dt,
3557                                        const struct dt_key *key,
3558                                        struct thandle *handle)
3559 {
3560         struct osd_thandle *oh;
3561         struct inode       *inode;
3562         int                 rc;
3563         ENTRY;
3564
3565         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3566         LASSERT(handle != NULL);
3567
3568         oh = container_of0(handle, struct osd_thandle, ot_super);
3569         LASSERT(oh->ot_handle == NULL);
3570
3571         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3572                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3573
3574         inode = osd_dt_obj(dt)->oo_inode;
3575         LASSERT(inode);
3576
3577         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3578                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3579         RETURN(rc);
3580 }
3581
3582 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3583                                           struct dt_rec *fid)
3584 {
3585         struct osd_fid_pack *rec;
3586         int                  rc = -ENODATA;
3587
3588         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3589                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3590                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3591         }
3592         return rc;
3593 }
3594
3595 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3596                           const struct lu_fid *fid)
3597 {
3598         struct seq_server_site  *ss = osd_seq_site(osd);
3599         ENTRY;
3600
3601         /* FID seqs not in FLDB, must be local seq */
3602         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3603                 RETURN(0);
3604
3605         /* If FLD is not being initialized yet, it only happens during the
3606          * initialization, likely during mgs initialization, and we assume
3607          * this is local FID. */
3608         if (ss == NULL || ss->ss_server_fld == NULL)
3609                 RETURN(0);
3610
3611         /* Only check the local FLDB here */
3612         if (osd_seq_exists(env, osd, fid_seq(fid)))
3613                 RETURN(0);
3614
3615         RETURN(1);
3616 }
3617
3618 /**
3619  * Index delete function for interoperability mode (b11826).
3620  * It will remove the directory entry added by osd_index_ea_insert().
3621  * This entry is needed to maintain name->fid mapping.
3622  *
3623  * \param key,  key i.e. file entry to be deleted
3624  *
3625  * \retval   0, on success
3626  * \retval -ve, on error
3627  */
3628 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3629                                const struct dt_key *key,
3630                                struct thandle *handle,
3631                                struct lustre_capa *capa)
3632 {
3633         struct osd_object          *obj    = osd_dt_obj(dt);
3634         struct inode               *dir    = obj->oo_inode;
3635         struct dentry              *dentry;
3636         struct osd_thandle         *oh;
3637         struct ldiskfs_dir_entry_2 *de = NULL;
3638         struct buffer_head         *bh;
3639         struct htree_lock          *hlock = NULL;
3640         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3641         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3642         int                        rc;
3643         ENTRY;
3644
3645         LINVRNT(osd_invariant(obj));
3646         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3647         LASSERT(handle != NULL);
3648
3649         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3650
3651         oh = container_of(handle, struct osd_thandle, ot_super);
3652         LASSERT(oh->ot_handle != NULL);
3653         LASSERT(oh->ot_handle->h_transaction != NULL);
3654
3655         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3656                 RETURN(-EACCES);
3657
3658         ll_vfs_dq_init(dir);
3659         dentry = osd_child_dentry_get(env, obj,
3660                                       (char *)key, strlen((char *)key));
3661
3662         if (obj->oo_hl_head != NULL) {
3663                 hlock = osd_oti_get(env)->oti_hlock;
3664                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3665                                    dir, LDISKFS_HLOCK_DEL);
3666         } else {
3667                 down_write(&obj->oo_ext_idx_sem);
3668         }
3669
3670         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3671         if (bh) {
3672                 __u32 ino = 0;
3673
3674                 /* If this is not the ".." entry, it might be a remote DNE
3675                  * entry and  we need to check if the FID is for a remote
3676                  * MDT.  If the FID is  not in the directory entry (e.g.
3677                  * upgraded 1.8 filesystem without dirdata enabled) then
3678                  * we need to get the FID from the LMA. For a remote directory
3679                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3680                  * case.
3681                  *
3682                  * Delete the entry before the agent inode in order to
3683                  * simplify error handling.  At worst an error after deleting
3684                  * the entry first might leak the agent inode afterward. The
3685                  * reverse would need filesystem abort in case of error deleting
3686                  * the entry after the agent had been removed, or leave a
3687                  * dangling entry pointing at a random inode. */
3688                 if (strcmp((char *)key, dotdot) != 0) {
3689                         LASSERT(de != NULL);
3690                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3691                         /* If Fid is not in dentry, try to get it from LMA */
3692                         if (rc == -ENODATA) {
3693                                 struct osd_inode_id *id;
3694                                 struct inode *inode;
3695
3696                                 /* Before trying to get fid from the inode,
3697                                  * check whether the inode is valid.
3698                                  *
3699                                  * If the inode has been deleted, do not go
3700                                  * ahead to do osd_ea_fid_get, which will set
3701                                  * the inode to bad inode, which might cause
3702                                  * the inode to be deleted uncorrectly */
3703                                 inode = ldiskfs_iget(osd_sb(osd),
3704                                                      le32_to_cpu(de->inode));
3705                                 if (IS_ERR(inode)) {
3706                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3707                                                "error.\n", osd_name(osd),
3708                                                PFID(fid));
3709                                         rc = PTR_ERR(inode);
3710                                 } else {
3711                                         if (likely(inode->i_nlink != 0)) {
3712                                                 id = &osd_oti_get(env)->oti_id;
3713                                                 rc = osd_ea_fid_get(env, obj,
3714                                                         le32_to_cpu(de->inode),
3715                                                                     fid, id);
3716                                         } else {
3717                                                 CDEBUG(D_INFO, "%s: %u "DFID
3718                                                        "deleted.\n",
3719                                                        osd_name(osd),
3720                                                        le32_to_cpu(de->inode),
3721                                                        PFID(fid));
3722                                                 rc = -ESTALE;
3723                                         }
3724                                         iput(inode);
3725                                 }
3726                         }
3727                         if (rc == 0 &&
3728                             unlikely(osd_remote_fid(env, osd, fid)))
3729                                 /* Need to delete agent inode */
3730                                 ino = le32_to_cpu(de->inode);
3731                 }
3732                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3733                 brelse(bh);
3734                 if (rc == 0 && unlikely(ino != 0)) {
3735                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3736                                                           oh);
3737                         if (rc != 0)
3738                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3739                                        osd_name(osd), PFID(fid), rc);
3740                 }
3741         } else {
3742                 rc = -ENOENT;
3743         }
3744         if (hlock != NULL)
3745                 ldiskfs_htree_unlock(hlock);
3746         else
3747                 up_write(&obj->oo_ext_idx_sem);
3748
3749         if (rc != 0)
3750                 GOTO(out, rc);
3751
3752         /* For inode on the remote MDT, .. will point to
3753          * /Agent directory, Check whether it needs to delete
3754          * from agent directory */
3755         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3756                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3757                                                    oh);
3758                 if (rc != 0 && rc != -ENOENT) {
3759                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3760                                osd_name(osd), PFID(fid), rc);
3761                 }
3762
3763                 if (rc == -ENOENT)
3764                         rc = 0;
3765
3766                 GOTO(out, rc);
3767         }
3768 out:
3769
3770         LASSERT(osd_invariant(obj));
3771         RETURN(rc);
3772 }
3773
3774 /**
3775  *      Lookup index for \a key and copy record to \a rec.
3776  *
3777  *      \param  dt      osd index object
3778  *      \param  key     key for index
3779  *      \param  rec     record reference
3780  *
3781  *      \retval  +ve  success : exact mach
3782  *      \retval  0    return record with key not greater than \a key
3783  *      \retval -ve   failure
3784  */
3785 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3786                                 struct dt_rec *rec, const struct dt_key *key,
3787                                 struct lustre_capa *capa)
3788 {
3789         struct osd_object      *obj = osd_dt_obj(dt);
3790         struct iam_path_descr  *ipd;
3791         struct iam_container   *bag = &obj->oo_dir->od_container;
3792         struct osd_thread_info *oti = osd_oti_get(env);
3793         struct iam_iterator    *it = &oti->oti_idx_it;
3794         struct iam_rec         *iam_rec;
3795         int                     rc;
3796
3797         ENTRY;
3798
3799         LASSERT(osd_invariant(obj));
3800         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3801         LASSERT(bag->ic_object == obj->oo_inode);
3802
3803         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3804                 RETURN(-EACCES);
3805
3806         ipd = osd_idx_ipd_get(env, bag);
3807         if (IS_ERR(ipd))
3808                 RETURN(-ENOMEM);
3809
3810         /* got ipd now we can start iterator. */
3811         iam_it_init(it, bag, 0, ipd);
3812
3813         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3814                 /* swab quota uid/gid provided by caller */
3815                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3816                 key = (const struct dt_key *)&oti->oti_quota_id;
3817         }
3818
3819         rc = iam_it_get(it, (struct iam_key *)key);
3820         if (rc >= 0) {
3821                 if (S_ISDIR(obj->oo_inode->i_mode))
3822                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3823                 else
3824                         iam_rec = (struct iam_rec *) rec;
3825
3826                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3827
3828                 if (S_ISDIR(obj->oo_inode->i_mode))
3829                         osd_fid_unpack((struct lu_fid *) rec,
3830                                        (struct osd_fid_pack *)iam_rec);
3831                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3832                         osd_quota_unpack(obj, rec);
3833         }
3834
3835         iam_it_put(it);
3836         iam_it_fini(it);
3837         osd_ipd_put(env, bag, ipd);
3838
3839         LINVRNT(osd_invariant(obj));
3840
3841         RETURN(rc);
3842 }
3843
3844 static int osd_index_declare_iam_insert(const struct lu_env *env,
3845                                         struct dt_object *dt,
3846                                         const struct dt_rec *rec,
3847                                         const struct dt_key *key,
3848                                         struct thandle *handle)
3849 {
3850         struct osd_thandle *oh;
3851
3852         LASSERT(handle != NULL);
3853
3854         oh = container_of0(handle, struct osd_thandle, ot_super);
3855         LASSERT(oh->ot_handle == NULL);
3856
3857         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3858                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3859
3860         return 0;
3861 }
3862
3863 /**
3864  *      Inserts (key, value) pair in \a dt index object.
3865  *
3866  *      \param  dt      osd index object
3867  *      \param  key     key for index
3868  *      \param  rec     record reference
3869  *      \param  th      transaction handler
3870  *
3871  *      \retval  0  success
3872  *      \retval -ve failure
3873  */
3874 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3875                                 const struct dt_rec *rec,
3876                                 const struct dt_key *key, struct thandle *th,
3877                                 struct lustre_capa *capa, int ignore_quota)
3878 {
3879         struct osd_object     *obj = osd_dt_obj(dt);
3880         struct iam_path_descr *ipd;
3881         struct osd_thandle    *oh;
3882         struct iam_container  *bag = &obj->oo_dir->od_container;
3883         struct osd_thread_info *oti = osd_oti_get(env);
3884         struct iam_rec         *iam_rec;
3885         int                     rc;
3886
3887         ENTRY;
3888
3889         LINVRNT(osd_invariant(obj));
3890         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3891         LASSERT(bag->ic_object == obj->oo_inode);
3892         LASSERT(th != NULL);
3893
3894         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3895                 RETURN(-EACCES);
3896
3897         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3898
3899         ipd = osd_idx_ipd_get(env, bag);
3900         if (unlikely(ipd == NULL))
3901                 RETURN(-ENOMEM);
3902
3903         oh = container_of0(th, struct osd_thandle, ot_super);
3904         LASSERT(oh->ot_handle != NULL);
3905         LASSERT(oh->ot_handle->h_transaction != NULL);
3906         if (S_ISDIR(obj->oo_inode->i_mode)) {
3907                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3908                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3909         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3910                 /* pack quota uid/gid */
3911                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3912                 key = (const struct dt_key *)&oti->oti_quota_id;
3913                 /* pack quota record */
3914                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3915                 iam_rec = (struct iam_rec *)rec;
3916         } else {
3917                 iam_rec = (struct iam_rec *)rec;
3918         }
3919
3920         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3921                         iam_rec, ipd);
3922         osd_ipd_put(env, bag, ipd);
3923         LINVRNT(osd_invariant(obj));
3924         RETURN(rc);
3925 }
3926
3927 /**
3928  * Calls ldiskfs_add_entry() to add directory entry
3929  * into the directory. This is required for
3930  * interoperability mode (b11826)
3931  *
3932  * \retval   0, on success
3933  * \retval -ve, on error
3934  */
3935 static int __osd_ea_add_rec(struct osd_thread_info *info,
3936                             struct osd_object *pobj, struct inode  *cinode,
3937                             const char *name, const struct lu_fid *fid,
3938                             struct htree_lock *hlock, struct thandle *th)
3939 {
3940         struct ldiskfs_dentry_param *ldp;
3941         struct dentry               *child;
3942         struct osd_thandle          *oth;
3943         int                          rc;
3944
3945         oth = container_of(th, struct osd_thandle, ot_super);
3946         LASSERT(oth->ot_handle != NULL);
3947         LASSERT(oth->ot_handle->h_transaction != NULL);
3948         LASSERT(pobj->oo_inode);
3949
3950         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3951         if (unlikely(pobj->oo_inode ==
3952                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3953                 ldp->edp_magic = 0;
3954         else
3955                 osd_get_ldiskfs_dirent_param(ldp, fid);
3956         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3957         child->d_fsdata = (void *)ldp;
3958         ll_vfs_dq_init(pobj->oo_inode);
3959         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3960         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3961                 struct ldiskfs_dir_entry_2      *de;
3962                 struct buffer_head              *bh;
3963                 int                              rc1;
3964
3965                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3966                                             NULL, hlock);
3967                 if (bh != NULL) {
3968                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3969                                                                bh);
3970                         if (rc1 == 0) {
3971                                 if (S_ISDIR(cinode->i_mode))
3972                                         de->file_type = LDISKFS_DIRENT_LUFID |
3973                                                         LDISKFS_FT_REG_FILE;
3974                                 else
3975                                         de->file_type = LDISKFS_DIRENT_LUFID |
3976                                                         LDISKFS_FT_DIR;
3977                                 ldiskfs_journal_dirty_metadata(oth->ot_handle,
3978                                                                bh);
3979                                 brelse(bh);
3980                         }
3981                 }
3982         }
3983
3984         RETURN(rc);
3985 }
3986
3987 /**
3988  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3989  * into the directory.Also sets flags into osd object to
3990  * indicate dot and dotdot are created. This is required for
3991  * interoperability mode (b11826)
3992  *
3993  * \param dir   directory for dot and dotdot fixup.
3994  * \param obj   child object for linking
3995  *
3996  * \retval   0, on success
3997  * \retval -ve, on error
3998  */
3999 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4000                               struct osd_object *dir,
4001                               struct inode *parent_dir, const char *name,
4002                               const struct lu_fid *dot_fid,
4003                               const struct lu_fid *dot_dot_fid,
4004                               struct thandle *th)
4005 {
4006         struct inode                *inode = dir->oo_inode;
4007         struct osd_thandle          *oth;
4008         int result = 0;
4009
4010         oth = container_of(th, struct osd_thandle, ot_super);
4011         LASSERT(oth->ot_handle->h_transaction != NULL);
4012         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4013
4014         if (strcmp(name, dot) == 0) {
4015                 if (dir->oo_compat_dot_created) {
4016                         result = -EEXIST;
4017                 } else {
4018                         LASSERT(inode == parent_dir);
4019                         dir->oo_compat_dot_created = 1;
4020                         result = 0;
4021                 }
4022         } else if (strcmp(name, dotdot) == 0) {
4023                 if (!dir->oo_compat_dot_created)
4024                         return -EINVAL;
4025                 /* in case of rename, dotdot is already created */
4026                 if (dir->oo_compat_dotdot_created) {
4027                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4028                                                 dot_dot_fid, NULL, th);
4029                 }
4030
4031                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT) ||
4032                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT2)) {
4033                         struct lu_fid tfid = *dot_dot_fid;
4034
4035                         tfid.f_oid--;
4036                         result = osd_add_dot_dotdot_internal(info,
4037                                         dir->oo_inode, parent_dir, dot_fid,
4038                                         &tfid, oth);
4039                 } else {
4040                         result = osd_add_dot_dotdot_internal(info,
4041                                         dir->oo_inode, parent_dir, dot_fid,
4042                                         dot_dot_fid, oth);
4043                 }
4044
4045                 if (result == 0)
4046                         dir->oo_compat_dotdot_created = 1;
4047         }
4048
4049         return result;
4050 }
4051
4052
4053 /**
4054  * It will call the appropriate osd_add* function and return the
4055  * value, return by respective functions.
4056  */
4057 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4058                           struct inode *cinode, const char *name,
4059                           const struct lu_fid *fid, struct thandle *th)
4060 {
4061         struct osd_thread_info *info   = osd_oti_get(env);
4062         struct htree_lock      *hlock;
4063         int                     rc;
4064
4065         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4066
4067         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4068                                                    name[2] =='\0'))) {
4069                 if (hlock != NULL) {
4070                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4071                                            pobj->oo_inode, 0);
4072                 } else {
4073                         down_write(&pobj->oo_ext_idx_sem);
4074                 }
4075
4076                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4077                                         lu_object_fid(&pobj->oo_dt.do_lu),
4078                                         fid, th);
4079         } else {
4080                 if (hlock != NULL) {
4081                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4082                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4083                 } else {
4084                         down_write(&pobj->oo_ext_idx_sem);
4085                 }
4086
4087                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4088                         struct lu_fid *tfid = &info->oti_fid;
4089
4090                         *tfid = *fid;
4091                         tfid->f_ver = ~0;
4092                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4093                                               tfid, hlock, th);
4094                 } else {
4095                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4096                                               hlock, th);
4097                 }
4098         }
4099         if (hlock != NULL)
4100                 ldiskfs_htree_unlock(hlock);
4101         else
4102                 up_write(&pobj->oo_ext_idx_sem);
4103
4104         return rc;
4105 }
4106
4107 static void
4108 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4109                       struct osd_idmap_cache *oic)
4110 {
4111         struct osd_scrub    *scrub = &dev->od_scrub;
4112         struct lu_fid       *fid   = &oic->oic_fid;
4113         struct osd_inode_id *id    = &oti->oti_id;
4114         int                  once  = 0;
4115         int                  rc;
4116         ENTRY;
4117
4118         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4119                 RETURN_EXIT;
4120
4121 again:
4122         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4123         if (rc != 0 && rc != -ENOENT)
4124                 RETURN_EXIT;
4125
4126         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4127                 RETURN_EXIT;
4128
4129         if (thread_is_running(&scrub->os_thread)) {
4130                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4131                 /* There is race condition between osd_oi_lookup and OI scrub.
4132                  * The OI scrub finished just after osd_oi_lookup() failure.
4133                  * Under such case, it is unnecessary to trigger OI scrub again,
4134                  * but try to call osd_oi_lookup() again. */
4135                 if (unlikely(rc == -EAGAIN))
4136                         goto again;
4137
4138                 RETURN_EXIT;
4139         }
4140
4141         if (!dev->od_noscrub && ++once == 1) {
4142                 rc = osd_scrub_start(dev);
4143                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4144                               ", rc = %d [2]\n",
4145                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4146                               PFID(fid), rc);
4147                 if (rc == 0)
4148                         goto again;
4149         }
4150
4151         EXIT;
4152 }
4153
4154 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4155                                struct osd_device *dev,
4156                                struct osd_idmap_cache *oic,
4157                                struct lu_fid *fid, __u32 ino)
4158 {
4159         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4160         struct inode            *inode;
4161         int                      rc;
4162
4163         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4164         inode = osd_iget(oti, dev, &oic->oic_lid);
4165         if (IS_ERR(inode)) {
4166                 fid_zero(&oic->oic_fid);
4167                 return PTR_ERR(inode);
4168         }
4169
4170         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4171         iput(inode);
4172         if (rc != 0)
4173                 fid_zero(&oic->oic_fid);
4174         else
4175                 *fid = oic->oic_fid = lma->lma_self_fid;
4176         return rc;
4177 }
4178
4179 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4180                      struct osd_inode_id *id, const struct lu_fid *fid)
4181 {
4182         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4183                id->oii_ino, id->oii_gen, info);
4184         info->oti_cache.oic_lid = *id;
4185         info->oti_cache.oic_fid = *fid;
4186         info->oti_cache.oic_dev = osd;
4187
4188         return 0;
4189 }
4190
4191 /**
4192  * Get parent FID from the linkEA.
4193  *
4194  * For a directory which parent resides on remote MDT, to satisfy the
4195  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4196  * the other hand, to make the lookup(..) on the directory can return
4197  * the real parent FID, we append the real parent FID after its ".."
4198  * name entry in the /REMOTE_PARENT_DIR.
4199  *
4200  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4201  * backup. So after the restore, we cannot get the right parent FID from
4202  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4203  * we have stored the real parent FID in the directory object's linkEA,
4204  * we can parse the linkEA for the real parent FID.
4205  *
4206  * \param[in] env       pointer to the thread context
4207  * \param[in] obj       pointer to the object to be handled
4208  * \param[out]fid       pointer to the buffer to hold the parent FID
4209  *
4210  * \retval              0 for getting the real parent FID successfully
4211  * \retval              negative error number on failure
4212  */
4213 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4214                                     struct osd_object *obj,
4215                                     struct lu_fid *fid)
4216 {
4217         struct osd_thread_info  *oti    = osd_oti_get(env);
4218         struct lu_buf           *buf    = &oti->oti_big_buf;
4219         struct dentry           *dentry = &oti->oti_obj_dentry;
4220         struct inode            *inode  = obj->oo_inode;
4221         struct linkea_data       ldata  = { 0 };
4222         int                      rc;
4223         ENTRY;
4224
4225         fid_zero(fid);
4226         if (!S_ISDIR(inode->i_mode))
4227                 RETURN(-EIO);
4228
4229 again:
4230         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4231                              buf->lb_buf, buf->lb_len);
4232         if (rc == -ERANGE) {
4233                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4234                                      NULL, 0);
4235                 if (rc > 0) {
4236                         lu_buf_realloc(buf, rc);
4237                         if (buf->lb_buf == NULL)
4238                                 RETURN(-ENOMEM);
4239
4240                         goto again;
4241                 }
4242         }
4243
4244         if (unlikely(rc == 0))
4245                 RETURN(-ENODATA);
4246
4247         if (rc < 0)
4248                 RETURN(rc);
4249
4250         if (unlikely(buf->lb_buf == NULL)) {
4251                 lu_buf_realloc(buf, rc);
4252                 if (buf->lb_buf == NULL)
4253                         RETURN(-ENOMEM);
4254
4255                 goto again;
4256         }
4257
4258         ldata.ld_buf = buf;
4259         rc = linkea_init(&ldata);
4260         if (rc == 0) {
4261                 linkea_first_entry(&ldata);
4262                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4263         }
4264
4265         RETURN(rc);
4266 }
4267
4268 /**
4269  * Calls ->lookup() to find dentry. From dentry get inode and
4270  * read inode's ea to get fid. This is required for  interoperability
4271  * mode (b11826)
4272  *
4273  * \retval   0, on success
4274  * \retval -ve, on error
4275  */
4276 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4277                              struct dt_rec *rec, const struct dt_key *key)
4278 {
4279         struct inode               *dir    = obj->oo_inode;
4280         struct dentry              *dentry;
4281         struct ldiskfs_dir_entry_2 *de;
4282         struct buffer_head         *bh;
4283         struct lu_fid              *fid = (struct lu_fid *) rec;
4284         struct htree_lock          *hlock = NULL;
4285         int                         ino;
4286         int                         rc;
4287         ENTRY;
4288
4289         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4290
4291         dentry = osd_child_dentry_get(env, obj,
4292                                       (char *)key, strlen((char *)key));
4293
4294         if (obj->oo_hl_head != NULL) {
4295                 hlock = osd_oti_get(env)->oti_hlock;
4296                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4297                                    dir, LDISKFS_HLOCK_LOOKUP);
4298         } else {
4299                 down_read(&obj->oo_ext_idx_sem);
4300         }
4301
4302         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4303         if (bh) {
4304                 struct osd_thread_info *oti = osd_oti_get(env);
4305                 struct osd_inode_id *id = &oti->oti_id;
4306                 struct osd_idmap_cache *oic = &oti->oti_cache;
4307                 struct osd_device *dev = osd_obj2dev(obj);
4308                 struct osd_scrub *scrub = &dev->od_scrub;
4309                 struct scrub_file *sf = &scrub->os_file;
4310
4311                 ino = le32_to_cpu(de->inode);
4312                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4313                         brelse(bh);
4314                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4315                         GOTO(out, rc);
4316                 }
4317
4318                 rc = osd_get_fid_from_dentry(de, rec);
4319
4320                 /* done with de, release bh */
4321                 brelse(bh);
4322                 if (rc != 0) {
4323                         if (unlikely(ino == osd_remote_parent_ino(dev)))
4324                                 /* If the parent is on remote MDT, and there
4325                                  * is no FID-in-dirent, then we have to get
4326                                  * the parent FID from the linkEA.  */
4327                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
4328                         else
4329                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4330                 } else {
4331                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4332                 }
4333
4334                 if (rc != 0) {
4335                         fid_zero(&oic->oic_fid);
4336                         GOTO(out, rc);
4337                 }
4338
4339                 if (osd_remote_fid(env, dev, fid))
4340                         GOTO(out, rc = 0);
4341
4342                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4343                                       fid);
4344                 if (rc != 0)
4345                         GOTO(out, rc);
4346                 if ((scrub->os_pos_current <= ino) &&
4347                     ((sf->sf_flags & SF_INCONSISTENT) ||
4348                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4349                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4350                                       sf->sf_oi_bitmap)))
4351                         osd_consistency_check(oti, dev, oic);
4352         } else {
4353                 rc = -ENOENT;
4354         }
4355
4356         GOTO(out, rc);
4357
4358 out:
4359         if (hlock != NULL)
4360                 ldiskfs_htree_unlock(hlock);
4361         else
4362                 up_read(&obj->oo_ext_idx_sem);
4363         return rc;
4364 }
4365
4366 /**
4367  * Find the osd object for given fid.
4368  *
4369  * \param fid need to find the osd object having this fid
4370  *
4371  * \retval osd_object on success
4372  * \retval        -ve on error
4373  */
4374 struct osd_object *osd_object_find(const struct lu_env *env,
4375                                    struct dt_object *dt,
4376                                    const struct lu_fid *fid)
4377 {
4378         struct lu_device  *ludev = dt->do_lu.lo_dev;
4379         struct osd_object *child = NULL;
4380         struct lu_object  *luch;
4381         struct lu_object  *lo;
4382
4383         /*
4384          * at this point topdev might not exist yet
4385          * (i.e. MGS is preparing profiles). so we can
4386          * not rely on topdev and instead lookup with
4387          * our device passed as topdev. this can't work
4388          * if the object isn't cached yet (as osd doesn't
4389          * allocate lu_header). IOW, the object must be
4390          * in the cache, otherwise lu_object_alloc() crashes
4391          * -bzzz
4392          */
4393         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4394                                  ludev : ludev->ld_site->ls_top_dev,
4395                                  fid, NULL);
4396         if (!IS_ERR(luch)) {
4397                 if (lu_object_exists(luch)) {
4398                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4399                         if (lo != NULL)
4400                                 child = osd_obj(lo);
4401                         else
4402                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4403                                                 "lu_object can't be located"
4404                                                 DFID"\n", PFID(fid));
4405
4406                         if (child == NULL) {
4407                                 lu_object_put(env, luch);
4408                                 CERROR("Unable to get osd_object\n");
4409                                 child = ERR_PTR(-ENOENT);
4410                         }
4411                 } else {
4412                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4413                                         "lu_object does not exists "DFID"\n",
4414                                         PFID(fid));
4415                         lu_object_put(env, luch);
4416                         child = ERR_PTR(-ENOENT);
4417                 }
4418         } else {
4419                 child = ERR_CAST(luch);
4420         }
4421
4422         return child;
4423 }
4424
4425 /**
4426  * Put the osd object once done with it.
4427  *
4428  * \param obj osd object that needs to be put
4429  */
4430 static inline void osd_object_put(const struct lu_env *env,
4431                                   struct osd_object *obj)
4432 {
4433         lu_object_put(env, &obj->oo_dt.do_lu);
4434 }
4435
4436 static int osd_index_declare_ea_insert(const struct lu_env *env,
4437                                        struct dt_object *dt,
4438                                        const struct dt_rec *rec,
4439                                        const struct dt_key *key,
4440                                        struct thandle *handle)
4441 {
4442         struct osd_thandle      *oh;
4443         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4444         struct lu_fid           *fid = (struct lu_fid *)rec;
4445         int                     rc;
4446         ENTRY;
4447
4448         LASSERT(!dt_object_remote(dt));
4449         LASSERT(handle != NULL);
4450
4451         oh = container_of0(handle, struct osd_thandle, ot_super);
4452         LASSERT(oh->ot_handle == NULL);
4453
4454         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4455                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4456
4457         if (osd_dt_obj(dt)->oo_inode != NULL) {
4458                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4459
4460                 /* We ignore block quota on meta pool (MDTs), so needn't
4461                  * calculate how many blocks will be consumed by this index
4462                  * insert */
4463                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4464                                            i_gid_read(inode), 0, oh,
4465                                            osd_dt_obj(dt), true, NULL, false);
4466         }
4467
4468         if (fid == NULL)
4469                 RETURN(0);
4470
4471         rc = osd_remote_fid(env, osd, fid);
4472         if (rc <= 0)
4473                 RETURN(rc);
4474
4475         rc = 0;
4476
4477         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4478                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4479         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4480                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4481         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4482                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4483
4484         RETURN(rc);
4485 }
4486
4487 /**
4488  * Index add function for interoperability mode (b11826).
4489  * It will add the directory entry.This entry is needed to
4490  * maintain name->fid mapping.
4491  *
4492  * \param key it is key i.e. file entry to be inserted
4493  * \param rec it is value of given key i.e. fid
4494  *
4495  * \retval   0, on success
4496  * \retval -ve, on error
4497  */
4498 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4499                                const struct dt_rec *rec,
4500                                const struct dt_key *key, struct thandle *th,
4501                                struct lustre_capa *capa, int ignore_quota)
4502 {
4503         struct osd_object       *obj = osd_dt_obj(dt);
4504         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4505         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4506         const struct lu_fid     *fid    = rec1->rec_fid;
4507         const char              *name = (const char *)key;
4508         struct osd_thread_info  *oti   = osd_oti_get(env);
4509         struct osd_inode_id     *id    = &oti->oti_id;
4510         struct inode            *child_inode = NULL;
4511         struct osd_object       *child = NULL;
4512         int                     rc;
4513         ENTRY;
4514
4515         LASSERT(osd_invariant(obj));
4516         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4517         LASSERT(th != NULL);
4518
4519         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4520
4521         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4522                 RETURN(-EACCES);
4523
4524         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4525
4526         rc = osd_remote_fid(env, osd, fid);
4527         if (rc < 0) {
4528                 CERROR("%s: Can not find object "DFID" rc %d\n",
4529                        osd_name(osd), PFID(fid), rc);
4530                 RETURN(rc);
4531         }
4532
4533         if (rc == 1) {
4534                 /* Insert remote entry */
4535                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4536                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4537                         struct osd_thandle      *oh;
4538
4539                         /* If parent on remote MDT, we need put this object
4540                          * under AGENT */
4541                         oh = container_of(th, typeof(*oh), ot_super);
4542                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4543                         if (rc != 0) {
4544                                 CERROR("%s: add "DFID" error: rc = %d\n",
4545                                        osd_name(osd),
4546                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4547                                 RETURN(rc);
4548                         }
4549
4550                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4551                 } else {
4552                         child_inode = osd_create_local_agent_inode(env, osd,
4553                                         obj, fid, rec1->rec_type & S_IFMT, th);
4554                         if (IS_ERR(child_inode))
4555                                 RETURN(PTR_ERR(child_inode));
4556                 }
4557         } else {
4558                 /* Insert local entry */
4559                 child = osd_object_find(env, dt, fid);
4560                 if (IS_ERR(child)) {
4561                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4562                                osd_name(osd), PFID(fid),
4563                                id->oii_ino, id->oii_gen,
4564                                (int)PTR_ERR(child));
4565                         RETURN(PTR_ERR(child));
4566                 }
4567                 child_inode = igrab(child->oo_inode);
4568         }
4569
4570         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4571
4572         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4573                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4574
4575         iput(child_inode);
4576         if (child != NULL)
4577                 osd_object_put(env, child);
4578         LASSERT(osd_invariant(obj));
4579         RETURN(rc);
4580 }
4581
4582 /**
4583  *  Initialize osd Iterator for given osd index object.
4584  *
4585  *  \param  dt      osd index object
4586  */
4587
4588 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4589                                      struct dt_object *dt,
4590                                      __u32 unused,
4591                                      struct lustre_capa *capa)
4592 {
4593         struct osd_it_iam      *it;
4594         struct osd_thread_info *oti = osd_oti_get(env);
4595         struct osd_object      *obj = osd_dt_obj(dt);
4596         struct lu_object       *lo  = &dt->do_lu;
4597         struct iam_path_descr  *ipd;
4598         struct iam_container   *bag = &obj->oo_dir->od_container;
4599
4600         LASSERT(lu_object_exists(lo));
4601
4602         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4603                 return ERR_PTR(-EACCES);
4604
4605         if (oti->oti_it_inline) {
4606                 OBD_ALLOC_PTR(it);
4607                 if (it == NULL)
4608                         return ERR_PTR(-ENOMEM);
4609         } else {
4610                 it = &oti->oti_it;
4611                 oti->oti_it_inline = 1;
4612         }
4613
4614         ipd = osd_it_ipd_get(env, bag);
4615         if (likely(ipd != NULL)) {
4616                 it->oi_obj = obj;
4617                 it->oi_ipd = ipd;
4618                 lu_object_get(lo);
4619                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4620                 return (struct dt_it *)it;
4621         } else {
4622                 if (it != &oti->oti_it)
4623                         OBD_FREE_PTR(it);
4624                 else
4625                         oti->oti_it_inline = 0;
4626
4627                 return ERR_PTR(-ENOMEM);
4628         }
4629 }
4630
4631 /**
4632  * free given Iterator.
4633  */
4634
4635 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4636 {
4637         struct osd_thread_info  *oti = osd_oti_get(env);
4638         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4639         struct osd_object       *obj = it->oi_obj;
4640
4641         iam_it_fini(&it->oi_it);
4642         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4643         lu_object_put(env, &obj->oo_dt.do_lu);
4644         if (it != &oti->oti_it)
4645                 OBD_FREE_PTR(it);
4646         else
4647                 oti->oti_it_inline = 0;
4648 }
4649
4650 /**
4651  *  Move Iterator to record specified by \a key
4652  *
4653  *  \param  di      osd iterator
4654  *  \param  key     key for index
4655  *
4656  *  \retval +ve  di points to record with least key not larger than key
4657  *  \retval  0   di points to exact matched key
4658  *  \retval -ve  failure
4659  */
4660
4661 static int osd_it_iam_get(const struct lu_env *env,
4662                           struct dt_it *di, const struct dt_key *key)
4663 {
4664         struct osd_thread_info  *oti = osd_oti_get(env);
4665         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4666
4667         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4668                 /* swab quota uid/gid */
4669                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4670                 key = (struct dt_key *)&oti->oti_quota_id;
4671         }
4672
4673         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4674 }
4675
4676 /**
4677  *  Release Iterator
4678  *
4679  *  \param  di      osd iterator
4680  */
4681 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4682 {
4683         struct osd_it_iam *it = (struct osd_it_iam *)di;
4684
4685         iam_it_put(&it->oi_it);
4686 }
4687
4688 /**
4689  *  Move iterator by one record
4690  *
4691  *  \param  di      osd iterator
4692  *
4693  *  \retval +1   end of container reached
4694  *  \retval  0   success
4695  *  \retval -ve  failure
4696  */
4697
4698 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4699 {
4700         struct osd_it_iam *it = (struct osd_it_iam *)di;
4701
4702         return iam_it_next(&it->oi_it);
4703 }
4704
4705 /**
4706  * Return pointer to the key under iterator.
4707  */
4708
4709 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4710                                  const struct dt_it *di)
4711 {
4712         struct osd_thread_info *oti = osd_oti_get(env);
4713         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4714         struct osd_object      *obj = it->oi_obj;
4715         struct dt_key          *key;
4716
4717         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4718
4719         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4720                 /* swab quota uid/gid */
4721                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4722                 key = (struct dt_key *)&oti->oti_quota_id;
4723         }
4724
4725         return key;
4726 }
4727
4728 /**
4729  * Return size of key under iterator (in bytes)
4730  */
4731
4732 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4733 {
4734         struct osd_it_iam *it = (struct osd_it_iam *)di;
4735
4736         return iam_it_key_size(&it->oi_it);
4737 }
4738
4739 static inline void
4740 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4741 {
4742         /* check if file type is required */
4743         if (ent->lde_attrs & LUDA_TYPE) {
4744                 struct luda_type *lt;
4745                 int align = sizeof(*lt) - 1;
4746
4747                 len = (len + align) & ~align;
4748                 lt = (struct luda_type *)(ent->lde_name + len);
4749                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4750         }
4751
4752         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4753 }
4754
4755 /**
4756  * build lu direct from backend fs dirent.
4757  */
4758
4759 static inline void
4760 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4761                    char *name, __u16 namelen, __u16 type, __u32 attr)
4762 {
4763         ent->lde_attrs = attr | LUDA_FID;
4764         fid_cpu_to_le(&ent->lde_fid, fid);
4765
4766         ent->lde_hash = cpu_to_le64(offset);
4767         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4768
4769         strncpy(ent->lde_name, name, namelen);
4770         ent->lde_name[namelen] = '\0';
4771         ent->lde_namelen = cpu_to_le16(namelen);
4772
4773         /* append lustre attributes */
4774         osd_it_append_attrs(ent, namelen, type);
4775 }
4776
4777 /**
4778  * Return pointer to the record under iterator.
4779  */
4780 static int osd_it_iam_rec(const struct lu_env *env,
4781                           const struct dt_it *di,
4782                           struct dt_rec *dtrec, __u32 attr)
4783 {
4784         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4785         struct osd_thread_info *info = osd_oti_get(env);
4786         ENTRY;
4787
4788         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4789                 const struct osd_fid_pack *rec;
4790                 struct lu_fid             *fid = &info->oti_fid;
4791                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4792                 char                      *name;
4793                 int                        namelen;
4794                 __u64                      hash;
4795                 int                        rc;
4796
4797                 name = (char *)iam_it_key_get(&it->oi_it);
4798                 if (IS_ERR(name))
4799                         RETURN(PTR_ERR(name));
4800
4801                 namelen = iam_it_key_size(&it->oi_it);
4802
4803                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4804                 if (IS_ERR(rec))
4805                         RETURN(PTR_ERR(rec));
4806
4807                 rc = osd_fid_unpack(fid, rec);
4808                 if (rc)
4809                         RETURN(rc);
4810
4811                 hash = iam_it_store(&it->oi_it);
4812
4813                 /* IAM does not store object type in IAM index (dir) */
4814                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4815                                    0, LUDA_FID);
4816         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4817                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4818                            (struct iam_rec *)dtrec);
4819                 osd_quota_unpack(it->oi_obj, dtrec);
4820         } else {
4821                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4822                            (struct iam_rec *)dtrec);
4823         }
4824
4825         RETURN(0);
4826 }
4827
4828 /**
4829  * Returns cookie for current Iterator position.
4830  */
4831 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4832 {
4833         struct osd_it_iam *it = (struct osd_it_iam *)di;
4834
4835         return iam_it_store(&it->oi_it);
4836 }
4837
4838 /**
4839  * Restore iterator from cookie.
4840  *
4841  * \param  di      osd iterator
4842  * \param  hash    Iterator location cookie
4843  *
4844  * \retval +ve  di points to record with least key not larger than key.
4845  * \retval  0   di points to exact matched key
4846  * \retval -ve  failure
4847  */
4848
4849 static int osd_it_iam_load(const struct lu_env *env,
4850                            const struct dt_it *di, __u64 hash)
4851 {
4852         struct osd_it_iam *it = (struct osd_it_iam *)di;
4853
4854         return iam_it_load(&it->oi_it, hash);
4855 }
4856
4857 static const struct dt_index_operations osd_index_iam_ops = {
4858         .dio_lookup         = osd_index_iam_lookup,
4859         .dio_declare_insert = osd_index_declare_iam_insert,
4860         .dio_insert         = osd_index_iam_insert,
4861         .dio_declare_delete = osd_index_declare_iam_delete,
4862         .dio_delete         = osd_index_iam_delete,
4863         .dio_it     = {
4864                 .init     = osd_it_iam_init,
4865                 .fini     = osd_it_iam_fini,
4866                 .get      = osd_it_iam_get,
4867                 .put      = osd_it_iam_put,
4868                 .next     = osd_it_iam_next,
4869                 .key      = osd_it_iam_key,
4870                 .key_size = osd_it_iam_key_size,
4871                 .rec      = osd_it_iam_rec,
4872                 .store    = osd_it_iam_store,
4873                 .load     = osd_it_iam_load
4874         }
4875 };
4876
4877
4878 /**
4879  * Creates or initializes iterator context.
4880  *
4881  * \retval struct osd_it_ea, iterator structure on success
4882  *
4883  */
4884 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4885                                     struct dt_object *dt,
4886                                     __u32 attr,
4887                                     struct lustre_capa *capa)
4888 {
4889         struct osd_object       *obj  = osd_dt_obj(dt);
4890         struct osd_thread_info  *info = osd_oti_get(env);
4891         struct osd_it_ea        *it;
4892         struct file             *file;
4893         struct lu_object        *lo   = &dt->do_lu;
4894         struct dentry           *obj_dentry = &info->oti_it_dentry;
4895         ENTRY;
4896
4897         LASSERT(lu_object_exists(lo));
4898
4899         if (info->oti_it_inline) {
4900                 OBD_ALLOC_PTR(it);
4901                 if (it == NULL)
4902                         RETURN(ERR_PTR(-ENOMEM));
4903         } else {
4904                 it = &info->oti_it_ea;
4905                 info->oti_it_inline = 1;
4906         }
4907
4908         obj_dentry->d_inode = obj->oo_inode;
4909         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4910         obj_dentry->d_name.hash = 0;
4911
4912         it->oie_rd_dirent       = 0;
4913         it->oie_it_dirent       = 0;
4914         it->oie_dirent          = NULL;
4915         it->oie_buf             = info->oti_it_ea_buf;
4916         it->oie_obj             = obj;
4917
4918         file = &it->oie_file;
4919         /* Reset the "file" totally to avoid to reuse any old value from
4920          * former readdir handling, the "file->f_pos" should be zero. */
4921         memset(file, 0, sizeof(*file));
4922         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4923         if (attr & LUDA_64BITHASH)
4924                 file->f_mode    = FMODE_64BITHASH;
4925         else
4926                 file->f_mode    = FMODE_32BITHASH;
4927         file->f_dentry          = obj_dentry;
4928         file->f_mapping         = obj->oo_inode->i_mapping;
4929         file->f_op              = obj->oo_inode->i_fop;
4930         set_file_inode(file, obj->oo_inode);
4931
4932         lu_object_get(lo);
4933         RETURN((struct dt_it *) it);
4934 }
4935
4936 /**
4937  * Destroy or finishes iterator context.
4938  *
4939  * \param di iterator structure to be destroyed
4940  */
4941 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4942 {
4943         struct osd_thread_info  *info   = osd_oti_get(env);
4944         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
4945         struct osd_object       *obj    = it->oie_obj;
4946         struct inode            *inode  = obj->oo_inode;
4947
4948         ENTRY;
4949         it->oie_file.f_op->release(inode, &it->oie_file);
4950         lu_object_put(env, &obj->oo_dt.do_lu);
4951         if (it != &info->oti_it_ea)
4952                 OBD_FREE_PTR(it);
4953         else
4954                 info->oti_it_inline = 0;
4955         EXIT;
4956 }
4957
4958 /**
4959  * It position the iterator at given key, so that next lookup continues from
4960  * that key Or it is similar to dio_it->load() but based on a key,
4961  * rather than file position.
4962  *
4963  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4964  * to the beginning.
4965  *
4966  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4967  */
4968 static int osd_it_ea_get(const struct lu_env *env,
4969                          struct dt_it *di, const struct dt_key *key)
4970 {
4971         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4972
4973         ENTRY;
4974         LASSERT(((const char *)key)[0] == '\0');
4975         it->oie_file.f_pos      = 0;
4976         it->oie_rd_dirent       = 0;
4977         it->oie_it_dirent       = 0;
4978         it->oie_dirent          = NULL;
4979
4980         RETURN(+1);
4981 }
4982
4983 /**
4984  * Does nothing
4985  */
4986 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4987 {
4988 }
4989
4990 struct osd_filldir_cbs {
4991 #ifdef HAVE_DIR_CONTEXT
4992         struct dir_context ctx;
4993 #endif
4994         struct osd_it_ea  *it;
4995 };
4996 /**
4997  * It is called internally by ->readdir(). It fills the
4998  * iterator's in-memory data structure with required
4999  * information i.e. name, namelen, rec_size etc.
5000  *
5001  * \param buf in which information to be filled in.
5002  * \param name name of the file in given dir
5003  *
5004  * \retval 0 on success
5005  * \retval 1 on buffer full
5006  */
5007 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5008                                loff_t offset, __u64 ino,
5009                                unsigned d_type)
5010 {
5011         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5012         struct osd_object       *obj  = it->oie_obj;
5013         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5014         struct lu_fid           *fid  = &ent->oied_fid;
5015         struct osd_fid_pack     *rec;
5016         ENTRY;
5017
5018         /* this should never happen */
5019         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5020                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5021                 RETURN(-EIO);
5022         }
5023
5024         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5025             OSD_IT_EA_BUFSIZE)
5026                 RETURN(1);
5027
5028         /* "." is just the object itself. */
5029         if (namelen == 1 && name[0] == '.') {
5030                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5031         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5032                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5033                 if (osd_fid_unpack(fid, rec) != 0)
5034                         fid_zero(fid);
5035         } else {
5036                 fid_zero(fid);
5037         }
5038         d_type &= ~LDISKFS_DIRENT_LUFID;
5039
5040         /* NOT export local root. */
5041         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5042                 ino = obj->oo_inode->i_ino;
5043                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5044         }
5045
5046         ent->oied_ino     = ino;
5047         ent->oied_off     = offset;
5048         ent->oied_namelen = namelen;
5049         ent->oied_type    = d_type;
5050
5051         memcpy(ent->oied_name, name, namelen);
5052
5053         it->oie_rd_dirent++;
5054         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5055         RETURN(0);
5056 }
5057
5058 /**
5059  * Calls ->readdir() to load a directory entry at a time
5060  * and stored it in iterator's in-memory data structure.
5061  *
5062  * \param di iterator's in memory structure
5063  *
5064  * \retval   0 on success
5065  * \retval -ve on error
5066  * \retval +1 reach the end of entry
5067  */
5068 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5069                                const struct dt_it *di)
5070 {
5071         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5072         struct osd_object  *obj   = it->oie_obj;
5073         struct inode       *inode = obj->oo_inode;
5074         struct htree_lock  *hlock = NULL;
5075         struct file        *filp  = &it->oie_file;
5076         int                 rc = 0;
5077         struct osd_filldir_cbs buf = {
5078 #ifdef HAVE_DIR_CONTEXT
5079                 .ctx.actor = osd_ldiskfs_filldir,
5080 #endif
5081                 .it = it
5082         };
5083
5084         ENTRY;
5085         it->oie_dirent = it->oie_buf;
5086         it->oie_rd_dirent = 0;
5087
5088         if (obj->oo_hl_head != NULL) {
5089                 hlock = osd_oti_get(env)->oti_hlock;
5090                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5091                                    inode, LDISKFS_HLOCK_READDIR);
5092         } else {
5093                 down_read(&obj->oo_ext_idx_sem);
5094         }
5095
5096 #ifdef HAVE_DIR_CONTEXT
5097         buf.ctx.pos = filp->f_pos;
5098         rc = inode->i_fop->iterate(filp, &buf.ctx);
5099         filp->f_pos = buf.ctx.pos;
5100 #else
5101         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5102 #endif
5103
5104         if (hlock != NULL)
5105                 ldiskfs_htree_unlock(hlock);
5106         else
5107                 up_read(&obj->oo_ext_idx_sem);
5108
5109         if (it->oie_rd_dirent == 0) {
5110                 /*If it does not get any dirent, it means it has been reached
5111                  *to the end of the dir */
5112                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5113                 if (rc == 0)
5114                         rc = 1;
5115         } else {
5116                 it->oie_dirent = it->oie_buf;
5117                 it->oie_it_dirent = 1;
5118         }
5119
5120         RETURN(rc);
5121 }
5122
5123 /**
5124  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5125  * to load a directory entry at a time and stored it in
5126  * iterator's in-memory data structure.
5127  *
5128  * \param di iterator's in memory structure
5129  *
5130  * \retval +ve iterator reached to end
5131  * \retval   0 iterator not reached to end
5132  * \retval -ve on error
5133  */
5134 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5135 {
5136         struct osd_it_ea *it = (struct osd_it_ea *)di;
5137         int rc;
5138
5139         ENTRY;
5140
5141         if (it->oie_it_dirent < it->oie_rd_dirent) {
5142                 it->oie_dirent =
5143                         (void *) it->oie_dirent +
5144                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5145                                        it->oie_dirent->oied_namelen);
5146                 it->oie_it_dirent++;
5147                 RETURN(0);
5148         } else {
5149                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5150                         rc = +1;
5151                 else
5152                         rc = osd_ldiskfs_it_fill(env, di);
5153         }
5154
5155         RETURN(rc);
5156 }
5157
5158 /**
5159  * Returns the key at current position from iterator's in memory structure.
5160  *
5161  * \param di iterator's in memory structure
5162  *
5163  * \retval key i.e. struct dt_key on success
5164  */
5165 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5166                                     const struct dt_it *di)
5167 {
5168         struct osd_it_ea *it = (struct osd_it_ea *)di;
5169
5170         return (struct dt_key *)it->oie_dirent->oied_name;
5171 }
5172
5173 /**
5174  * Returns the key's size at current position from iterator's in memory structure.
5175  *
5176  * \param di iterator's in memory structure
5177  *
5178  * \retval key_size i.e. struct dt_key on success
5179  */
5180 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5181 {
5182         struct osd_it_ea *it = (struct osd_it_ea *)di;
5183
5184         return it->oie_dirent->oied_namelen;
5185 }
5186
5187 static int
5188 osd_dirent_update(handle_t *jh, struct super_block *sb,
5189                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5190                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5191 {
5192         struct osd_fid_pack *rec;
5193         int                  rc;
5194         ENTRY;
5195
5196         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5197         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5198
5199         rc = ldiskfs_journal_get_write_access(jh, bh);
5200         if (rc != 0)
5201                 RETURN(rc);
5202
5203         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5204         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5205         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5206
5207         RETURN(rc);
5208 }
5209
5210 static inline int
5211 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5212 {
5213         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5214             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5215                 return 1;
5216         else
5217                 return 0;
5218 }
5219
5220 static inline int
5221 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5222 {
5223         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5224                  "dot_dotdot = %d\n", dot_dotdot);
5225
5226         if (LDISKFS_DIR_REC_LEN(de) >=
5227             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5228                 return 1;
5229         else
5230                 return 0;
5231 }
5232
5233 static int
5234 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5235                     struct inode *dir, struct inode *inode,
5236                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5237                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5238                     struct htree_lock *hlock)
5239 {
5240         struct dentry               *dentry;
5241         struct osd_fid_pack         *rec;
5242         struct ldiskfs_dentry_param *ldp;
5243         int                          rc;
5244         ENTRY;
5245
5246         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5247                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5248                 RETURN(0);
5249
5250         /* There is enough space to hold the FID-in-dirent. */
5251         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5252                                  dir->i_sb->s_blocksize)) {
5253                 rc = ldiskfs_journal_get_write_access(jh, bh);
5254                 if (rc != 0)
5255                         RETURN(rc);
5256
5257                 de->name[de->name_len] = 0;
5258                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5259                 rec->fp_len = sizeof(struct lu_fid) + 1;
5260                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5261                 de->file_type |= LDISKFS_DIRENT_LUFID;
5262
5263                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5264
5265                 RETURN(rc);
5266         }
5267
5268         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5269         if (rc != 0)
5270                 RETURN(rc);
5271
5272         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5273                                            ent->oied_namelen);
5274         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5275         osd_get_ldiskfs_dirent_param(ldp, fid);
5276         dentry->d_fsdata = (void *)ldp;
5277         ll_vfs_dq_init(dir);
5278         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5279         /* It is too bad, we cannot reinsert the name entry back.
5280          * That means we lose it! */
5281         if (rc != 0)
5282                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5283                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5284                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5285                        dir->i_ino, dir->i_generation,
5286                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5287
5288         RETURN(rc);
5289 }
5290
5291 static int
5292 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5293                         struct osd_it_ea *it, struct lu_fid *fid,
5294                         struct osd_inode_id *id, __u32 *attr)
5295 {
5296         struct osd_thread_info     *info        = osd_oti_get(env);
5297         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5298         struct osd_device          *dev         = osd_obj2dev(obj);
5299         struct super_block         *sb          = osd_sb(dev);
5300         const char                 *devname     =
5301                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5302         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5303         struct inode               *dir         = obj->oo_inode;
5304         struct htree_lock          *hlock       = NULL;
5305         struct buffer_head         *bh          = NULL;
5306         handle_t                   *jh          = NULL;
5307         struct ldiskfs_dir_entry_2 *de;
5308         struct dentry              *dentry;
5309         struct inode               *inode;
5310         int                         credits;
5311         int                         rc;
5312         int                         dot_dotdot  = 0;
5313         bool                        dirty       = false;
5314         ENTRY;
5315
5316         if (ent->oied_name[0] == '.') {
5317                 if (ent->oied_namelen == 1)
5318                         dot_dotdot = 1;
5319                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5320                         dot_dotdot = 2;
5321         }
5322
5323         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5324                                       ent->oied_namelen);
5325
5326         /* We need to ensure that the name entry is still valid.
5327          * Because it may be removed or renamed by other already.
5328          *
5329          * The unlink or rename operation will start journal before PDO lock,
5330          * so to avoid deadlock, here we need to start journal handle before
5331          * related PDO lock also. But because we do not know whether there
5332          * will be something to be repaired before PDO lock, we just start
5333          * journal without conditions.
5334          *
5335          * We may need to remove the name entry firstly, then insert back.
5336          * One credit is for user quota file update.
5337          * One credit is for group quota file update.
5338          * Two credits are for dirty inode. */
5339         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5340                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5341
5342 again:
5343         if (dev->od_dirent_journal) {
5344                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5345                 if (IS_ERR(jh)) {
5346                         rc = PTR_ERR(jh);
5347                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5348                                "check_repair, dir = %lu/%u, credits = %d, "
5349                                "name = %.*s: rc = %d\n",
5350                                devname, dir->i_ino, dir->i_generation, credits,
5351                                ent->oied_namelen, ent->oied_name, rc);
5352                         RETURN(rc);
5353                 }
5354
5355                 if (obj->oo_hl_head != NULL) {
5356                         hlock = osd_oti_get(env)->oti_hlock;
5357                         /* "0" means exclusive lock for the whole directory.
5358                          * We need to prevent others access such name entry
5359                          * during the delete + insert. Neither HLOCK_ADD nor
5360                          * HLOCK_DEL cannot guarantee the atomicity. */
5361                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5362                 } else {
5363                         down_write(&obj->oo_ext_idx_sem);
5364                 }
5365         } else {
5366                 if (obj->oo_hl_head != NULL) {
5367                         hlock = osd_oti_get(env)->oti_hlock;
5368                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5369                                            LDISKFS_HLOCK_LOOKUP);
5370                 } else {
5371                         down_read(&obj->oo_ext_idx_sem);
5372                 }
5373         }
5374
5375         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5376         /* For dot/dotdot entry, if there is not enough space to hold the
5377          * FID-in-dirent, just keep them there. It only happens when the
5378          * device upgraded from 1.8 or restored from MDT file-level backup.
5379          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5380          * and needs to get FID from LMA when readdir, it will not affect the
5381          * performance much. */
5382         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5383             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5384                 *attr |= LUDA_IGNORE;
5385                 GOTO(out_journal, rc = 0);
5386         }
5387
5388         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5389         inode = osd_iget(info, dev, id);
5390         if (IS_ERR(inode)) {
5391                 rc = PTR_ERR(inode);
5392                 if (rc == -ENOENT || rc == -ESTALE)
5393                         rc = 1;
5394                 else
5395                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5396                                "check_repair, dir = %lu/%u, name = %.*s: "
5397                                "rc = %d\n",
5398                                devname, dir->i_ino, dir->i_generation,
5399                                ent->oied_namelen, ent->oied_name, rc);
5400
5401                 GOTO(out_journal, rc);
5402         }
5403
5404         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5405         if (rc == 0) {
5406                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5407
5408                 if (fid_is_sane(fid)) {
5409                         /* FID-in-dirent is valid. */
5410                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5411                                 GOTO(out_inode, rc = 0);
5412
5413                         /* Do not repair under dryrun mode. */
5414                         if (*attr & LUDA_VERIFY_DRYRUN) {
5415                                 *attr |= LUDA_REPAIR;
5416                                 GOTO(out_inode, rc = 0);
5417                         }
5418
5419                         if (!dev->od_dirent_journal) {
5420                                 iput(inode);
5421                                 brelse(bh);
5422                                 if (hlock != NULL)
5423                                         ldiskfs_htree_unlock(hlock);
5424                                 else
5425                                         up_read(&obj->oo_ext_idx_sem);
5426                                 dev->od_dirent_journal = 1;
5427                                 goto again;
5428                         }
5429
5430                         *fid = lma->lma_self_fid;
5431                         dirty = true;
5432                         /* Update the FID-in-dirent. */
5433                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5434                         if (rc == 0)
5435                                 *attr |= LUDA_REPAIR;
5436                         else
5437                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5438                                        "in the dirent, dir = %lu/%u, "
5439                                        "name = %.*s, "DFID": rc = %d\n",
5440                                        devname, dir->i_ino, dir->i_generation,
5441                                        ent->oied_namelen, ent->oied_name,
5442                                        PFID(fid), rc);
5443                 } else {
5444                         /* Do not repair under dryrun mode. */
5445                         if (*attr & LUDA_VERIFY_DRYRUN) {
5446                                 *fid = lma->lma_self_fid;
5447                                 *attr |= LUDA_REPAIR;
5448                                 GOTO(out_inode, rc = 0);
5449                         }
5450
5451                         if (!dev->od_dirent_journal) {
5452                                 iput(inode);
5453                                 brelse(bh);
5454                                 if (hlock != NULL)
5455                                         ldiskfs_htree_unlock(hlock);
5456                                 else
5457                                         up_read(&obj->oo_ext_idx_sem);
5458                                 dev->od_dirent_journal = 1;
5459                                 goto again;
5460                         }
5461
5462                         *fid = lma->lma_self_fid;
5463                         dirty = true;
5464                         /* Append the FID-in-dirent. */
5465                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5466                                                  fid, bh, de, hlock);
5467                         if (rc == 0)
5468                                 *attr |= LUDA_REPAIR;
5469                         else
5470                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5471                                        "after the dirent, dir = %lu/%u, "
5472                                        "name = %.*s, "DFID": rc = %d\n",
5473                                        devname, dir->i_ino, dir->i_generation,
5474                                        ent->oied_namelen, ent->oied_name,
5475                                        PFID(fid), rc);
5476                 }
5477         } else if (rc == -ENODATA) {
5478                 /* Do not repair under dryrun mode. */
5479                 if (*attr & LUDA_VERIFY_DRYRUN) {
5480                         if (fid_is_sane(fid)) {
5481                                 *attr |= LUDA_REPAIR;
5482                         } else {
5483                                 lu_igif_build(fid, inode->i_ino,
5484                                               inode->i_generation);
5485                                 *attr |= LUDA_UPGRADE;
5486                         }
5487                         GOTO(out_inode, rc = 0);
5488                 }
5489
5490                 if (!dev->od_dirent_journal) {
5491                         iput(inode);
5492                         brelse(bh);
5493                         if (hlock != NULL)
5494                                 ldiskfs_htree_unlock(hlock);
5495                         else
5496                                 up_read(&obj->oo_ext_idx_sem);
5497                         dev->od_dirent_journal = 1;
5498                         goto again;
5499                 }
5500
5501                 dirty = true;
5502                 if (unlikely(fid_is_sane(fid))) {
5503                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5504                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5505                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5506                         if (rc == 0)
5507                                 *attr |= LUDA_REPAIR;
5508                         else
5509                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5510                                        "update dirent, dir = %lu/%u, "
5511                                        "name = %.*s, "DFID": rc = %d\n",
5512                                        devname, dir->i_ino, dir->i_generation,
5513                                        ent->oied_namelen, ent->oied_name,
5514                                        PFID(fid), rc);
5515                 } else {
5516                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5517                         /* It is probably IGIF object. Only aappend the
5518                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5519                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5520                                                  fid, bh, de, hlock);
5521                         if (rc == 0)
5522                                 *attr |= LUDA_UPGRADE;
5523                         else
5524                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5525                                        "after the dirent, dir = %lu/%u, "
5526                                        "name = %.*s, "DFID": rc = %d\n",
5527                                        devname, dir->i_ino, dir->i_generation,
5528                                        ent->oied_namelen, ent->oied_name,
5529                                        PFID(fid), rc);
5530                 }
5531         }
5532
5533         GOTO(out_inode, rc);
5534
5535 out_inode:
5536         iput(inode);
5537
5538 out_journal:
5539         brelse(bh);
5540         if (hlock != NULL) {
5541                 ldiskfs_htree_unlock(hlock);
5542         } else {
5543                 if (dev->od_dirent_journal)
5544                         up_write(&obj->oo_ext_idx_sem);
5545                 else
5546                         up_read(&obj->oo_ext_idx_sem);
5547         }
5548         if (jh != NULL)
5549                 ldiskfs_journal_stop(jh);
5550         if (rc >= 0 && !dirty)
5551                 dev->od_dirent_journal = 0;
5552         return rc;
5553 }
5554
5555 /**
5556  * Returns the value at current position from iterator's in memory structure.
5557  *
5558  * \param di struct osd_it_ea, iterator's in memory structure
5559  * \param attr attr requested for dirent.
5560  * \param lde lustre dirent
5561  *
5562  * \retval   0 no error and \param lde has correct lustre dirent.
5563  * \retval -ve on error
5564  */
5565 static inline int osd_it_ea_rec(const struct lu_env *env,
5566                                 const struct dt_it *di,
5567                                 struct dt_rec *dtrec, __u32 attr)
5568 {
5569         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5570         struct osd_object      *obj   = it->oie_obj;
5571         struct osd_device      *dev   = osd_obj2dev(obj);
5572         struct osd_scrub       *scrub = &dev->od_scrub;
5573         struct scrub_file      *sf    = &scrub->os_file;
5574         struct osd_thread_info *oti   = osd_oti_get(env);
5575         struct osd_inode_id    *id    = &oti->oti_id;
5576         struct osd_idmap_cache *oic   = &oti->oti_cache;
5577         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5578         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5579         __u32                   ino   = it->oie_dirent->oied_ino;
5580         int                     rc    = 0;
5581         ENTRY;
5582
5583         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5584
5585         if (attr & LUDA_VERIFY) {
5586                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5587                         attr |= LUDA_IGNORE;
5588                         /* If the parent is on remote MDT, and there
5589                          * is no FID-in-dirent, then we have to get
5590                          * the parent FID from the linkEA.  */
5591                         osd_get_pfid_from_linkea(env, obj, fid);
5592                 } else {
5593                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5594                                                      &attr);
5595                 }
5596         } else {
5597                 attr &= ~LU_DIRENT_ATTRS_MASK;
5598                 if (!fid_is_sane(fid)) {
5599                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5600                             likely(it->oie_dirent->oied_namelen != 2 ||
5601                                    it->oie_dirent->oied_name[0] != '.' ||
5602                                    it->oie_dirent->oied_name[1] != '.'))
5603                                 RETURN(-ENOENT);
5604
5605                         if (unlikely(ino == osd_remote_parent_ino(dev)))
5606                                 /* If the parent is on remote MDT, and there
5607                                  * is no FID-in-dirent, then we have to get
5608                                  * the parent FID from the linkEA.  */
5609                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5610                         else
5611                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5612                 } else {
5613                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5614                 }
5615         }
5616
5617         /* Pack the entry anyway, at least the offset is right. */
5618         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5619                            it->oie_dirent->oied_name,
5620                            it->oie_dirent->oied_namelen,
5621                            it->oie_dirent->oied_type, attr);
5622
5623         if (rc < 0)
5624                 RETURN(rc);
5625
5626         if (osd_remote_fid(env, dev, fid))
5627                 RETURN(0);
5628
5629         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5630                 rc = osd_add_oi_cache(oti, dev, id, fid);
5631
5632         if (!(attr & LUDA_VERIFY) &&
5633             (scrub->os_pos_current <= ino) &&
5634             ((sf->sf_flags & SF_INCONSISTENT) ||
5635              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5636              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5637                 osd_consistency_check(oti, dev, oic);
5638
5639         RETURN(rc > 0 ? 0 : rc);
5640 }
5641
5642 /**
5643  * Returns the record size size at current position.
5644  *
5645  * This function will return record(lu_dirent) size in bytes.
5646  *
5647  * \param[in] env       execution environment
5648  * \param[in] di        iterator's in memory structure
5649  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5650  *                      calculate the lu_dirent size.
5651  *
5652  * \retval      record size(in bytes & in memory) of the current lu_dirent
5653  *              entry.
5654  */
5655 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5656                               __u32 attr)
5657 {
5658         struct osd_it_ea *it = (struct osd_it_ea *)di;
5659
5660         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5661 }
5662
5663 /**
5664  * Returns a cookie for current position of the iterator head, so that
5665  * user can use this cookie to load/start the iterator next time.
5666  *
5667  * \param di iterator's in memory structure
5668  *
5669  * \retval cookie for current position, on success
5670  */
5671 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5672 {
5673         struct osd_it_ea *it = (struct osd_it_ea *)di;
5674
5675         return it->oie_dirent->oied_off;
5676 }
5677
5678 /**
5679  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5680  * to load a directory entry at a time and stored it i inn,
5681  * in iterator's in-memory data structure.
5682  *
5683  * \param di struct osd_it_ea, iterator's in memory structure
5684  *
5685  * \retval +ve on success
5686  * \retval -ve on error
5687  */
5688 static int osd_it_ea_load(const struct lu_env *env,
5689                           const struct dt_it *di, __u64 hash)
5690 {
5691         struct osd_it_ea *it = (struct osd_it_ea *)di;
5692         int rc;
5693
5694         ENTRY;
5695         it->oie_file.f_pos = hash;
5696
5697         rc =  osd_ldiskfs_it_fill(env, di);
5698         if (rc > 0)
5699                 rc = -ENODATA;
5700
5701         if (rc == 0)
5702                 rc = +1;
5703
5704         RETURN(rc);
5705 }
5706
5707 /**
5708  * Index lookup function for interoperability mode (b11826).
5709  *
5710  * \param key,  key i.e. file name to be searched
5711  *
5712  * \retval +ve, on success
5713  * \retval -ve, on error
5714  */
5715 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5716                                struct dt_rec *rec, const struct dt_key *key,
5717                                struct lustre_capa *capa)
5718 {
5719         struct osd_object *obj = osd_dt_obj(dt);
5720         int rc = 0;
5721
5722         ENTRY;
5723
5724         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5725         LINVRNT(osd_invariant(obj));
5726
5727         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5728                 return -EACCES;
5729
5730         rc = osd_ea_lookup_rec(env, obj, rec, key);
5731         if (rc == 0)
5732                 rc = +1;
5733         RETURN(rc);
5734 }
5735
5736 /**
5737  * Index and Iterator operations for interoperability
5738  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5739  */
5740 static const struct dt_index_operations osd_index_ea_ops = {
5741         .dio_lookup         = osd_index_ea_lookup,
5742         .dio_declare_insert = osd_index_declare_ea_insert,
5743         .dio_insert         = osd_index_ea_insert,
5744         .dio_declare_delete = osd_index_declare_ea_delete,
5745         .dio_delete         = osd_index_ea_delete,
5746         .dio_it     = {
5747                 .init     = osd_it_ea_init,
5748                 .fini     = osd_it_ea_fini,
5749                 .get      = osd_it_ea_get,
5750                 .put      = osd_it_ea_put,
5751                 .next     = osd_it_ea_next,
5752                 .key      = osd_it_ea_key,
5753                 .key_size = osd_it_ea_key_size,
5754                 .rec      = osd_it_ea_rec,
5755                 .rec_size = osd_it_ea_rec_size,
5756                 .store    = osd_it_ea_store,
5757                 .load     = osd_it_ea_load
5758         }
5759 };
5760
5761 static void *osd_key_init(const struct lu_context *ctx,
5762                           struct lu_context_key *key)
5763 {
5764         struct osd_thread_info *info;
5765
5766         OBD_ALLOC_PTR(info);
5767         if (info == NULL)
5768                 return ERR_PTR(-ENOMEM);
5769
5770         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5771         if (info->oti_it_ea_buf == NULL)
5772                 goto out_free_info;
5773
5774         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5775
5776         info->oti_hlock = ldiskfs_htree_lock_alloc();
5777         if (info->oti_hlock == NULL)
5778                 goto out_free_ea;
5779
5780         return info;
5781
5782  out_free_ea:
5783         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5784  out_free_info:
5785         OBD_FREE_PTR(info);
5786         return ERR_PTR(-ENOMEM);
5787 }
5788
5789 static void osd_key_fini(const struct lu_context *ctx,
5790                          struct lu_context_key *key, void* data)
5791 {
5792         struct osd_thread_info *info = data;
5793
5794         if (info->oti_hlock != NULL)
5795                 ldiskfs_htree_lock_free(info->oti_hlock);
5796         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5797         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5798         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5799         lu_buf_free(&info->oti_big_buf);
5800         OBD_FREE_PTR(info);
5801 }
5802
5803 static void osd_key_exit(const struct lu_context *ctx,
5804                          struct lu_context_key *key, void *data)
5805 {
5806         struct osd_thread_info *info = data;
5807
5808         LASSERT(info->oti_r_locks == 0);
5809         LASSERT(info->oti_w_locks == 0);
5810         LASSERT(info->oti_txns    == 0);
5811 }
5812
5813 /* type constructor/destructor: osd_type_init, osd_type_fini */
5814 LU_TYPE_INIT_FINI(osd, &osd_key);
5815
5816 struct lu_context_key osd_key = {
5817         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5818         .lct_init = osd_key_init,
5819         .lct_fini = osd_key_fini,
5820         .lct_exit = osd_key_exit
5821 };
5822
5823
5824 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5825                            const char *name, struct lu_device *next)
5826 {
5827         struct osd_device *osd = osd_dev(d);
5828
5829         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5830             >= sizeof(osd->od_svname))
5831                 return -E2BIG;
5832         return osd_procfs_init(osd, name);
5833 }
5834
5835 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5836 {
5837         struct seq_server_site  *ss = osd_seq_site(osd);
5838         int                     rc;
5839         ENTRY;
5840
5841         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5842                 RETURN(0);
5843
5844         if (unlikely(ss == NULL))
5845                 RETURN(-ENODEV);
5846
5847         OBD_ALLOC_PTR(osd->od_cl_seq);
5848         if (osd->od_cl_seq == NULL)
5849                 RETURN(-ENOMEM);
5850
5851         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5852                              osd->od_svname, ss->ss_server_seq);
5853
5854         if (rc != 0) {
5855                 OBD_FREE_PTR(osd->od_cl_seq);
5856                 osd->od_cl_seq = NULL;
5857         }
5858
5859         RETURN(rc);
5860 }
5861
5862 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5863 {
5864         if (osd->od_cl_seq == NULL)
5865                 return;
5866
5867         seq_client_fini(osd->od_cl_seq);
5868         OBD_FREE_PTR(osd->od_cl_seq);
5869         osd->od_cl_seq = NULL;
5870 }
5871
5872 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5873 {
5874         ENTRY;
5875
5876         /* shutdown quota slave instance associated with the device */
5877         if (o->od_quota_slave != NULL) {
5878                 qsd_fini(env, o->od_quota_slave);
5879                 o->od_quota_slave = NULL;
5880         }
5881
5882         osd_fid_fini(env, o);
5883
5884         RETURN(0);
5885 }
5886
5887 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5888 {
5889         ENTRY;
5890
5891         if (o->od_mnt != NULL) {
5892                 shrink_dcache_sb(osd_sb(o));
5893                 osd_sync(env, &o->od_dt_dev);
5894
5895                 mntput(o->od_mnt);
5896                 o->od_mnt = NULL;
5897         }
5898
5899         EXIT;
5900 }
5901
5902 static int osd_mount(const struct lu_env *env,
5903                      struct osd_device *o, struct lustre_cfg *cfg)
5904 {
5905         const char              *name  = lustre_cfg_string(cfg, 0);
5906         const char              *dev  = lustre_cfg_string(cfg, 1);
5907         const char              *opts;
5908         unsigned long            page, s_flags, lmd_flags = 0;
5909         struct page             *__page;
5910         struct file_system_type *type;
5911         char                    *options = NULL;
5912         char                    *str;
5913         struct osd_thread_info  *info = osd_oti_get(env);
5914         struct lu_fid           *fid = &info->oti_fid;
5915         struct inode            *inode;
5916         int                      rc = 0;
5917         ENTRY;
5918
5919         if (o->od_mnt != NULL)
5920                 RETURN(0);
5921
5922         if (strlen(dev) >= sizeof(o->od_mntdev))
5923                 RETURN(-E2BIG);
5924         strcpy(o->od_mntdev, dev);
5925
5926         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5927         if (__page == NULL)
5928                 GOTO(out, rc = -ENOMEM);
5929
5930         str = lustre_cfg_string(cfg, 2);
5931         s_flags = simple_strtoul(str, NULL, 0);
5932         str = strstr(str, ":");
5933         if (str)
5934                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5935         opts = lustre_cfg_string(cfg, 3);
5936         page = (unsigned long)page_address(__page);
5937         options = (char *)page;
5938         *options = '\0';
5939         if (opts == NULL)
5940                 strcat(options, "user_xattr,acl");
5941         else
5942                 strcat(options, opts);
5943
5944         /* Glom up mount options */
5945         if (*options != '\0')
5946                 strcat(options, ",");
5947         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5948
5949         type = get_fs_type("ldiskfs");
5950         if (!type) {
5951                 CERROR("%s: cannot find ldiskfs module\n", name);
5952                 GOTO(out, rc = -ENODEV);
5953         }
5954
5955         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5956         module_put(type->owner);
5957
5958         if (IS_ERR(o->od_mnt)) {
5959                 rc = PTR_ERR(o->od_mnt);
5960                 o->od_mnt = NULL;
5961                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5962                 GOTO(out, rc);
5963         }
5964
5965 #ifdef HAVE_DEV_SET_RDONLY
5966         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5967                 CERROR("%s: underlying device %s is marked as read-only. "
5968                        "Setup failed\n", name, dev);
5969                 GOTO(out_mnt, rc = -EROFS);
5970         }
5971 #endif
5972
5973         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5974                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5975                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5976                 GOTO(out_mnt, rc = -EINVAL);
5977         }
5978
5979 #ifdef LDISKFS_MOUNT_DIRDATA
5980         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5981                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5982                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5983         else if (!o->od_is_ost)
5984                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5985                       "enabling the dirdata feature. If you do not want to "
5986                       "downgrade to Lustre-1.x again, you can enable it via "
5987                       "'tune2fs -O dirdata device'\n", name, dev);
5988 #endif
5989         inode = osd_sb(o)->s_root->d_inode;
5990         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5991         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5992         if (rc != 0) {
5993                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5994                 GOTO(out_mnt, rc);
5995         }
5996
5997         if (lmd_flags & LMD_FLG_NOSCRUB)
5998                 o->od_noscrub = 1;
5999
6000         GOTO(out, rc = 0);
6001
6002 out_mnt:
6003         mntput(o->od_mnt);
6004         o->od_mnt = NULL;
6005
6006 out:
6007         if (__page)
6008                 OBD_PAGE_FREE(__page);
6009
6010         return rc;
6011 }
6012
6013 static struct lu_device *osd_device_fini(const struct lu_env *env,
6014                                          struct lu_device *d)
6015 {
6016         struct osd_device *o = osd_dev(d);
6017         ENTRY;
6018
6019         osd_shutdown(env, o);
6020         osd_procfs_fini(o);
6021         osd_scrub_cleanup(env, o);
6022         osd_obj_map_fini(o);
6023         osd_umount(env, o);
6024
6025         RETURN(NULL);
6026 }
6027
6028 static int osd_device_init0(const struct lu_env *env,
6029                             struct osd_device *o,
6030                             struct lustre_cfg *cfg)
6031 {
6032         struct lu_device        *l = osd2lu_dev(o);
6033         struct osd_thread_info *info;
6034         int                     rc;
6035         int                     cplen = 0;
6036
6037         /* if the module was re-loaded, env can loose its keys */
6038         rc = lu_env_refill((struct lu_env *) env);
6039         if (rc)
6040                 GOTO(out, rc);
6041         info = osd_oti_get(env);
6042         LASSERT(info);
6043
6044         l->ld_ops = &osd_lu_ops;
6045         o->od_dt_dev.dd_ops = &osd_dt_ops;
6046
6047         spin_lock_init(&o->od_osfs_lock);
6048         mutex_init(&o->od_otable_mutex);
6049
6050         o->od_capa_hash = init_capa_hash();
6051         if (o->od_capa_hash == NULL)
6052                 GOTO(out, rc = -ENOMEM);
6053
6054         o->od_read_cache = 1;
6055         o->od_writethrough_cache = 1;
6056         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6057
6058         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6059                         sizeof(o->od_svname));
6060         if (cplen >= sizeof(o->od_svname)) {
6061                 rc = -E2BIG;
6062                 GOTO(out_capa, rc);
6063         }
6064
6065         if (server_name_is_ost(o->od_svname))
6066                 o->od_is_ost = 1;
6067
6068         rc = osd_mount(env, o, cfg);
6069         if (rc != 0)
6070                 GOTO(out_capa, rc);
6071
6072         rc = osd_obj_map_init(env, o);
6073         if (rc != 0)
6074                 GOTO(out_mnt, rc);
6075
6076         rc = lu_site_init(&o->od_site, l);
6077         if (rc != 0)
6078                 GOTO(out_compat, rc);
6079         o->od_site.ls_bottom_dev = l;
6080
6081         rc = lu_site_init_finish(&o->od_site);
6082         if (rc != 0)
6083                 GOTO(out_site, rc);
6084
6085         /* self-repair LMA by default */
6086         o->od_lma_self_repair = 1;
6087
6088         INIT_LIST_HEAD(&o->od_ios_list);
6089         /* setup scrub, including OI files initialization */
6090         rc = osd_scrub_setup(env, o);
6091         if (rc < 0)
6092                 GOTO(out_site, rc);
6093
6094         rc = osd_procfs_init(o, o->od_svname);
6095         if (rc != 0) {
6096                 CERROR("%s: can't initialize procfs: rc = %d\n",
6097                        o->od_svname, rc);
6098                 GOTO(out_scrub, rc);
6099         }
6100
6101         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
6102
6103         /* initialize quota slave instance */
6104         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6105                                      o->od_proc_entry);
6106         if (IS_ERR(o->od_quota_slave)) {
6107                 rc = PTR_ERR(o->od_quota_slave);
6108                 o->od_quota_slave = NULL;
6109                 GOTO(out_procfs, rc);
6110         }
6111
6112         RETURN(0);
6113
6114 out_procfs:
6115         osd_procfs_fini(o);
6116 out_scrub:
6117         osd_scrub_cleanup(env, o);
6118 out_site:
6119         lu_site_fini(&o->od_site);
6120 out_compat:
6121         osd_obj_map_fini(o);
6122 out_mnt:
6123         osd_umount(env, o);
6124 out_capa:
6125         cleanup_capa_hash(o->od_capa_hash);
6126 out:
6127         return rc;
6128 }
6129
6130 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6131                                           struct lu_device_type *t,
6132                                           struct lustre_cfg *cfg)
6133 {
6134         struct osd_device *o;
6135         int                rc;
6136
6137         OBD_ALLOC_PTR(o);
6138         if (o == NULL)
6139                 return ERR_PTR(-ENOMEM);
6140
6141         rc = dt_device_init(&o->od_dt_dev, t);
6142         if (rc == 0) {
6143                 /* Because the ctx might be revived in dt_device_init,
6144                  * refill the env here */
6145                 lu_env_refill((struct lu_env *)env);
6146                 rc = osd_device_init0(env, o, cfg);
6147                 if (rc)
6148                         dt_device_fini(&o->od_dt_dev);
6149         }
6150
6151         if (unlikely(rc != 0))
6152                 OBD_FREE_PTR(o);
6153
6154         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6155 }
6156
6157 static struct lu_device *osd_device_free(const struct lu_env *env,
6158                                          struct lu_device *d)
6159 {
6160         struct osd_device *o = osd_dev(d);
6161         ENTRY;
6162
6163         cleanup_capa_hash(o->od_capa_hash);
6164         /* XXX: make osd top device in order to release reference */
6165         d->ld_site->ls_top_dev = d;
6166         lu_site_purge(env, d->ld_site, -1);
6167         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6168                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6169                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6170         }
6171         lu_site_fini(&o->od_site);
6172         dt_device_fini(&o->od_dt_dev);
6173         OBD_FREE_PTR(o);
6174         RETURN(NULL);
6175 }
6176
6177 static int osd_process_config(const struct lu_env *env,
6178                               struct lu_device *d, struct lustre_cfg *cfg)
6179 {
6180         struct osd_device               *o = osd_dev(d);
6181         int                             rc;
6182         ENTRY;
6183
6184         switch (cfg->lcfg_command) {
6185         case LCFG_SETUP:
6186                 rc = osd_mount(env, o, cfg);
6187                 break;
6188         case LCFG_CLEANUP:
6189                 lu_dev_del_linkage(d->ld_site, d);
6190                 rc = osd_shutdown(env, o);
6191                 break;
6192         case LCFG_PARAM:
6193                 LASSERT(&o->od_dt_dev);
6194                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6195                                               cfg, &o->od_dt_dev);
6196                 if (rc > 0 || rc == -ENOSYS)
6197                         rc = class_process_proc_param(PARAM_OST,
6198                                                       lprocfs_osd_obd_vars,
6199                                                       cfg, &o->od_dt_dev);
6200                 break;
6201         default:
6202                 rc = -ENOSYS;
6203         }
6204
6205         RETURN(rc);
6206 }
6207
6208 static int osd_recovery_complete(const struct lu_env *env,
6209                                  struct lu_device *d)
6210 {
6211         struct osd_device       *osd = osd_dev(d);
6212         int                      rc = 0;
6213         ENTRY;
6214
6215         if (osd->od_quota_slave == NULL)
6216                 RETURN(0);
6217
6218         /* start qsd instance on recovery completion, this notifies the quota
6219          * slave code that we are about to process new requests now */
6220         rc = qsd_start(env, osd->od_quota_slave);
6221         RETURN(rc);
6222 }
6223
6224 /*
6225  * we use exports to track all osd users
6226  */
6227 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6228                            struct obd_device *obd, struct obd_uuid *cluuid,
6229                            struct obd_connect_data *data, void *localdata)
6230 {
6231         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6232         struct lustre_handle  conn;
6233         int                   rc;
6234         ENTRY;
6235
6236         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6237
6238         rc = class_connect(&conn, obd, cluuid);
6239         if (rc)
6240                 RETURN(rc);
6241
6242         *exp = class_conn2export(&conn);
6243
6244         spin_lock(&osd->od_osfs_lock);
6245         osd->od_connects++;
6246         spin_unlock(&osd->od_osfs_lock);
6247
6248         RETURN(0);
6249 }
6250
6251 /*
6252  * once last export (we don't count self-export) disappeared
6253  * osd can be released
6254  */
6255 static int osd_obd_disconnect(struct obd_export *exp)
6256 {
6257         struct obd_device *obd = exp->exp_obd;
6258         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6259         int                rc, release = 0;
6260         ENTRY;
6261
6262         /* Only disconnect the underlying layers on the final disconnect. */
6263         spin_lock(&osd->od_osfs_lock);
6264         osd->od_connects--;
6265         if (osd->od_connects == 0)
6266                 release = 1;
6267         spin_unlock(&osd->od_osfs_lock);
6268
6269         rc = class_disconnect(exp); /* bz 9811 */
6270
6271         if (rc == 0 && release)
6272                 class_manual_cleanup(obd);
6273         RETURN(rc);
6274 }
6275
6276 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6277                        struct lu_device *dev)
6278 {
6279         struct osd_device *osd = osd_dev(dev);
6280         int                result = 0;
6281         ENTRY;
6282
6283         if (osd->od_quota_slave != NULL) {
6284                 /* set up quota slave objects */
6285                 result = qsd_prepare(env, osd->od_quota_slave);
6286                 if (result != 0)
6287                         RETURN(result);
6288         }
6289
6290         result = osd_fid_init(env, osd);
6291
6292         RETURN(result);
6293 }
6294
6295 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6296                   struct lu_fid *fid, struct md_op_data *op_data)
6297 {
6298         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6299
6300         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6301 }
6302
6303 static const struct lu_object_operations osd_lu_obj_ops = {
6304         .loo_object_init      = osd_object_init,
6305         .loo_object_delete    = osd_object_delete,
6306         .loo_object_release   = osd_object_release,
6307         .loo_object_free      = osd_object_free,
6308         .loo_object_print     = osd_object_print,
6309         .loo_object_invariant = osd_object_invariant
6310 };
6311
6312 const struct lu_device_operations osd_lu_ops = {
6313         .ldo_object_alloc      = osd_object_alloc,
6314         .ldo_process_config    = osd_process_config,
6315         .ldo_recovery_complete = osd_recovery_complete,
6316         .ldo_prepare           = osd_prepare,
6317 };
6318
6319 static const struct lu_device_type_operations osd_device_type_ops = {
6320         .ldto_init = osd_type_init,
6321         .ldto_fini = osd_type_fini,
6322
6323         .ldto_start = osd_type_start,
6324         .ldto_stop  = osd_type_stop,
6325
6326         .ldto_device_alloc = osd_device_alloc,
6327         .ldto_device_free  = osd_device_free,
6328
6329         .ldto_device_init    = osd_device_init,
6330         .ldto_device_fini    = osd_device_fini
6331 };
6332
6333 struct lu_device_type osd_device_type = {
6334         .ldt_tags     = LU_DEVICE_DT,
6335         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6336         .ldt_ops      = &osd_device_type_ops,
6337         .ldt_ctx_tags = LCT_LOCAL,
6338 };
6339
6340 /*
6341  * lprocfs legacy support.
6342  */
6343 static struct obd_ops osd_obd_device_ops = {
6344         .o_owner = THIS_MODULE,
6345         .o_connect      = osd_obd_connect,
6346         .o_disconnect   = osd_obd_disconnect,
6347         .o_fid_alloc    = osd_fid_alloc,
6348 };
6349
6350 static int __init osd_mod_init(void)
6351 {
6352         int rc;
6353
6354         osd_oi_mod_init();
6355
6356         rc = lu_kmem_init(ldiskfs_caches);
6357         if (rc)
6358                 return rc;
6359
6360         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6361                                  lprocfs_osd_module_vars,
6362                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6363         if (rc)
6364                 lu_kmem_fini(ldiskfs_caches);
6365         return rc;
6366 }
6367
6368 static void __exit osd_mod_exit(void)
6369 {
6370         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6371         lu_kmem_fini(ldiskfs_caches);
6372 }
6373
6374 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6375 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6376 MODULE_LICENSE("GPL");
6377
6378 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);