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 + 60 * HZ)) {
1034                         libcfs_debug_dumpstack(NULL);
1035                         last_credits = oh->ot_credits;
1036                         last_printed = jiffies;
1037                 }
1038                 /* XXX Limit the credits to 'max_transaction_buffers', and
1039                  *     let the underlying filesystem to catch the error if
1040                  *     we really need so many credits.
1041                  *
1042                  *     This should be removed when we can calculate the
1043                  *     credits precisely. */
1044                 oh->ot_credits = osd_transaction_size(dev);
1045         }
1046
1047         /*
1048          * XXX temporary stuff. Some abstraction layer should
1049          * be used.
1050          */
1051         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1052         osd_th_started(oh);
1053         if (!IS_ERR(jh)) {
1054                 oh->ot_handle = jh;
1055                 LASSERT(oti->oti_txns == 0);
1056                 lu_context_init(&th->th_ctx, th->th_tags);
1057                 lu_context_enter(&th->th_ctx);
1058
1059                 lu_device_get(&d->dd_lu_dev);
1060                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1061                               "osd-tx", th);
1062                 oti->oti_txns++;
1063                 rc = 0;
1064         } else {
1065                 rc = PTR_ERR(jh);
1066         }
1067 out:
1068         RETURN(rc);
1069 }
1070
1071 static int osd_seq_exists(const struct lu_env *env,
1072                               struct osd_device *osd, obd_seq seq)
1073 {
1074         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1075         struct seq_server_site  *ss = osd_seq_site(osd);
1076         int                     rc;
1077         ENTRY;
1078
1079         LASSERT(ss != NULL);
1080         LASSERT(ss->ss_server_fld != NULL);
1081
1082         rc = osd_fld_lookup(env, osd, seq, range);
1083         if (rc != 0) {
1084                 if (rc != -ENOENT)
1085                         CERROR("%s: can't lookup FLD sequence "LPX64
1086                                ": rc = %d\n", osd_name(osd), seq, rc);
1087                 RETURN(0);
1088         }
1089
1090         RETURN(ss->ss_node_id == range->lsr_index);
1091 }
1092
1093 /*
1094  * Concurrency: shouldn't matter.
1095  */
1096 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1097                           struct thandle *th)
1098 {
1099         int                     rc = 0;
1100         struct osd_thandle     *oh;
1101         struct osd_thread_info *oti = osd_oti_get(env);
1102         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1103         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1104         struct lquota_trans    *qtrans;
1105         ENTRY;
1106
1107         oh = container_of0(th, struct osd_thandle, ot_super);
1108
1109         qtrans = oh->ot_quota_trans;
1110         oh->ot_quota_trans = NULL;
1111
1112         if (oh->ot_handle != NULL) {
1113                 handle_t *hdl = oh->ot_handle;
1114
1115                 /*
1116                  * add commit callback
1117                  * notice we don't do this in osd_trans_start()
1118                  * as underlying transaction can change during truncate
1119                  */
1120                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1121                                          &oh->ot_jcb);
1122
1123                 LASSERT(oti->oti_txns == 1);
1124                 oti->oti_txns--;
1125                 rc = dt_txn_hook_stop(env, th);
1126                 if (rc != 0)
1127                         CERROR("Failure in transaction hook: %d\n", rc);
1128
1129                 /* hook functions might modify th_sync */
1130                 hdl->h_sync = th->th_sync;
1131
1132                 oh->ot_handle = NULL;
1133                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1134                                   rc = ldiskfs_journal_stop(hdl));
1135                 if (rc != 0)
1136                         CERROR("Failure to stop transaction: %d\n", rc);
1137         } else {
1138                 thandle_put(&oh->ot_super);
1139         }
1140
1141         /* inform the quota slave device that the transaction is stopping */
1142         qsd_op_end(env, qsd, qtrans);
1143
1144         /* as we want IO to journal and data IO be concurrent, we don't block
1145          * awaiting data IO completion in osd_do_bio(), instead we wait here
1146          * once transaction is submitted to the journal. all reqular requests
1147          * don't do direct IO (except read/write), thus this wait_event becomes
1148          * no-op for them.
1149          *
1150          * IMPORTANT: we have to wait till any IO submited by the thread is
1151          * completed otherwise iobuf may be corrupted by different request
1152          */
1153         wait_event(iobuf->dr_wait,
1154                        atomic_read(&iobuf->dr_numreqs) == 0);
1155         osd_fini_iobuf(oti->oti_dev, iobuf);
1156         if (!rc)
1157                 rc = iobuf->dr_error;
1158
1159         RETURN(rc);
1160 }
1161
1162 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1163 {
1164         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1165                                                ot_super);
1166
1167         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1168         LASSERT(&dcb->dcb_func != NULL);
1169         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1170
1171         return 0;
1172 }
1173
1174 /*
1175  * Called just before object is freed. Releases all resources except for
1176  * object itself (that is released by osd_object_free()).
1177  *
1178  * Concurrency: no concurrent access is possible that late in object
1179  * life-cycle.
1180  */
1181 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1182 {
1183         struct osd_object *obj   = osd_obj(l);
1184         struct inode      *inode = obj->oo_inode;
1185
1186         LINVRNT(osd_invariant(obj));
1187
1188         /*
1189          * If object is unlinked remove fid->ino mapping from object index.
1190          */
1191
1192         osd_index_fini(obj);
1193         if (inode != NULL) {
1194                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1195                 qid_t                    uid = i_uid_read(inode);
1196                 qid_t                    gid = i_gid_read(inode);
1197
1198                 iput(inode);
1199                 obj->oo_inode = NULL;
1200
1201                 if (qsd != NULL) {
1202                         struct osd_thread_info  *info = osd_oti_get(env);
1203                         struct lquota_id_info   *qi = &info->oti_qi;
1204
1205                         /* Release granted quota to master if necessary */
1206                         qi->lqi_id.qid_uid = uid;
1207                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1208
1209                         qi->lqi_id.qid_uid = gid;
1210                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1211                 }
1212         }
1213 }
1214
1215 /*
1216  * Concurrency: ->loo_object_release() is called under site spin-lock.
1217  */
1218 static void osd_object_release(const struct lu_env *env,
1219                                struct lu_object *l)
1220 {
1221 }
1222
1223 /*
1224  * Concurrency: shouldn't matter.
1225  */
1226 static int osd_object_print(const struct lu_env *env, void *cookie,
1227                             lu_printer_t p, const struct lu_object *l)
1228 {
1229         struct osd_object *o = osd_obj(l);
1230         struct iam_descr  *d;
1231
1232         if (o->oo_dir != NULL)
1233                 d = o->oo_dir->od_container.ic_descr;
1234         else
1235                 d = NULL;
1236         return (*p)(env, cookie,
1237                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1238                     o, o->oo_inode,
1239                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1240                     o->oo_inode ? o->oo_inode->i_generation : 0,
1241                     d ? d->id_ops->id_name : "plain");
1242 }
1243
1244 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1245
1246 /*
1247  * Concurrency: shouldn't matter.
1248  */
1249 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1250                struct obd_statfs *sfs)
1251 {
1252         struct osd_device  *osd = osd_dt_dev(d);
1253         struct super_block *sb = osd_sb(osd);
1254         struct kstatfs     *ksfs;
1255         int result = 0;
1256
1257         if (unlikely(osd->od_mnt == NULL))
1258                 return -EINPROGRESS;
1259
1260         /* osd_lproc.c call this without env, allocate ksfs for that case */
1261         if (unlikely(env == NULL)) {
1262                 OBD_ALLOC_PTR(ksfs);
1263                 if (ksfs == NULL)
1264                         return -ENOMEM;
1265         } else {
1266                 ksfs = &osd_oti_get(env)->oti_ksfs;
1267         }
1268
1269         spin_lock(&osd->od_osfs_lock);
1270         result = sb->s_op->statfs(sb->s_root, ksfs);
1271         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1272                 statfs_pack(sfs, ksfs);
1273                 if (sb->s_flags & MS_RDONLY)
1274                         sfs->os_state = OS_STATE_READONLY;
1275         }
1276
1277         spin_unlock(&osd->od_osfs_lock);
1278
1279         if (unlikely(env == NULL))
1280                 OBD_FREE_PTR(ksfs);
1281
1282         /* Reserve a small amount of space for local objects like last_rcvd,
1283          * llog, quota files, ... */
1284         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1285                 sfs->os_bavail = 0;
1286         } else {
1287                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1288                 /** Take out metadata overhead for indirect blocks */
1289                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1290         }
1291
1292         return result;
1293 }
1294
1295 /**
1296  * Estimate space needed for file creations. We assume the largest filename
1297  * which is 2^64 - 1, hence a filename of 20 chars.
1298  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1299  */
1300 #ifdef __LDISKFS_DIR_REC_LEN
1301 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1302 #else
1303 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1304 #endif
1305
1306 /*
1307  * Concurrency: doesn't access mutable data.
1308  */
1309 static void osd_conf_get(const struct lu_env *env,
1310                          const struct dt_device *dev,
1311                          struct dt_device_param *param)
1312 {
1313         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1314         int                ea_overhead;
1315
1316         /*
1317          * XXX should be taken from not-yet-existing fs abstraction layer.
1318          */
1319         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1320         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1321         param->ddp_block_shift  = sb->s_blocksize_bits;
1322         param->ddp_mount_type     = LDD_MT_LDISKFS;
1323         param->ddp_maxbytes       = sb->s_maxbytes;
1324         /* Overhead estimate should be fairly accurate, so we really take a tiny
1325          * error margin which also avoids fragmenting the filesystem too much */
1326         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1327         /* inode are statically allocated, so per-inode space consumption
1328          * is the space consumed by the directory entry */
1329         param->ddp_inodespace     = PER_OBJ_USAGE;
1330         /* per-fragment overhead to be used by the client code */
1331         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1332         param->ddp_mntopts      = 0;
1333         if (test_opt(sb, XATTR_USER))
1334                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1335         if (test_opt(sb, POSIX_ACL))
1336                 param->ddp_mntopts |= MNTOPT_ACL;
1337
1338         /* LOD might calculate the max stripe count based on max_ea_size,
1339          * so we need take account in the overhead as well,
1340          * xattr_header + magic + xattr_entry_head */
1341         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1342                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1343
1344 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1345         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1346                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1347                                                                 ea_overhead;
1348         else
1349 #endif
1350                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1351 }
1352
1353 /*
1354  * Concurrency: shouldn't matter.
1355  */
1356 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1357 {
1358         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1359         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1360 }
1361
1362 /**
1363  * Start commit for OSD device.
1364  *
1365  * An implementation of dt_commit_async method for OSD device.
1366  * Asychronously starts underlayng fs sync and thereby a transaction
1367  * commit.
1368  *
1369  * \param env environment
1370  * \param d dt device
1371  *
1372  * \see dt_device_operations
1373  */
1374 static int osd_commit_async(const struct lu_env *env,
1375                             struct dt_device *d)
1376 {
1377         struct super_block *s = osd_sb(osd_dt_dev(d));
1378         ENTRY;
1379
1380         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1381         RETURN(s->s_op->sync_fs(s, 0));
1382 }
1383
1384 /*
1385  * Concurrency: shouldn't matter.
1386  */
1387
1388 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1389 {
1390         struct super_block *sb = osd_sb(osd_dt_dev(d));
1391         struct block_device *dev = sb->s_bdev;
1392 #ifdef HAVE_DEV_SET_RDONLY
1393         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1394         int rc = 0;
1395 #else
1396         int rc = -EOPNOTSUPP;
1397 #endif
1398         ENTRY;
1399
1400 #ifdef HAVE_DEV_SET_RDONLY
1401         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1402
1403         if (jdev && (jdev != dev)) {
1404                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1405                        (long)jdev);
1406                 dev_set_rdonly(jdev);
1407         }
1408         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1409         dev_set_rdonly(dev);
1410 #else
1411         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1412                osd_dt_dev(d)->od_svname, (long)dev, rc);
1413 #endif
1414         RETURN(rc);
1415 }
1416
1417 /*
1418  * Concurrency: serialization provided by callers.
1419  */
1420 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1421                               int mode, unsigned long timeout, __u32 alg,
1422                               struct lustre_capa_key *keys)
1423 {
1424         struct osd_device *dev = osd_dt_dev(d);
1425         ENTRY;
1426
1427         dev->od_fl_capa = mode;
1428         dev->od_capa_timeout = timeout;
1429         dev->od_capa_alg = alg;
1430         dev->od_capa_keys = keys;
1431         RETURN(0);
1432 }
1433
1434 /**
1435  * Note: we do not count into QUOTA here.
1436  * If we mount with --data_journal we may need more.
1437  */
1438 const int osd_dto_credits_noquota[DTO_NR] = {
1439         /**
1440          * Insert.
1441          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1442          * SINGLEDATA_TRANS_BLOCKS(8)
1443          * XXX Note: maybe iam need more, since iam have more level than
1444          *           EXT3 htree.
1445          */
1446         [DTO_INDEX_INSERT]  = 16,
1447         /**
1448          * Delete
1449          * just modify a single entry, probably merge few within a block
1450          */
1451         [DTO_INDEX_DELETE]  = 1,
1452         /**
1453          * Used for OI scrub
1454          */
1455         [DTO_INDEX_UPDATE]  = 16,
1456         /**
1457          * 4(inode, inode bits, groups, GDT)
1458          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1459          */
1460         [DTO_OBJECT_CREATE] = 4,
1461         /**
1462          * 4(inode, inode bits, groups, GDT)
1463          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1464          */
1465         [DTO_OBJECT_DELETE] = 4,
1466         /**
1467          * Attr set credits (inode)
1468          */
1469         [DTO_ATTR_SET_BASE] = 1,
1470         /**
1471          * Xattr set. The same as xattr of EXT3.
1472          * DATA_TRANS_BLOCKS(14)
1473          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1474          * are also counted in. Do not know why?
1475          */
1476         [DTO_XATTR_SET]     = 14,
1477         /**
1478          * credits for inode change during write.
1479          */
1480         [DTO_WRITE_BASE]    = 3,
1481         /**
1482          * credits for single block write.
1483          */
1484         [DTO_WRITE_BLOCK]   = 14,
1485         /**
1486          * Attr set credits for chown.
1487          * This is extra credits for setattr, and it is null without quota
1488          */
1489         [DTO_ATTR_SET_CHOWN] = 0
1490 };
1491
1492 static const struct dt_device_operations osd_dt_ops = {
1493         .dt_root_get       = osd_root_get,
1494         .dt_statfs         = osd_statfs,
1495         .dt_trans_create   = osd_trans_create,
1496         .dt_trans_start    = osd_trans_start,
1497         .dt_trans_stop     = osd_trans_stop,
1498         .dt_trans_cb_add   = osd_trans_cb_add,
1499         .dt_conf_get       = osd_conf_get,
1500         .dt_sync           = osd_sync,
1501         .dt_ro             = osd_ro,
1502         .dt_commit_async   = osd_commit_async,
1503         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1504 };
1505
1506 static void osd_object_read_lock(const struct lu_env *env,
1507                                  struct dt_object *dt, unsigned role)
1508 {
1509         struct osd_object *obj = osd_dt_obj(dt);
1510         struct osd_thread_info *oti = osd_oti_get(env);
1511
1512         LINVRNT(osd_invariant(obj));
1513
1514         LASSERT(obj->oo_owner != env);
1515         down_read_nested(&obj->oo_sem, role);
1516
1517         LASSERT(obj->oo_owner == NULL);
1518         oti->oti_r_locks++;
1519 }
1520
1521 static void osd_object_write_lock(const struct lu_env *env,
1522                                   struct dt_object *dt, unsigned role)
1523 {
1524         struct osd_object *obj = osd_dt_obj(dt);
1525         struct osd_thread_info *oti = osd_oti_get(env);
1526
1527         LINVRNT(osd_invariant(obj));
1528
1529         LASSERT(obj->oo_owner != env);
1530         down_write_nested(&obj->oo_sem, role);
1531
1532         LASSERT(obj->oo_owner == NULL);
1533         obj->oo_owner = env;
1534         oti->oti_w_locks++;
1535 }
1536
1537 static void osd_object_read_unlock(const struct lu_env *env,
1538                                    struct dt_object *dt)
1539 {
1540         struct osd_object *obj = osd_dt_obj(dt);
1541         struct osd_thread_info *oti = osd_oti_get(env);
1542
1543         LINVRNT(osd_invariant(obj));
1544
1545         LASSERT(oti->oti_r_locks > 0);
1546         oti->oti_r_locks--;
1547         up_read(&obj->oo_sem);
1548 }
1549
1550 static void osd_object_write_unlock(const struct lu_env *env,
1551                                     struct dt_object *dt)
1552 {
1553         struct osd_object *obj = osd_dt_obj(dt);
1554         struct osd_thread_info *oti = osd_oti_get(env);
1555
1556         LINVRNT(osd_invariant(obj));
1557
1558         LASSERT(obj->oo_owner == env);
1559         LASSERT(oti->oti_w_locks > 0);
1560         oti->oti_w_locks--;
1561         obj->oo_owner = NULL;
1562         up_write(&obj->oo_sem);
1563 }
1564
1565 static int osd_object_write_locked(const struct lu_env *env,
1566                                    struct dt_object *dt)
1567 {
1568         struct osd_object *obj = osd_dt_obj(dt);
1569
1570         LINVRNT(osd_invariant(obj));
1571
1572         return obj->oo_owner == env;
1573 }
1574
1575 static int capa_is_sane(const struct lu_env *env,
1576                         struct osd_device *dev,
1577                         struct lustre_capa *capa,
1578                         struct lustre_capa_key *keys)
1579 {
1580         struct osd_thread_info *oti = osd_oti_get(env);
1581         struct lustre_capa *tcapa = &oti->oti_capa;
1582         struct obd_capa *oc;
1583         int i, rc = 0;
1584         ENTRY;
1585
1586         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1587         if (oc) {
1588                 if (capa_is_expired(oc)) {
1589                         DEBUG_CAPA(D_ERROR, capa, "expired");
1590                         rc = -ESTALE;
1591                 }
1592                 capa_put(oc);
1593                 RETURN(rc);
1594         }
1595
1596         if (capa_is_expired_sec(capa)) {
1597                 DEBUG_CAPA(D_ERROR, capa, "expired");
1598                 RETURN(-ESTALE);
1599         }
1600
1601         spin_lock(&capa_lock);
1602         for (i = 0; i < 2; i++) {
1603                 if (keys[i].lk_keyid == capa->lc_keyid) {
1604                         oti->oti_capa_key = keys[i];
1605                         break;
1606                 }
1607         }
1608         spin_unlock(&capa_lock);
1609
1610         if (i == 2) {
1611                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1612                 RETURN(-ESTALE);
1613         }
1614
1615         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1616         if (rc)
1617                 RETURN(rc);
1618
1619         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1620                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1621                 RETURN(-EACCES);
1622         }
1623
1624         oc = capa_add(dev->od_capa_hash, capa);
1625         capa_put(oc);
1626
1627         RETURN(0);
1628 }
1629
1630 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1631                     struct lustre_capa *capa, __u64 opc)
1632 {
1633         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1634         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1635         struct lu_capainfo *lci;
1636         int rc;
1637
1638         if (!osd->od_fl_capa)
1639                 return 0;
1640
1641         if (capa == BYPASS_CAPA)
1642                 return 0;
1643
1644         lci = lu_capainfo_get(env);
1645         if (unlikely(lci == NULL))
1646                 return 0;
1647
1648         if (lci->lci_auth == LC_ID_NONE)
1649                 return 0;
1650
1651         if (capa == NULL) {
1652                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1653                        osd_name(osd), PFID(fid), -EACCES);
1654                 return -EACCES;
1655         }
1656
1657         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1658                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1659                            PFID(fid));
1660                 return -EACCES;
1661         }
1662
1663         if (!capa_opc_supported(capa, opc)) {
1664                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1665                 return -EACCES;
1666         }
1667
1668         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1669         if (rc != 0) {
1670                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1671                 return -EACCES;
1672         }
1673
1674         return 0;
1675 }
1676
1677 static struct timespec *osd_inode_time(const struct lu_env *env,
1678                                        struct inode *inode, __u64 seconds)
1679 {
1680         struct osd_thread_info  *oti = osd_oti_get(env);
1681         struct timespec         *t   = &oti->oti_time;
1682
1683         t->tv_sec = seconds;
1684         t->tv_nsec = 0;
1685         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1686         return t;
1687 }
1688
1689
1690 static void osd_inode_getattr(const struct lu_env *env,
1691                               struct inode *inode, struct lu_attr *attr)
1692 {
1693         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1694                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1695                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1696                            LA_TYPE;
1697
1698         attr->la_atime   = LTIME_S(inode->i_atime);
1699         attr->la_mtime   = LTIME_S(inode->i_mtime);
1700         attr->la_ctime   = LTIME_S(inode->i_ctime);
1701         attr->la_mode    = inode->i_mode;
1702         attr->la_size    = i_size_read(inode);
1703         attr->la_blocks  = inode->i_blocks;
1704         attr->la_uid     = i_uid_read(inode);
1705         attr->la_gid     = i_gid_read(inode);
1706         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1707         attr->la_nlink   = inode->i_nlink;
1708         attr->la_rdev    = inode->i_rdev;
1709         attr->la_blksize = 1 << inode->i_blkbits;
1710         attr->la_blkbits = inode->i_blkbits;
1711 }
1712
1713 static int osd_attr_get(const struct lu_env *env,
1714                         struct dt_object *dt,
1715                         struct lu_attr *attr,
1716                         struct lustre_capa *capa)
1717 {
1718         struct osd_object *obj = osd_dt_obj(dt);
1719
1720         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1721         LINVRNT(osd_invariant(obj));
1722
1723         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1724                 return -EACCES;
1725
1726         spin_lock(&obj->oo_guard);
1727         osd_inode_getattr(env, obj->oo_inode, attr);
1728         spin_unlock(&obj->oo_guard);
1729         return 0;
1730 }
1731
1732 static int osd_declare_attr_set(const struct lu_env *env,
1733                                 struct dt_object *dt,
1734                                 const struct lu_attr *attr,
1735                                 struct thandle *handle)
1736 {
1737         struct osd_thandle     *oh;
1738         struct osd_object      *obj;
1739         struct osd_thread_info *info = osd_oti_get(env);
1740         struct lquota_id_info  *qi = &info->oti_qi;
1741         qid_t                   uid;
1742         qid_t                   gid;
1743         long long               bspace;
1744         int                     rc = 0;
1745         bool                    enforce;
1746         ENTRY;
1747
1748         LASSERT(dt != NULL);
1749         LASSERT(handle != NULL);
1750
1751         obj = osd_dt_obj(dt);
1752         LASSERT(osd_invariant(obj));
1753
1754         oh = container_of0(handle, struct osd_thandle, ot_super);
1755         LASSERT(oh->ot_handle == NULL);
1756
1757         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1758                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1759
1760         if (attr == NULL || obj->oo_inode == NULL)
1761                 RETURN(rc);
1762
1763         bspace   = obj->oo_inode->i_blocks;
1764         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1765         bspace   = toqb(bspace);
1766
1767         /* Changing ownership is always preformed by super user, it should not
1768          * fail with EDQUOT.
1769          *
1770          * We still need to call the osd_declare_qid() to calculate the journal
1771          * credits for updating quota accounting files and to trigger quota
1772          * space adjustment once the operation is completed.*/
1773         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1774                 /* USERQUOTA */
1775                 uid = i_uid_read(obj->oo_inode);
1776                 qi->lqi_type = USRQUOTA;
1777                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1778                 /* inode accounting */
1779                 qi->lqi_is_blk = false;
1780
1781                 /* one more inode for the new uid ... */
1782                 qi->lqi_id.qid_uid = attr->la_uid;
1783                 qi->lqi_space      = 1;
1784                 /* Reserve credits for the new uid */
1785                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1786                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1787                         rc = 0;
1788                 if (rc)
1789                         RETURN(rc);
1790
1791                 /* and one less inode for the current uid */
1792                 qi->lqi_id.qid_uid = uid;
1793                 qi->lqi_space      = -1;
1794                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1795                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1796                         rc = 0;
1797                 if (rc)
1798                         RETURN(rc);
1799
1800                 /* block accounting */
1801                 qi->lqi_is_blk = true;
1802
1803                 /* more blocks for the new uid ... */
1804                 qi->lqi_id.qid_uid = attr->la_uid;
1805                 qi->lqi_space      = bspace;
1806                 /*
1807                  * Credits for the new uid has been reserved, re-use "obj"
1808                  * to save credit reservation.
1809                  */
1810                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1811                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1812                         rc = 0;
1813                 if (rc)
1814                         RETURN(rc);
1815
1816                 /* and finally less blocks for the current uid */
1817                 qi->lqi_id.qid_uid = uid;
1818                 qi->lqi_space      = -bspace;
1819                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1820                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1821                         rc = 0;
1822                 if (rc)
1823                         RETURN(rc);
1824
1825                 /* GROUP QUOTA */
1826                 gid = i_gid_read(obj->oo_inode);
1827                 qi->lqi_type = GRPQUOTA;
1828                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1829
1830                 /* inode accounting */
1831                 qi->lqi_is_blk = false;
1832
1833                 /* one more inode for the new gid ... */
1834                 qi->lqi_id.qid_gid = attr->la_gid;
1835                 qi->lqi_space      = 1;
1836                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1837                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1838                         rc = 0;
1839                 if (rc)
1840                         RETURN(rc);
1841
1842                 /* and one less inode for the current gid */
1843                 qi->lqi_id.qid_gid = gid;
1844                 qi->lqi_space      = -1;
1845                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1846                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1847                         rc = 0;
1848                 if (rc)
1849                         RETURN(rc);
1850
1851                 /* block accounting */
1852                 qi->lqi_is_blk = true;
1853
1854                 /* more blocks for the new gid ... */
1855                 qi->lqi_id.qid_gid = attr->la_gid;
1856                 qi->lqi_space      = bspace;
1857                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1858                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1859                         rc = 0;
1860                 if (rc)
1861                         RETURN(rc);
1862
1863                 /* and finally less blocks for the current gid */
1864                 qi->lqi_id.qid_gid = gid;
1865                 qi->lqi_space      = -bspace;
1866                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1867                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1868                         rc = 0;
1869                 if (rc)
1870                         RETURN(rc);
1871         }
1872
1873         RETURN(rc);
1874 }
1875
1876 static int osd_inode_setattr(const struct lu_env *env,
1877                              struct inode *inode, const struct lu_attr *attr)
1878 {
1879         __u64 bits = attr->la_valid;
1880
1881         /* Only allow set size for regular file */
1882         if (!S_ISREG(inode->i_mode))
1883                 bits &= ~(LA_SIZE | LA_BLOCKS);
1884
1885         if (bits == 0)
1886                 return 0;
1887
1888         if (bits & LA_ATIME)
1889                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1890         if (bits & LA_CTIME)
1891                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1892         if (bits & LA_MTIME)
1893                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1894         if (bits & LA_SIZE) {
1895                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1896                 i_size_write(inode, attr->la_size);
1897         }
1898
1899 #if 0
1900         /* OSD should not change "i_blocks" which is used by quota.
1901          * "i_blocks" should be changed by ldiskfs only. */
1902         if (bits & LA_BLOCKS)
1903                 inode->i_blocks = attr->la_blocks;
1904 #endif
1905         if (bits & LA_MODE)
1906                 inode->i_mode = (inode->i_mode & S_IFMT) |
1907                                 (attr->la_mode & ~S_IFMT);
1908         if (bits & LA_UID)
1909                 i_uid_write(inode, attr->la_uid);
1910         if (bits & LA_GID)
1911                 i_gid_write(inode, attr->la_gid);
1912         if (bits & LA_NLINK)
1913                 set_nlink(inode, attr->la_nlink);
1914         if (bits & LA_RDEV)
1915                 inode->i_rdev = attr->la_rdev;
1916
1917         if (bits & LA_FLAGS) {
1918                 /* always keep S_NOCMTIME */
1919                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1920                                  S_NOCMTIME;
1921         }
1922         return 0;
1923 }
1924
1925 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1926 {
1927         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1928             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1929                 struct iattr    iattr;
1930                 int             rc;
1931
1932                 ll_vfs_dq_init(inode);
1933                 iattr.ia_valid = 0;
1934                 if (attr->la_valid & LA_UID)
1935                         iattr.ia_valid |= ATTR_UID;
1936                 if (attr->la_valid & LA_GID)
1937                         iattr.ia_valid |= ATTR_GID;
1938                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1939                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1940
1941                 rc = ll_vfs_dq_transfer(inode, &iattr);
1942                 if (rc) {
1943                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1944                                "enforcement enabled on the ldiskfs "
1945                                "filesystem?\n", inode->i_sb->s_id, rc);
1946                         return rc;
1947                 }
1948         }
1949         return 0;
1950 }
1951
1952 static int osd_attr_set(const struct lu_env *env,
1953                         struct dt_object *dt,
1954                         const struct lu_attr *attr,
1955                         struct thandle *handle,
1956                         struct lustre_capa *capa)
1957 {
1958         struct osd_object *obj = osd_dt_obj(dt);
1959         struct inode      *inode;
1960         int rc;
1961
1962         LASSERT(handle != NULL);
1963         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1964         LASSERT(osd_invariant(obj));
1965
1966         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1967                 return -EACCES;
1968
1969         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1970
1971         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1972                 struct osd_thread_info  *oti  = osd_oti_get(env);
1973                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1974                 struct lu_fid           *fid1 = &oti->oti_fid;
1975                 struct osd_inode_id     *id   = &oti->oti_id;
1976                 struct iam_path_descr   *ipd;
1977                 struct iam_container    *bag;
1978                 struct osd_thandle      *oh;
1979                 int                      rc;
1980
1981                 fid_cpu_to_be(fid1, fid0);
1982                 memset(id, 1, sizeof(*id));
1983                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1984                                   fid0)->oi_dir.od_container;
1985                 ipd = osd_idx_ipd_get(env, bag);
1986                 if (unlikely(ipd == NULL))
1987                         RETURN(-ENOMEM);
1988
1989                 oh = container_of0(handle, struct osd_thandle, ot_super);
1990                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1991                                 (const struct iam_rec *)id, ipd);
1992                 osd_ipd_put(env, bag, ipd);
1993                 return(rc > 0 ? 0 : rc);
1994         }
1995
1996         inode = obj->oo_inode;
1997
1998         rc = osd_quota_transfer(inode, attr);
1999         if (rc)
2000                 return rc;
2001
2002         spin_lock(&obj->oo_guard);
2003         rc = osd_inode_setattr(env, inode, attr);
2004         spin_unlock(&obj->oo_guard);
2005
2006         if (!rc)
2007                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2008         return rc;
2009 }
2010
2011 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2012                                     struct osd_object *obj,
2013                                     const char *name, const int namelen)
2014 {
2015         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2016 }
2017
2018 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2019                       umode_t mode, struct dt_allocation_hint *hint,
2020                       struct thandle *th)
2021 {
2022         int result;
2023         struct osd_device  *osd = osd_obj2dev(obj);
2024         struct osd_thandle *oth;
2025         struct dt_object   *parent = NULL;
2026         struct inode       *inode;
2027
2028         LINVRNT(osd_invariant(obj));
2029         LASSERT(obj->oo_inode == NULL);
2030         LASSERT(obj->oo_hl_head == NULL);
2031
2032         if (S_ISDIR(mode) && ldiskfs_pdo) {
2033                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2034                 if (obj->oo_hl_head == NULL)
2035                         return -ENOMEM;
2036         }
2037
2038         oth = container_of(th, struct osd_thandle, ot_super);
2039         LASSERT(oth->ot_handle->h_transaction != NULL);
2040
2041         if (hint && hint->dah_parent)
2042                 parent = hint->dah_parent;
2043
2044         inode = ldiskfs_create_inode(oth->ot_handle,
2045                                      parent ? osd_dt_obj(parent)->oo_inode :
2046                                               osd_sb(osd)->s_root->d_inode,
2047                                      mode);
2048         if (!IS_ERR(inode)) {
2049                 /* Do not update file c/mtime in ldiskfs.
2050                  * NB: don't need any lock because no contention at this
2051                  * early stage */
2052                 inode->i_flags |= S_NOCMTIME;
2053
2054                 /* For new created object, it must be consistent,
2055                  * and it is unnecessary to scrub against it. */
2056                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2057                 obj->oo_inode = inode;
2058                 result = 0;
2059         } else {
2060                 if (obj->oo_hl_head != NULL) {
2061                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2062                         obj->oo_hl_head = NULL;
2063                 }
2064                 result = PTR_ERR(inode);
2065         }
2066         LINVRNT(osd_invariant(obj));
2067         return result;
2068 }
2069
2070 enum {
2071         OSD_NAME_LEN = 255
2072 };
2073
2074 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2075                      struct lu_attr *attr,
2076                      struct dt_allocation_hint *hint,
2077                      struct dt_object_format *dof,
2078                      struct thandle *th)
2079 {
2080         int result;
2081         struct osd_thandle *oth;
2082         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2083
2084         LASSERT(S_ISDIR(attr->la_mode));
2085
2086         oth = container_of(th, struct osd_thandle, ot_super);
2087         LASSERT(oth->ot_handle->h_transaction != NULL);
2088         result = osd_mkfile(info, obj, mode, hint, th);
2089
2090         return result;
2091 }
2092
2093 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2094                         struct lu_attr *attr,
2095                         struct dt_allocation_hint *hint,
2096                         struct dt_object_format *dof,
2097                         struct thandle *th)
2098 {
2099         int result;
2100         struct osd_thandle *oth;
2101         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2102
2103         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2104
2105         LASSERT(S_ISREG(attr->la_mode));
2106
2107         oth = container_of(th, struct osd_thandle, ot_super);
2108         LASSERT(oth->ot_handle->h_transaction != NULL);
2109
2110         result = osd_mkfile(info, obj, mode, hint, th);
2111         if (result == 0) {
2112                 LASSERT(obj->oo_inode != NULL);
2113                 if (feat->dif_flags & DT_IND_VARKEY)
2114                         result = iam_lvar_create(obj->oo_inode,
2115                                                  feat->dif_keysize_max,
2116                                                  feat->dif_ptrsize,
2117                                                  feat->dif_recsize_max,
2118                                                  oth->ot_handle);
2119                 else
2120                         result = iam_lfix_create(obj->oo_inode,
2121                                                  feat->dif_keysize_max,
2122                                                  feat->dif_ptrsize,
2123                                                  feat->dif_recsize_max,
2124                                                  oth->ot_handle);
2125
2126         }
2127         return result;
2128 }
2129
2130 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2131                      struct lu_attr *attr,
2132                      struct dt_allocation_hint *hint,
2133                      struct dt_object_format *dof,
2134                      struct thandle *th)
2135 {
2136         LASSERT(S_ISREG(attr->la_mode));
2137         return osd_mkfile(info, obj, (attr->la_mode &
2138                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2139 }
2140
2141 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2142                      struct lu_attr *attr,
2143                      struct dt_allocation_hint *hint,
2144                      struct dt_object_format *dof,
2145                      struct thandle *th)
2146 {
2147         LASSERT(S_ISLNK(attr->la_mode));
2148         return osd_mkfile(info, obj, (attr->la_mode &
2149                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2150 }
2151
2152 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2153                      struct lu_attr *attr,
2154                      struct dt_allocation_hint *hint,
2155                      struct dt_object_format *dof,
2156                      struct thandle *th)
2157 {
2158         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2159         int result;
2160
2161         LINVRNT(osd_invariant(obj));
2162         LASSERT(obj->oo_inode == NULL);
2163         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2164                 S_ISFIFO(mode) || S_ISSOCK(mode));
2165
2166         result = osd_mkfile(info, obj, mode, hint, th);
2167         if (result == 0) {
2168                 LASSERT(obj->oo_inode != NULL);
2169                 /*
2170                  * This inode should be marked dirty for i_rdev.  Currently
2171                  * that is done in the osd_attr_init().
2172                  */
2173                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2174                                    attr->la_rdev);
2175         }
2176         LINVRNT(osd_invariant(obj));
2177         return result;
2178 }
2179
2180 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2181                               struct lu_attr *,
2182                               struct dt_allocation_hint *hint,
2183                               struct dt_object_format *dof,
2184                               struct thandle *);
2185
2186 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2187 {
2188         osd_obj_type_f result;
2189
2190         switch (type) {
2191         case DFT_DIR:
2192                 result = osd_mkdir;
2193                 break;
2194         case DFT_REGULAR:
2195                 result = osd_mkreg;
2196                 break;
2197         case DFT_SYM:
2198                 result = osd_mksym;
2199                 break;
2200         case DFT_NODE:
2201                 result = osd_mknod;
2202                 break;
2203         case DFT_INDEX:
2204                 result = osd_mk_index;
2205                 break;
2206
2207         default:
2208                 LBUG();
2209                 break;
2210         }
2211         return result;
2212 }
2213
2214
2215 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2216                         struct dt_object *parent, struct dt_object *child,
2217                         umode_t child_mode)
2218 {
2219         LASSERT(ah);
2220
2221         ah->dah_parent = parent;
2222         ah->dah_mode = child_mode;
2223 }
2224
2225 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2226                           struct lu_attr *attr, struct dt_object_format *dof)
2227 {
2228         struct inode   *inode = obj->oo_inode;
2229         __u64           valid = attr->la_valid;
2230         int             result;
2231
2232         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2233
2234         if (dof->dof_type != DFT_NODE)
2235                 attr->la_valid &= ~LA_RDEV;
2236         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2237                 attr->la_valid &= ~LA_ATIME;
2238         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2239                 attr->la_valid &= ~LA_CTIME;
2240         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2241                 attr->la_valid &= ~LA_MTIME;
2242
2243         result = osd_quota_transfer(inode, attr);
2244         if (result)
2245                 return;
2246
2247         if (attr->la_valid != 0) {
2248                 result = osd_inode_setattr(info->oti_env, inode, attr);
2249                 /*
2250                  * The osd_inode_setattr() should always succeed here.  The
2251                  * only error that could be returned is EDQUOT when we are
2252                  * trying to change the UID or GID of the inode. However, this
2253                  * should not happen since quota enforcement is no longer
2254                  * enabled on ldiskfs (lquota takes care of it).
2255                  */
2256                 LASSERTF(result == 0, "%d\n", result);
2257                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2258         }
2259
2260         attr->la_valid = valid;
2261 }
2262
2263 /**
2264  * Helper function for osd_object_create()
2265  *
2266  * \retval 0, on success
2267  */
2268 static int __osd_object_create(struct osd_thread_info *info,
2269                                struct osd_object *obj, struct lu_attr *attr,
2270                                struct dt_allocation_hint *hint,
2271                                struct dt_object_format *dof,
2272                                struct thandle *th)
2273 {
2274         int     result;
2275         __u32   umask;
2276
2277         /* we drop umask so that permissions we pass are not affected */
2278         umask = current->fs->umask;
2279         current->fs->umask = 0;
2280
2281         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2282                                                   th);
2283         if (result == 0) {
2284                 osd_attr_init(info, obj, attr, dof);
2285                 osd_object_init0(obj);
2286                 /* bz 24037 */
2287                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2288                         unlock_new_inode(obj->oo_inode);
2289         }
2290
2291         /* restore previous umask value */
2292         current->fs->umask = umask;
2293
2294         return result;
2295 }
2296
2297 /**
2298  * Helper function for osd_object_create()
2299  *
2300  * \retval 0, on success
2301  */
2302 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2303                            const struct lu_fid *fid, struct thandle *th)
2304 {
2305         struct osd_thread_info *info = osd_oti_get(env);
2306         struct osd_inode_id    *id   = &info->oti_id;
2307         struct osd_device      *osd  = osd_obj2dev(obj);
2308         struct osd_thandle     *oh;
2309
2310         LASSERT(obj->oo_inode != NULL);
2311
2312         oh = container_of0(th, struct osd_thandle, ot_super);
2313         LASSERT(oh->ot_handle);
2314
2315         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2316         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2317 }
2318
2319 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2320                    obd_seq seq, struct lu_seq_range *range)
2321 {
2322         struct seq_server_site  *ss = osd_seq_site(osd);
2323
2324         if (fid_seq_is_idif(seq)) {
2325                 fld_range_set_ost(range);
2326                 range->lsr_index = idif_ost_idx(seq);
2327                 return 0;
2328         }
2329
2330         if (!fid_seq_in_fldb(seq)) {
2331                 fld_range_set_mdt(range);
2332                 if (ss != NULL)
2333                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2334                          * at all */
2335                         range->lsr_index = ss->ss_node_id;
2336                 return 0;
2337         }
2338
2339         LASSERT(ss != NULL);
2340         fld_range_set_any(range);
2341         /* OSD will only do local fld lookup */
2342         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2343 }
2344
2345 /*
2346  * Concurrency: no external locking is necessary.
2347  */
2348 static int osd_declare_object_create(const struct lu_env *env,
2349                                      struct dt_object *dt,
2350                                      struct lu_attr *attr,
2351                                      struct dt_allocation_hint *hint,
2352                                      struct dt_object_format *dof,
2353                                      struct thandle *handle)
2354 {
2355         struct osd_thandle      *oh;
2356         int                      rc;
2357         ENTRY;
2358
2359         LASSERT(handle != NULL);
2360
2361         oh = container_of0(handle, struct osd_thandle, ot_super);
2362         LASSERT(oh->ot_handle == NULL);
2363
2364         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2365                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2366         /* Reuse idle OI block may cause additional one OI block
2367          * to be changed. */
2368         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2369                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2370
2371         /* If this is directory, then we expect . and .. to be inserted as
2372          * well. The one directory block always needs to be created for the
2373          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2374          * block), there is no danger of needing a tree for the first block.
2375          */
2376         if (attr && S_ISDIR(attr->la_mode)) {
2377                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2378                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2379                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2380         }
2381
2382         if (!attr)
2383                 RETURN(0);
2384
2385         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2386                                    osd_dt_obj(dt), false, NULL, false);
2387         if (rc != 0)
2388                 RETURN(rc);
2389
2390         RETURN(rc);
2391 }
2392
2393 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2394                              struct lu_attr *attr,
2395                              struct dt_allocation_hint *hint,
2396                              struct dt_object_format *dof,
2397                              struct thandle *th)
2398 {
2399         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2400         struct osd_object      *obj    = osd_dt_obj(dt);
2401         struct osd_thread_info *info   = osd_oti_get(env);
2402         int result;
2403
2404         ENTRY;
2405
2406         LINVRNT(osd_invariant(obj));
2407         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2408         LASSERT(osd_write_locked(env, obj));
2409         LASSERT(th != NULL);
2410
2411         if (unlikely(fid_is_acct(fid)))
2412                 /* Quota files can't be created from the kernel any more,
2413                  * 'tune2fs -O quota' will take care of creating them */
2414                 RETURN(-EPERM);
2415
2416         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2417         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2418
2419         result = __osd_object_create(info, obj, attr, hint, dof, th);
2420         if (result == 0)
2421                 result = __osd_oi_insert(env, obj, fid, th);
2422
2423         LASSERT(ergo(result == 0,
2424                      dt_object_exists(dt) && !dt_object_remote(dt)));
2425
2426         LASSERT(osd_invariant(obj));
2427         RETURN(result);
2428 }
2429
2430 /**
2431  * Called to destroy on-disk representation of the object
2432  *
2433  * Concurrency: must be locked
2434  */
2435 static int osd_declare_object_destroy(const struct lu_env *env,
2436                                       struct dt_object *dt,
2437                                       struct thandle *th)
2438 {
2439         struct osd_object  *obj = osd_dt_obj(dt);
2440         struct inode       *inode = obj->oo_inode;
2441         struct osd_thandle *oh;
2442         int                 rc;
2443         ENTRY;
2444
2445         oh = container_of0(th, struct osd_thandle, ot_super);
2446         LASSERT(oh->ot_handle == NULL);
2447         LASSERT(inode);
2448
2449         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2450                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2451         /* Recycle idle OI leaf may cause additional three OI blocks
2452          * to be changed. */
2453         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2454                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2455         /* one less inode */
2456         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2457                                    -1, oh, obj, false, NULL, false);
2458         if (rc)
2459                 RETURN(rc);
2460         /* data to be truncated */
2461         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2462                                    0, oh, obj, true, NULL, false);
2463         RETURN(rc);
2464 }
2465
2466 static int osd_object_destroy(const struct lu_env *env,
2467                               struct dt_object *dt,
2468                               struct thandle *th)
2469 {
2470         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2471         struct osd_object      *obj = osd_dt_obj(dt);
2472         struct inode           *inode = obj->oo_inode;
2473         struct osd_device      *osd = osd_obj2dev(obj);
2474         struct osd_thandle     *oh;
2475         int                     result;
2476         ENTRY;
2477
2478         oh = container_of0(th, struct osd_thandle, ot_super);
2479         LASSERT(oh->ot_handle);
2480         LASSERT(inode);
2481         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2482
2483         if (unlikely(fid_is_acct(fid)))
2484                 RETURN(-EPERM);
2485
2486         if (S_ISDIR(inode->i_mode)) {
2487                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2488                         inode->i_nlink == 2);
2489                 /* it will check/delete the inode from remote parent,
2490                  * how to optimize it? unlink performance impaction XXX */
2491                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2492                 if (result != 0 && result != -ENOENT) {
2493                         CERROR("%s: delete inode "DFID": rc = %d\n",
2494                                osd_name(osd), PFID(fid), result);
2495                 }
2496                 spin_lock(&obj->oo_guard);
2497                 clear_nlink(inode);
2498                 spin_unlock(&obj->oo_guard);
2499                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2500         }
2501
2502         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2503
2504         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2505                                OI_CHECK_FLD);
2506
2507         /* XXX: add to ext3 orphan list */
2508         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2509
2510         /* not needed in the cache anymore */
2511         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2512
2513         RETURN(0);
2514 }
2515
2516 /**
2517  * Put the fid into lustre_mdt_attrs, and then place the structure
2518  * inode's ea. This fid should not be altered during the life time
2519  * of the inode.
2520  *
2521  * \retval +ve, on success
2522  * \retval -ve, on error
2523  *
2524  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2525  */
2526 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2527                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2528 {
2529         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2530         int                      rc;
2531         ENTRY;
2532
2533         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2534                 RETURN(0);
2535
2536         lustre_lma_init(lma, fid, compat, incompat);
2537         lustre_lma_swab(lma);
2538
2539         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2540                              XATTR_CREATE);
2541         /* LMA may already exist, but we need to check that all the
2542          * desired compat/incompat flags have been added. */
2543         if (unlikely(rc == -EEXIST)) {
2544                 if (compat == 0 && incompat == 0)
2545                         RETURN(0);
2546
2547                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2548                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2549                                      LMA_OLD_SIZE);
2550                 if (rc <= 0)
2551                         RETURN(-EINVAL);
2552
2553                 lustre_lma_swab(lma);
2554                 if (!(~lma->lma_compat & compat) &&
2555                     !(~lma->lma_incompat & incompat))
2556                         RETURN(0);
2557
2558                 lma->lma_compat |= compat;
2559                 lma->lma_incompat |= incompat;
2560                 lustre_lma_swab(lma);
2561                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2562                                      sizeof(*lma), XATTR_REPLACE);
2563         }
2564
2565         RETURN(rc);
2566 }
2567
2568 /**
2569  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2570  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2571  * To have compatilibility with 1.8 ldiskfs driver we need to have
2572  * magic number at start of fid data.
2573  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2574  * its inmemory API.
2575  */
2576 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2577                                   const struct lu_fid *fid)
2578 {
2579         if (!fid_is_namespace_visible(fid) ||
2580             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2581                 param->edp_magic = 0;
2582                 return;
2583         }
2584
2585         param->edp_magic = LDISKFS_LUFID_MAGIC;
2586         param->edp_len =  sizeof(struct lu_fid) + 1;
2587         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2588 }
2589
2590 /**
2591  * Try to read the fid from inode ea into dt_rec.
2592  *
2593  * \param fid object fid.
2594  *
2595  * \retval 0 on success
2596  */
2597 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2598                           __u32 ino, struct lu_fid *fid,
2599                           struct osd_inode_id *id)
2600 {
2601         struct osd_thread_info *info  = osd_oti_get(env);
2602         struct inode           *inode;
2603         ENTRY;
2604
2605         osd_id_gen(id, ino, OSD_OII_NOGEN);
2606         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2607         if (IS_ERR(inode))
2608                 RETURN(PTR_ERR(inode));
2609
2610         iput(inode);
2611         RETURN(0);
2612 }
2613
2614 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2615                                         struct inode *dir,
2616                                         struct inode *parent_dir,
2617                                         const struct lu_fid *dot_fid,
2618                                         const struct lu_fid *dot_dot_fid,
2619                                         struct osd_thandle *oth)
2620 {
2621         struct ldiskfs_dentry_param *dot_ldp;
2622         struct ldiskfs_dentry_param *dot_dot_ldp;
2623
2624         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2625         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2626
2627         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2628         dot_ldp->edp_magic = 0;
2629         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2630                                         dir, dot_ldp, dot_dot_ldp);
2631 }
2632
2633 /**
2634  * Create an local agent inode for remote entry
2635  */
2636 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2637                                                   struct osd_device *osd,
2638                                                   struct osd_object *pobj,
2639                                                   const struct lu_fid *fid,
2640                                                   __u32 type,
2641                                                   struct thandle *th)
2642 {
2643         struct osd_thread_info  *info = osd_oti_get(env);
2644         struct inode            *local;
2645         struct osd_thandle      *oh;
2646         int                     rc;
2647         ENTRY;
2648
2649         LASSERT(th);
2650         oh = container_of(th, struct osd_thandle, ot_super);
2651         LASSERT(oh->ot_handle->h_transaction != NULL);
2652
2653         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2654         if (IS_ERR(local)) {
2655                 CERROR("%s: create local error %d\n", osd_name(osd),
2656                        (int)PTR_ERR(local));
2657                 RETURN(local);
2658         }
2659
2660         /* Set special LMA flag for local agent inode */
2661         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2662         if (rc != 0) {
2663                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2664                        osd_name(osd), PFID(fid), rc);
2665                 RETURN(ERR_PTR(rc));
2666         }
2667
2668         if (!S_ISDIR(type))
2669                 RETURN(local);
2670
2671         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2672                                          lu_object_fid(&pobj->oo_dt.do_lu),
2673                                          fid, oh);
2674         if (rc != 0) {
2675                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2676                         osd_name(osd), PFID(fid), rc);
2677                 RETURN(ERR_PTR(rc));
2678         }
2679
2680         RETURN(local);
2681 }
2682
2683 /**
2684  * Delete local agent inode for remote entry
2685  */
2686 static int osd_delete_local_agent_inode(const struct lu_env *env,
2687                                         struct osd_device *osd,
2688                                         const struct lu_fid *fid,
2689                                         __u32 ino, struct osd_thandle *oh)
2690 {
2691         struct osd_thread_info  *oti = osd_oti_get(env);
2692         struct osd_inode_id     *id = &oti->oti_id;
2693         struct inode            *inode;
2694         ENTRY;
2695
2696         id->oii_ino = le32_to_cpu(ino);
2697         id->oii_gen = OSD_OII_NOGEN;
2698         inode = osd_iget(oti, osd, id);
2699         if (IS_ERR(inode)) {
2700                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2701                        PFID(fid), id->oii_ino, id->oii_gen);
2702                 RETURN(PTR_ERR(inode));
2703         }
2704
2705         clear_nlink(inode);
2706         mark_inode_dirty(inode);
2707         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2708                 osd_name(osd), PFID(fid), inode->i_ino);
2709         iput(inode);
2710         RETURN(0);
2711 }
2712
2713 /**
2714  * OSD layer object create function for interoperability mode (b11826).
2715  * This is mostly similar to osd_object_create(). Only difference being, fid is
2716  * inserted into inode ea here.
2717  *
2718  * \retval   0, on success
2719  * \retval -ve, on error
2720  */
2721 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2722                                 struct lu_attr *attr,
2723                                 struct dt_allocation_hint *hint,
2724                                 struct dt_object_format *dof,
2725                                 struct thandle *th)
2726 {
2727         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2728         struct osd_object      *obj    = osd_dt_obj(dt);
2729         struct osd_thread_info *info   = osd_oti_get(env);
2730         int                     result;
2731
2732         ENTRY;
2733
2734         LASSERT(osd_invariant(obj));
2735         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2736         LASSERT(osd_write_locked(env, obj));
2737         LASSERT(th != NULL);
2738
2739         if (unlikely(fid_is_acct(fid)))
2740                 /* Quota files can't be created from the kernel any more,
2741                  * 'tune2fs -O quota' will take care of creating them */
2742                 RETURN(-EPERM);
2743
2744         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2745         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2746
2747         result = __osd_object_create(info, obj, attr, hint, dof, th);
2748         if (result == 0)
2749                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2750                                 fid_is_on_ost(info, osd_obj2dev(obj),
2751                                               fid, OI_CHECK_FLD) ?
2752                                 LMAC_FID_ON_OST : 0, 0);
2753
2754         if (result == 0)
2755                 result = __osd_oi_insert(env, obj, fid, th);
2756
2757         LASSERT(ergo(result == 0,
2758                      dt_object_exists(dt) && !dt_object_remote(dt)));
2759         LINVRNT(osd_invariant(obj));
2760         RETURN(result);
2761 }
2762
2763 static int osd_declare_object_ref_add(const struct lu_env *env,
2764                                       struct dt_object *dt,
2765                                       struct thandle *handle)
2766 {
2767         struct osd_thandle       *oh;
2768
2769         /* it's possible that object doesn't exist yet */
2770         LASSERT(handle != NULL);
2771
2772         oh = container_of0(handle, struct osd_thandle, ot_super);
2773         LASSERT(oh->ot_handle == NULL);
2774
2775         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2776                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2777
2778         return 0;
2779 }
2780
2781 /*
2782  * Concurrency: @dt is write locked.
2783  */
2784 static int osd_object_ref_add(const struct lu_env *env,
2785                               struct dt_object *dt, struct thandle *th)
2786 {
2787         struct osd_object  *obj = osd_dt_obj(dt);
2788         struct inode       *inode = obj->oo_inode;
2789         struct osd_thandle *oh;
2790         int                 rc = 0;
2791
2792         LINVRNT(osd_invariant(obj));
2793         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2794         LASSERT(osd_write_locked(env, obj));
2795         LASSERT(th != NULL);
2796
2797         oh = container_of0(th, struct osd_thandle, ot_super);
2798         LASSERT(oh->ot_handle != NULL);
2799
2800         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2801
2802         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2803                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2804         /*
2805          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2806          * (65000) subdirectories by storing "1" in i_nlink if the link count
2807          * would otherwise overflow. Directory tranversal tools understand
2808          * that (st_nlink == 1) indicates that the filesystem dose not track
2809          * hard links count on the directory, and will not abort subdirectory
2810          * scanning early once (st_nlink - 2) subdirs have been found.
2811          *
2812          * This also has to properly handle the case of inodes with nlink == 0
2813          * in case they are being linked into the PENDING directory
2814          */
2815         spin_lock(&obj->oo_guard);
2816         if (unlikely(inode->i_nlink == 0))
2817                 /* inc_nlink from 0 may cause WARN_ON */
2818                 set_nlink(inode, 1);
2819         else {
2820                 ldiskfs_inc_count(oh->ot_handle, inode);
2821                 if (!S_ISDIR(inode->i_mode))
2822                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2823         }
2824         spin_unlock(&obj->oo_guard);
2825
2826         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2827         LINVRNT(osd_invariant(obj));
2828
2829         return rc;
2830 }
2831
2832 static int osd_declare_object_ref_del(const struct lu_env *env,
2833                                       struct dt_object *dt,
2834                                       struct thandle *handle)
2835 {
2836         struct osd_thandle *oh;
2837
2838         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2839         LASSERT(handle != NULL);
2840
2841         oh = container_of0(handle, struct osd_thandle, ot_super);
2842         LASSERT(oh->ot_handle == NULL);
2843
2844         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2845                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2846
2847         return 0;
2848 }
2849
2850 /*
2851  * Concurrency: @dt is write locked.
2852  */
2853 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2854                               struct thandle *th)
2855 {
2856         struct osd_object       *obj = osd_dt_obj(dt);
2857         struct inode            *inode = obj->oo_inode;
2858         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2859         struct osd_thandle      *oh;
2860
2861         LINVRNT(osd_invariant(obj));
2862         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2863         LASSERT(osd_write_locked(env, obj));
2864         LASSERT(th != NULL);
2865
2866         oh = container_of0(th, struct osd_thandle, ot_super);
2867         LASSERT(oh->ot_handle != NULL);
2868
2869         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2870
2871         spin_lock(&obj->oo_guard);
2872         /* That can be result of upgrade from old Lustre version and
2873          * applied only to local files.  Just skip this ref_del call.
2874          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2875         if (inode->i_nlink == 0) {
2876                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2877                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2878                              ", maybe an upgraded file? (LU-3915)\n",
2879                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2880                 spin_unlock(&obj->oo_guard);
2881                 return 0;
2882         }
2883
2884         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2885                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2886
2887         ldiskfs_dec_count(oh->ot_handle, inode);
2888         spin_unlock(&obj->oo_guard);
2889
2890         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2891         LINVRNT(osd_invariant(obj));
2892
2893         return 0;
2894 }
2895
2896 /*
2897  * Get the 64-bit version for an inode.
2898  */
2899 static int osd_object_version_get(const struct lu_env *env,
2900                                   struct dt_object *dt, dt_obj_version_t *ver)
2901 {
2902         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2903
2904         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2905                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2906         *ver = LDISKFS_I(inode)->i_fs_version;
2907         return 0;
2908 }
2909
2910 /*
2911  * Concurrency: @dt is read locked.
2912  */
2913 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2914                          struct lu_buf *buf, const char *name,
2915                          struct lustre_capa *capa)
2916 {
2917         struct osd_object      *obj    = osd_dt_obj(dt);
2918         struct inode           *inode  = obj->oo_inode;
2919         struct osd_thread_info *info   = osd_oti_get(env);
2920         struct dentry          *dentry = &info->oti_obj_dentry;
2921
2922         /* version get is not real XATTR but uses xattr API */
2923         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2924                 /* for version we are just using xattr API but change inode
2925                  * field instead */
2926                 if (buf->lb_len == 0)
2927                         return sizeof(dt_obj_version_t);
2928
2929                 if (buf->lb_len < sizeof(dt_obj_version_t))
2930                         return -ERANGE;
2931
2932                 osd_object_version_get(env, dt, buf->lb_buf);
2933
2934                 return sizeof(dt_obj_version_t);
2935         }
2936
2937         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2938         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2939
2940         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2941                 return -EACCES;
2942
2943         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2944 }
2945
2946
2947 static int osd_declare_xattr_set(const struct lu_env *env,
2948                                  struct dt_object *dt,
2949                                  const struct lu_buf *buf, const char *name,
2950                                  int fl, struct thandle *handle)
2951 {
2952         struct osd_thandle *oh;
2953         int credits;
2954         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2955
2956         LASSERT(handle != NULL);
2957
2958         oh = container_of0(handle, struct osd_thandle, ot_super);
2959         LASSERT(oh->ot_handle == NULL);
2960
2961         /* optimistic optimization: LMA is set first and usually fit inode */
2962         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2963                 if (dt_object_exists(dt))
2964                         credits = 0;
2965                 else
2966                         credits = 1;
2967         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2968                 credits = 1;
2969         } else {
2970                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2971                 if (buf && buf->lb_len > sb->s_blocksize) {
2972                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2973                                         sb->s_blocksize_bits;
2974                 }
2975                 /*
2976                  * xattr set may involve inode quota change, reserve credits for
2977                  * dquot_initialize()
2978                  */
2979                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2980         }
2981
2982         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2983
2984         return 0;
2985 }
2986
2987 /*
2988  * Set the 64-bit version for object
2989  */
2990 static void osd_object_version_set(const struct lu_env *env,
2991                                    struct dt_object *dt,
2992                                    dt_obj_version_t *new_version)
2993 {
2994         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2995
2996         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2997                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2998
2999         LDISKFS_I(inode)->i_fs_version = *new_version;
3000         /** Version is set after all inode operations are finished,
3001          *  so we should mark it dirty here */
3002         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3003 }
3004
3005 /*
3006  * Concurrency: @dt is write locked.
3007  */
3008 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3009                          const struct lu_buf *buf, const char *name, int fl,
3010                          struct thandle *handle, struct lustre_capa *capa)
3011 {
3012         struct osd_object      *obj      = osd_dt_obj(dt);
3013         struct inode           *inode    = obj->oo_inode;
3014         struct osd_thread_info *info     = osd_oti_get(env);
3015         int                     fs_flags = 0;
3016         ENTRY;
3017
3018         LASSERT(handle != NULL);
3019
3020         /* version set is not real XATTR */
3021         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3022                 /* for version we are just using xattr API but change inode
3023                  * field instead */
3024                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3025                 osd_object_version_set(env, dt, buf->lb_buf);
3026                 return sizeof(dt_obj_version_t);
3027         }
3028
3029         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3030                 return -EACCES;
3031
3032         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3033                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3034
3035         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3036         if (fl & LU_XATTR_REPLACE)
3037                 fs_flags |= XATTR_REPLACE;
3038
3039         if (fl & LU_XATTR_CREATE)
3040                 fs_flags |= XATTR_CREATE;
3041
3042         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3043                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3044                 int                      rc;
3045
3046                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3047                 if (rc != 0)
3048                         RETURN(rc);
3049
3050                 lma->lma_incompat |= LMAI_STRIPED;
3051                 lustre_lma_swab(lma);
3052                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3053                                      sizeof(*lma), XATTR_REPLACE);
3054                 if (rc != 0)
3055                         RETURN(rc);
3056         }
3057
3058         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3059                                fs_flags);
3060 }
3061
3062 /*
3063  * Concurrency: @dt is read locked.
3064  */
3065 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3066                           struct lu_buf *buf, struct lustre_capa *capa)
3067 {
3068         struct osd_object      *obj    = osd_dt_obj(dt);
3069         struct inode           *inode  = obj->oo_inode;
3070         struct osd_thread_info *info   = osd_oti_get(env);
3071         struct dentry          *dentry = &info->oti_obj_dentry;
3072
3073         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3074         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3075
3076         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3077                 return -EACCES;
3078
3079         dentry->d_inode = inode;
3080         dentry->d_sb = inode->i_sb;
3081         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3082 }
3083
3084 static int osd_declare_xattr_del(const struct lu_env *env,
3085                                  struct dt_object *dt, const char *name,
3086                                  struct thandle *handle)
3087 {
3088         struct osd_thandle *oh;
3089         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3090
3091         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3092         LASSERT(handle != NULL);
3093
3094         oh = container_of0(handle, struct osd_thandle, ot_super);
3095         LASSERT(oh->ot_handle == NULL);
3096
3097         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3098                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3099         /*
3100          * xattr del may involve inode quota change, reserve credits for
3101          * dquot_initialize()
3102          */
3103         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3104
3105         return 0;
3106 }
3107
3108 /*
3109  * Concurrency: @dt is write locked.
3110  */
3111 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3112                          const char *name, struct thandle *handle,
3113                          struct lustre_capa *capa)
3114 {
3115         struct osd_object      *obj    = osd_dt_obj(dt);
3116         struct inode           *inode  = obj->oo_inode;
3117         struct osd_thread_info *info   = osd_oti_get(env);
3118         struct dentry          *dentry = &info->oti_obj_dentry;
3119         int                     rc;
3120
3121         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3122         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3123         LASSERT(handle != NULL);
3124
3125         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3126                 return -EACCES;
3127
3128         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3129
3130         ll_vfs_dq_init(inode);
3131         dentry->d_inode = inode;
3132         dentry->d_sb = inode->i_sb;
3133         rc = inode->i_op->removexattr(dentry, name);
3134         return rc;
3135 }
3136
3137 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3138                                      struct dt_object *dt,
3139                                      struct lustre_capa *old, __u64 opc)
3140 {
3141         struct osd_thread_info *info = osd_oti_get(env);
3142         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3143         struct osd_object *obj = osd_dt_obj(dt);
3144         struct osd_device *osd = osd_obj2dev(obj);
3145         struct lustre_capa_key *key = &info->oti_capa_key;
3146         struct lustre_capa *capa = &info->oti_capa;
3147         struct obd_capa *oc;
3148         struct lu_capainfo *lci;
3149         int rc;
3150         ENTRY;
3151
3152         if (!osd->od_fl_capa)
3153                 RETURN(ERR_PTR(-ENOENT));
3154
3155         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3156         LINVRNT(osd_invariant(obj));
3157
3158         /* renewal sanity check */
3159         if (old && osd_object_auth(env, dt, old, opc))
3160                 RETURN(ERR_PTR(-EACCES));
3161
3162         lci = lu_capainfo_get(env);
3163         if (unlikely(lci == NULL))
3164                 RETURN(ERR_PTR(-ENOENT));
3165
3166         switch (lci->lci_auth) {
3167         case LC_ID_NONE:
3168                 RETURN(NULL);
3169         case LC_ID_PLAIN:
3170                 capa->lc_uid = i_uid_read(obj->oo_inode);
3171                 capa->lc_gid = i_gid_read(obj->oo_inode);
3172                 capa->lc_flags = LC_ID_PLAIN;
3173                 break;
3174         case LC_ID_CONVERT: {
3175                 __u32 d[4], s[4];
3176
3177                 s[0] = i_uid_read(obj->oo_inode);
3178                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3179                 s[2] = i_uid_read(obj->oo_inode);
3180                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3181                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3182                 if (unlikely(rc))
3183                         RETURN(ERR_PTR(rc));
3184
3185                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3186                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3187                 capa->lc_flags = LC_ID_CONVERT;
3188                 break;
3189         }
3190         default:
3191                 RETURN(ERR_PTR(-EINVAL));
3192         }
3193
3194         capa->lc_fid = *fid;
3195         capa->lc_opc = opc;
3196         capa->lc_flags |= osd->od_capa_alg << 24;
3197         capa->lc_timeout = osd->od_capa_timeout;
3198         capa->lc_expiry = 0;
3199
3200         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3201         if (oc) {
3202                 LASSERT(!capa_is_expired(oc));
3203                 RETURN(oc);
3204         }
3205
3206         spin_lock(&capa_lock);
3207         *key = osd->od_capa_keys[1];
3208         spin_unlock(&capa_lock);
3209
3210         capa->lc_keyid = key->lk_keyid;
3211         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3212
3213         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3214         if (rc) {
3215                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3216                 RETURN(ERR_PTR(rc));
3217         }
3218
3219         oc = capa_add(osd->od_capa_hash, capa);
3220         RETURN(oc);
3221 }
3222
3223 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3224                            __u64 start, __u64 end)
3225 {
3226         struct osd_object       *obj    = osd_dt_obj(dt);
3227         struct inode            *inode  = obj->oo_inode;
3228         struct osd_thread_info  *info   = osd_oti_get(env);
3229         struct dentry           *dentry = &info->oti_obj_dentry;
3230         struct file             *file   = &info->oti_file;
3231         int                     rc;
3232
3233         ENTRY;
3234
3235         dentry->d_inode = inode;
3236         dentry->d_sb = inode->i_sb;
3237         file->f_dentry = dentry;
3238         file->f_mapping = inode->i_mapping;
3239         file->f_op = inode->i_fop;
3240         set_file_inode(file, inode);
3241
3242 #ifdef HAVE_FILE_FSYNC_4ARGS
3243         rc = file->f_op->fsync(file, start, end, 0);
3244 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3245         mutex_lock(&inode->i_mutex);
3246         rc = file->f_op->fsync(file, 0);
3247         mutex_unlock(&inode->i_mutex);
3248 #else
3249         mutex_lock(&inode->i_mutex);
3250         rc = file->f_op->fsync(file, dentry, 0);
3251         mutex_unlock(&inode->i_mutex);
3252 #endif
3253
3254         RETURN(rc);
3255 }
3256
3257 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3258                         void **data)
3259 {
3260         struct osd_object *obj = osd_dt_obj(dt);
3261         ENTRY;
3262
3263         *data = (void *)obj->oo_inode;
3264         RETURN(0);
3265 }
3266
3267 /*
3268  * Index operations.
3269  */
3270
3271 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3272                            const struct dt_index_features *feat)
3273 {
3274         struct iam_descr *descr;
3275
3276         if (osd_object_is_root(o))
3277                 return feat == &dt_directory_features;
3278
3279         LASSERT(o->oo_dir != NULL);
3280
3281         descr = o->oo_dir->od_container.ic_descr;
3282         if (feat == &dt_directory_features) {
3283                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3284                         return 1;
3285                 else
3286                         return 0;
3287         } else {
3288                 return
3289                         feat->dif_keysize_min <= descr->id_key_size &&
3290                         descr->id_key_size <= feat->dif_keysize_max &&
3291                         feat->dif_recsize_min <= descr->id_rec_size &&
3292                         descr->id_rec_size <= feat->dif_recsize_max &&
3293                         !(feat->dif_flags & (DT_IND_VARKEY |
3294                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3295                         ergo(feat->dif_flags & DT_IND_UPDATE,
3296                              1 /* XXX check that object (and file system) is
3297                                 * writable */);
3298         }
3299 }
3300
3301 static int osd_iam_container_init(const struct lu_env *env,
3302                                   struct osd_object *obj,
3303                                   struct osd_directory *dir)
3304 {
3305         struct iam_container *bag = &dir->od_container;
3306         int result;
3307
3308         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3309         if (result != 0)
3310                 return result;
3311
3312         result = iam_container_setup(bag);
3313         if (result == 0)
3314                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3315         else
3316                 iam_container_fini(bag);
3317
3318         return result;
3319 }
3320
3321
3322 /*
3323  * Concurrency: no external locking is necessary.
3324  */
3325 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3326                          const struct dt_index_features *feat)
3327 {
3328         int                      result;
3329         int                      skip_iam = 0;
3330         struct osd_object       *obj = osd_dt_obj(dt);
3331
3332         LINVRNT(osd_invariant(obj));
3333
3334         if (osd_object_is_root(obj)) {
3335                 dt->do_index_ops = &osd_index_ea_ops;
3336                 result = 0;
3337         } else if (feat == &dt_directory_features) {
3338                 dt->do_index_ops = &osd_index_ea_ops;
3339                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3340                         result = 0;
3341                 else
3342                         result = -ENOTDIR;
3343                 skip_iam = 1;
3344         } else if (unlikely(feat == &dt_otable_features)) {
3345                 dt->do_index_ops = &osd_otable_ops;
3346                 return 0;
3347         } else if (unlikely(feat == &dt_acct_features)) {
3348                 dt->do_index_ops = &osd_acct_index_ops;
3349                 result = 0;
3350                 skip_iam = 1;
3351         } else if (!osd_has_index(obj)) {
3352                 struct osd_directory *dir;
3353
3354                 OBD_ALLOC_PTR(dir);
3355                 if (dir != NULL) {
3356
3357                         spin_lock(&obj->oo_guard);
3358                         if (obj->oo_dir == NULL)
3359                                 obj->oo_dir = dir;
3360                         else
3361                                 /*
3362                                  * Concurrent thread allocated container data.
3363                                  */
3364                                 OBD_FREE_PTR(dir);
3365                         spin_unlock(&obj->oo_guard);
3366                         /*
3367                          * Now, that we have container data, serialize its
3368                          * initialization.
3369                          */
3370                         down_write(&obj->oo_ext_idx_sem);
3371                         /*
3372                          * recheck under lock.
3373                          */
3374                         if (!osd_has_index(obj))
3375                                 result = osd_iam_container_init(env, obj,
3376                                                                 obj->oo_dir);
3377                         else
3378                                 result = 0;
3379                         up_write(&obj->oo_ext_idx_sem);
3380                 } else {
3381                         result = -ENOMEM;
3382                 }
3383         } else {
3384                 result = 0;
3385         }
3386
3387         if (result == 0 && skip_iam == 0) {
3388                 if (!osd_iam_index_probe(env, obj, feat))
3389                         result = -ENOTDIR;
3390         }
3391         LINVRNT(osd_invariant(obj));
3392
3393         if (result == 0 && is_quota_glb_feat(feat) &&
3394             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3395                 result = osd_quota_migration(env, dt, feat);
3396
3397         return result;
3398 }
3399
3400 static int osd_otable_it_attr_get(const struct lu_env *env,
3401                                  struct dt_object *dt,
3402                                  struct lu_attr *attr,
3403                                  struct lustre_capa *capa)
3404 {
3405         attr->la_valid = 0;
3406         return 0;
3407 }
3408
3409 static const struct dt_object_operations osd_obj_ops = {
3410         .do_read_lock         = osd_object_read_lock,
3411         .do_write_lock        = osd_object_write_lock,
3412         .do_read_unlock       = osd_object_read_unlock,
3413         .do_write_unlock      = osd_object_write_unlock,
3414         .do_write_locked      = osd_object_write_locked,
3415         .do_attr_get          = osd_attr_get,
3416         .do_declare_attr_set  = osd_declare_attr_set,
3417         .do_attr_set          = osd_attr_set,
3418         .do_ah_init           = osd_ah_init,
3419         .do_declare_create    = osd_declare_object_create,
3420         .do_create            = osd_object_create,
3421         .do_declare_destroy   = osd_declare_object_destroy,
3422         .do_destroy           = osd_object_destroy,
3423         .do_index_try         = osd_index_try,
3424         .do_declare_ref_add   = osd_declare_object_ref_add,
3425         .do_ref_add           = osd_object_ref_add,
3426         .do_declare_ref_del   = osd_declare_object_ref_del,
3427         .do_ref_del           = osd_object_ref_del,
3428         .do_xattr_get         = osd_xattr_get,
3429         .do_declare_xattr_set = osd_declare_xattr_set,
3430         .do_xattr_set         = osd_xattr_set,
3431         .do_declare_xattr_del = osd_declare_xattr_del,
3432         .do_xattr_del         = osd_xattr_del,
3433         .do_xattr_list        = osd_xattr_list,
3434         .do_capa_get          = osd_capa_get,
3435         .do_object_sync       = osd_object_sync,
3436         .do_data_get          = osd_data_get,
3437 };
3438
3439 /**
3440  * dt_object_operations for interoperability mode
3441  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3442  */
3443 static const struct dt_object_operations osd_obj_ea_ops = {
3444         .do_read_lock         = osd_object_read_lock,
3445         .do_write_lock        = osd_object_write_lock,
3446         .do_read_unlock       = osd_object_read_unlock,
3447         .do_write_unlock      = osd_object_write_unlock,
3448         .do_write_locked      = osd_object_write_locked,
3449         .do_attr_get          = osd_attr_get,
3450         .do_declare_attr_set  = osd_declare_attr_set,
3451         .do_attr_set          = osd_attr_set,
3452         .do_ah_init           = osd_ah_init,
3453         .do_declare_create    = osd_declare_object_create,
3454         .do_create            = osd_object_ea_create,
3455         .do_declare_destroy   = osd_declare_object_destroy,
3456         .do_destroy           = osd_object_destroy,
3457         .do_index_try         = osd_index_try,
3458         .do_declare_ref_add   = osd_declare_object_ref_add,
3459         .do_ref_add           = osd_object_ref_add,
3460         .do_declare_ref_del   = osd_declare_object_ref_del,
3461         .do_ref_del           = osd_object_ref_del,
3462         .do_xattr_get         = osd_xattr_get,
3463         .do_declare_xattr_set = osd_declare_xattr_set,
3464         .do_xattr_set         = osd_xattr_set,
3465         .do_declare_xattr_del = osd_declare_xattr_del,
3466         .do_xattr_del         = osd_xattr_del,
3467         .do_xattr_list        = osd_xattr_list,
3468         .do_capa_get          = osd_capa_get,
3469         .do_object_sync       = osd_object_sync,
3470         .do_data_get          = osd_data_get,
3471 };
3472
3473 static const struct dt_object_operations osd_obj_otable_it_ops = {
3474         .do_attr_get    = osd_otable_it_attr_get,
3475         .do_index_try   = osd_index_try,
3476 };
3477
3478 static int osd_index_declare_iam_delete(const struct lu_env *env,
3479                                         struct dt_object *dt,
3480                                         const struct dt_key *key,
3481                                         struct thandle *handle)
3482 {
3483         struct osd_thandle    *oh;
3484
3485         oh = container_of0(handle, struct osd_thandle, ot_super);
3486         LASSERT(oh->ot_handle == NULL);
3487
3488         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3489                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3490
3491         return 0;
3492 }
3493
3494 /**
3495  *      delete a (key, value) pair from index \a dt specified by \a key
3496  *
3497  *      \param  dt      osd index object
3498  *      \param  key     key for index
3499  *      \param  rec     record reference
3500  *      \param  handle  transaction handler
3501  *
3502  *      \retval  0  success
3503  *      \retval -ve   failure
3504  */
3505
3506 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3507                                 const struct dt_key *key,
3508                                 struct thandle *handle,
3509                                 struct lustre_capa *capa)
3510 {
3511         struct osd_thread_info *oti = osd_oti_get(env);
3512         struct osd_object      *obj = osd_dt_obj(dt);
3513         struct osd_thandle     *oh;
3514         struct iam_path_descr  *ipd;
3515         struct iam_container   *bag = &obj->oo_dir->od_container;
3516         int                     rc;
3517
3518         ENTRY;
3519
3520         LINVRNT(osd_invariant(obj));
3521         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3522         LASSERT(bag->ic_object == obj->oo_inode);
3523         LASSERT(handle != NULL);
3524
3525         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3526                 RETURN(-EACCES);
3527
3528         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3529
3530         ipd = osd_idx_ipd_get(env, bag);
3531         if (unlikely(ipd == NULL))
3532                 RETURN(-ENOMEM);
3533
3534         oh = container_of0(handle, struct osd_thandle, ot_super);
3535         LASSERT(oh->ot_handle != NULL);
3536         LASSERT(oh->ot_handle->h_transaction != NULL);
3537
3538         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3539                 /* swab quota uid/gid provided by caller */
3540                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3541                 key = (const struct dt_key *)&oti->oti_quota_id;
3542         }
3543
3544         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3545         osd_ipd_put(env, bag, ipd);
3546         LINVRNT(osd_invariant(obj));
3547         RETURN(rc);
3548 }
3549
3550 static int osd_index_declare_ea_delete(const struct lu_env *env,
3551                                        struct dt_object *dt,
3552                                        const struct dt_key *key,
3553                                        struct thandle *handle)
3554 {
3555         struct osd_thandle *oh;
3556         struct inode       *inode;
3557         int                 rc;
3558         ENTRY;
3559
3560         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3561         LASSERT(handle != NULL);
3562
3563         oh = container_of0(handle, struct osd_thandle, ot_super);
3564         LASSERT(oh->ot_handle == NULL);
3565
3566         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3567                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3568
3569         inode = osd_dt_obj(dt)->oo_inode;
3570         LASSERT(inode);
3571
3572         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3573                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3574         RETURN(rc);
3575 }
3576
3577 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3578                                           struct dt_rec *fid)
3579 {
3580         struct osd_fid_pack *rec;
3581         int                  rc = -ENODATA;
3582
3583         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3584                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3585                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3586         }
3587         return rc;
3588 }
3589
3590 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3591                           const struct lu_fid *fid)
3592 {
3593         struct seq_server_site  *ss = osd_seq_site(osd);
3594         ENTRY;
3595
3596         /* FID seqs not in FLDB, must be local seq */
3597         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3598                 RETURN(0);
3599
3600         /* If FLD is not being initialized yet, it only happens during the
3601          * initialization, likely during mgs initialization, and we assume
3602          * this is local FID. */
3603         if (ss == NULL || ss->ss_server_fld == NULL)
3604                 RETURN(0);
3605
3606         /* Only check the local FLDB here */
3607         if (osd_seq_exists(env, osd, fid_seq(fid)))
3608                 RETURN(0);
3609
3610         RETURN(1);
3611 }
3612
3613 /**
3614  * Index delete function for interoperability mode (b11826).
3615  * It will remove the directory entry added by osd_index_ea_insert().
3616  * This entry is needed to maintain name->fid mapping.
3617  *
3618  * \param key,  key i.e. file entry to be deleted
3619  *
3620  * \retval   0, on success
3621  * \retval -ve, on error
3622  */
3623 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3624                                const struct dt_key *key,
3625                                struct thandle *handle,
3626                                struct lustre_capa *capa)
3627 {
3628         struct osd_object          *obj    = osd_dt_obj(dt);
3629         struct inode               *dir    = obj->oo_inode;
3630         struct dentry              *dentry;
3631         struct osd_thandle         *oh;
3632         struct ldiskfs_dir_entry_2 *de = NULL;
3633         struct buffer_head         *bh;
3634         struct htree_lock          *hlock = NULL;
3635         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3636         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3637         int                        rc;
3638         ENTRY;
3639
3640         LINVRNT(osd_invariant(obj));
3641         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3642         LASSERT(handle != NULL);
3643
3644         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3645
3646         oh = container_of(handle, struct osd_thandle, ot_super);
3647         LASSERT(oh->ot_handle != NULL);
3648         LASSERT(oh->ot_handle->h_transaction != NULL);
3649
3650         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3651                 RETURN(-EACCES);
3652
3653         ll_vfs_dq_init(dir);
3654         dentry = osd_child_dentry_get(env, obj,
3655                                       (char *)key, strlen((char *)key));
3656
3657         if (obj->oo_hl_head != NULL) {
3658                 hlock = osd_oti_get(env)->oti_hlock;
3659                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3660                                    dir, LDISKFS_HLOCK_DEL);
3661         } else {
3662                 down_write(&obj->oo_ext_idx_sem);
3663         }
3664
3665         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3666         if (bh) {
3667                 __u32 ino = 0;
3668
3669                 /* If this is not the ".." entry, it might be a remote DNE
3670                  * entry and  we need to check if the FID is for a remote
3671                  * MDT.  If the FID is  not in the directory entry (e.g.
3672                  * upgraded 1.8 filesystem without dirdata enabled) then
3673                  * we need to get the FID from the LMA. For a remote directory
3674                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3675                  * case.
3676                  *
3677                  * Delete the entry before the agent inode in order to
3678                  * simplify error handling.  At worst an error after deleting
3679                  * the entry first might leak the agent inode afterward. The
3680                  * reverse would need filesystem abort in case of error deleting
3681                  * the entry after the agent had been removed, or leave a
3682                  * dangling entry pointing at a random inode. */
3683                 if (strcmp((char *)key, dotdot) != 0) {
3684                         LASSERT(de != NULL);
3685                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3686                         /* If Fid is not in dentry, try to get it from LMA */
3687                         if (rc == -ENODATA) {
3688                                 struct osd_inode_id *id;
3689                                 struct inode *inode;
3690
3691                                 /* Before trying to get fid from the inode,
3692                                  * check whether the inode is valid.
3693                                  *
3694                                  * If the inode has been deleted, do not go
3695                                  * ahead to do osd_ea_fid_get, which will set
3696                                  * the inode to bad inode, which might cause
3697                                  * the inode to be deleted uncorrectly */
3698                                 inode = ldiskfs_iget(osd_sb(osd),
3699                                                      le32_to_cpu(de->inode));
3700                                 if (IS_ERR(inode)) {
3701                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3702                                                "error.\n", osd_name(osd),
3703                                                PFID(fid));
3704                                         rc = PTR_ERR(inode);
3705                                 } else {
3706                                         if (likely(inode->i_nlink != 0)) {
3707                                                 id = &osd_oti_get(env)->oti_id;
3708                                                 rc = osd_ea_fid_get(env, obj,
3709                                                         le32_to_cpu(de->inode),
3710                                                                     fid, id);
3711                                         } else {
3712                                                 CDEBUG(D_INFO, "%s: %u "DFID
3713                                                        "deleted.\n",
3714                                                        osd_name(osd),
3715                                                        le32_to_cpu(de->inode),
3716                                                        PFID(fid));
3717                                                 rc = -ESTALE;
3718                                         }
3719                                         iput(inode);
3720                                 }
3721                         }
3722                         if (rc == 0 &&
3723                             unlikely(osd_remote_fid(env, osd, fid)))
3724                                 /* Need to delete agent inode */
3725                                 ino = le32_to_cpu(de->inode);
3726                 }
3727                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3728                 brelse(bh);
3729                 if (rc == 0 && unlikely(ino != 0)) {
3730                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3731                                                           oh);
3732                         if (rc != 0)
3733                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3734                                        osd_name(osd), PFID(fid), rc);
3735                 }
3736         } else {
3737                 rc = -ENOENT;
3738         }
3739         if (hlock != NULL)
3740                 ldiskfs_htree_unlock(hlock);
3741         else
3742                 up_write(&obj->oo_ext_idx_sem);
3743
3744         if (rc != 0)
3745                 GOTO(out, rc);
3746
3747         /* For inode on the remote MDT, .. will point to
3748          * /Agent directory, Check whether it needs to delete
3749          * from agent directory */
3750         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3751                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3752                                                    oh);
3753                 if (rc != 0 && rc != -ENOENT) {
3754                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3755                                osd_name(osd), PFID(fid), rc);
3756                 }
3757
3758                 if (rc == -ENOENT)
3759                         rc = 0;
3760
3761                 GOTO(out, rc);
3762         }
3763 out:
3764
3765         LASSERT(osd_invariant(obj));
3766         RETURN(rc);
3767 }
3768
3769 /**
3770  *      Lookup index for \a key and copy record to \a rec.
3771  *
3772  *      \param  dt      osd index object
3773  *      \param  key     key for index
3774  *      \param  rec     record reference
3775  *
3776  *      \retval  +ve  success : exact mach
3777  *      \retval  0    return record with key not greater than \a key
3778  *      \retval -ve   failure
3779  */
3780 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3781                                 struct dt_rec *rec, const struct dt_key *key,
3782                                 struct lustre_capa *capa)
3783 {
3784         struct osd_object      *obj = osd_dt_obj(dt);
3785         struct iam_path_descr  *ipd;
3786         struct iam_container   *bag = &obj->oo_dir->od_container;
3787         struct osd_thread_info *oti = osd_oti_get(env);
3788         struct iam_iterator    *it = &oti->oti_idx_it;
3789         struct iam_rec         *iam_rec;
3790         int                     rc;
3791
3792         ENTRY;
3793
3794         LASSERT(osd_invariant(obj));
3795         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3796         LASSERT(bag->ic_object == obj->oo_inode);
3797
3798         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3799                 RETURN(-EACCES);
3800
3801         ipd = osd_idx_ipd_get(env, bag);
3802         if (IS_ERR(ipd))
3803                 RETURN(-ENOMEM);
3804
3805         /* got ipd now we can start iterator. */
3806         iam_it_init(it, bag, 0, ipd);
3807
3808         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3809                 /* swab quota uid/gid provided by caller */
3810                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3811                 key = (const struct dt_key *)&oti->oti_quota_id;
3812         }
3813
3814         rc = iam_it_get(it, (struct iam_key *)key);
3815         if (rc >= 0) {
3816                 if (S_ISDIR(obj->oo_inode->i_mode))
3817                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3818                 else
3819                         iam_rec = (struct iam_rec *) rec;
3820
3821                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3822
3823                 if (S_ISDIR(obj->oo_inode->i_mode))
3824                         osd_fid_unpack((struct lu_fid *) rec,
3825                                        (struct osd_fid_pack *)iam_rec);
3826                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3827                         osd_quota_unpack(obj, rec);
3828         }
3829
3830         iam_it_put(it);
3831         iam_it_fini(it);
3832         osd_ipd_put(env, bag, ipd);
3833
3834         LINVRNT(osd_invariant(obj));
3835
3836         RETURN(rc);
3837 }
3838
3839 static int osd_index_declare_iam_insert(const struct lu_env *env,
3840                                         struct dt_object *dt,
3841                                         const struct dt_rec *rec,
3842                                         const struct dt_key *key,
3843                                         struct thandle *handle)
3844 {
3845         struct osd_thandle *oh;
3846
3847         LASSERT(handle != NULL);
3848
3849         oh = container_of0(handle, struct osd_thandle, ot_super);
3850         LASSERT(oh->ot_handle == NULL);
3851
3852         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3853                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3854
3855         return 0;
3856 }
3857
3858 /**
3859  *      Inserts (key, value) pair in \a dt index object.
3860  *
3861  *      \param  dt      osd index object
3862  *      \param  key     key for index
3863  *      \param  rec     record reference
3864  *      \param  th      transaction handler
3865  *
3866  *      \retval  0  success
3867  *      \retval -ve failure
3868  */
3869 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3870                                 const struct dt_rec *rec,
3871                                 const struct dt_key *key, struct thandle *th,
3872                                 struct lustre_capa *capa, int ignore_quota)
3873 {
3874         struct osd_object     *obj = osd_dt_obj(dt);
3875         struct iam_path_descr *ipd;
3876         struct osd_thandle    *oh;
3877         struct iam_container  *bag = &obj->oo_dir->od_container;
3878         struct osd_thread_info *oti = osd_oti_get(env);
3879         struct iam_rec         *iam_rec;
3880         int                     rc;
3881
3882         ENTRY;
3883
3884         LINVRNT(osd_invariant(obj));
3885         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3886         LASSERT(bag->ic_object == obj->oo_inode);
3887         LASSERT(th != NULL);
3888
3889         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3890                 RETURN(-EACCES);
3891
3892         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3893
3894         ipd = osd_idx_ipd_get(env, bag);
3895         if (unlikely(ipd == NULL))
3896                 RETURN(-ENOMEM);
3897
3898         oh = container_of0(th, struct osd_thandle, ot_super);
3899         LASSERT(oh->ot_handle != NULL);
3900         LASSERT(oh->ot_handle->h_transaction != NULL);
3901         if (S_ISDIR(obj->oo_inode->i_mode)) {
3902                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3903                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3904         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3905                 /* pack quota uid/gid */
3906                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3907                 key = (const struct dt_key *)&oti->oti_quota_id;
3908                 /* pack quota record */
3909                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3910                 iam_rec = (struct iam_rec *)rec;
3911         } else {
3912                 iam_rec = (struct iam_rec *)rec;
3913         }
3914
3915         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3916                         iam_rec, ipd);
3917         osd_ipd_put(env, bag, ipd);
3918         LINVRNT(osd_invariant(obj));
3919         RETURN(rc);
3920 }
3921
3922 /**
3923  * Calls ldiskfs_add_entry() to add directory entry
3924  * into the directory. This is required for
3925  * interoperability mode (b11826)
3926  *
3927  * \retval   0, on success
3928  * \retval -ve, on error
3929  */
3930 static int __osd_ea_add_rec(struct osd_thread_info *info,
3931                             struct osd_object *pobj, struct inode  *cinode,
3932                             const char *name, const struct lu_fid *fid,
3933                             struct htree_lock *hlock, struct thandle *th)
3934 {
3935         struct ldiskfs_dentry_param *ldp;
3936         struct dentry               *child;
3937         struct osd_thandle          *oth;
3938         int                          rc;
3939
3940         oth = container_of(th, struct osd_thandle, ot_super);
3941         LASSERT(oth->ot_handle != NULL);
3942         LASSERT(oth->ot_handle->h_transaction != NULL);
3943         LASSERT(pobj->oo_inode);
3944
3945         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3946         if (unlikely(pobj->oo_inode ==
3947                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3948                 ldp->edp_magic = 0;
3949         else
3950                 osd_get_ldiskfs_dirent_param(ldp, fid);
3951         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3952         child->d_fsdata = (void *)ldp;
3953         ll_vfs_dq_init(pobj->oo_inode);
3954         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3955         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3956                 struct ldiskfs_dir_entry_2      *de;
3957                 struct buffer_head              *bh;
3958                 int                              rc1;
3959
3960                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3961                                             NULL, hlock);
3962                 if (bh != NULL) {
3963                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3964                                                                bh);
3965                         if (rc1 == 0) {
3966                                 if (S_ISDIR(cinode->i_mode))
3967                                         de->file_type = LDISKFS_DIRENT_LUFID |
3968                                                         LDISKFS_FT_REG_FILE;
3969                                 else
3970                                         de->file_type = LDISKFS_DIRENT_LUFID |
3971                                                         LDISKFS_FT_DIR;
3972                                 ldiskfs_journal_dirty_metadata(oth->ot_handle,
3973                                                                bh);
3974                                 brelse(bh);
3975                         }
3976                 }
3977         }
3978
3979         RETURN(rc);
3980 }
3981
3982 /**
3983  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3984  * into the directory.Also sets flags into osd object to
3985  * indicate dot and dotdot are created. This is required for
3986  * interoperability mode (b11826)
3987  *
3988  * \param dir   directory for dot and dotdot fixup.
3989  * \param obj   child object for linking
3990  *
3991  * \retval   0, on success
3992  * \retval -ve, on error
3993  */
3994 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3995                               struct osd_object *dir,
3996                               struct inode *parent_dir, const char *name,
3997                               const struct lu_fid *dot_fid,
3998                               const struct lu_fid *dot_dot_fid,
3999                               struct thandle *th)
4000 {
4001         struct inode                *inode = dir->oo_inode;
4002         struct osd_thandle          *oth;
4003         int result = 0;
4004
4005         oth = container_of(th, struct osd_thandle, ot_super);
4006         LASSERT(oth->ot_handle->h_transaction != NULL);
4007         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4008
4009         if (strcmp(name, dot) == 0) {
4010                 if (dir->oo_compat_dot_created) {
4011                         result = -EEXIST;
4012                 } else {
4013                         LASSERT(inode == parent_dir);
4014                         dir->oo_compat_dot_created = 1;
4015                         result = 0;
4016                 }
4017         } else if (strcmp(name, dotdot) == 0) {
4018                 if (!dir->oo_compat_dot_created)
4019                         return -EINVAL;
4020                 /* in case of rename, dotdot is already created */
4021                 if (dir->oo_compat_dotdot_created) {
4022                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4023                                                 dot_dot_fid, NULL, th);
4024                 }
4025
4026                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT) ||
4027                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT2)) {
4028                         struct lu_fid tfid = *dot_dot_fid;
4029
4030                         tfid.f_oid--;
4031                         result = osd_add_dot_dotdot_internal(info,
4032                                         dir->oo_inode, parent_dir, dot_fid,
4033                                         &tfid, oth);
4034                 } else {
4035                         result = osd_add_dot_dotdot_internal(info,
4036                                         dir->oo_inode, parent_dir, dot_fid,
4037                                         dot_dot_fid, oth);
4038                 }
4039
4040                 if (result == 0)
4041                         dir->oo_compat_dotdot_created = 1;
4042         }
4043
4044         return result;
4045 }
4046
4047
4048 /**
4049  * It will call the appropriate osd_add* function and return the
4050  * value, return by respective functions.
4051  */
4052 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4053                           struct inode *cinode, const char *name,
4054                           const struct lu_fid *fid, struct thandle *th)
4055 {
4056         struct osd_thread_info *info   = osd_oti_get(env);
4057         struct htree_lock      *hlock;
4058         int                     rc;
4059
4060         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4061
4062         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4063                                                    name[2] =='\0'))) {
4064                 if (hlock != NULL) {
4065                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4066                                            pobj->oo_inode, 0);
4067                 } else {
4068                         down_write(&pobj->oo_ext_idx_sem);
4069                 }
4070
4071                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4072                                         lu_object_fid(&pobj->oo_dt.do_lu),
4073                                         fid, th);
4074         } else {
4075                 if (hlock != NULL) {
4076                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4077                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4078                 } else {
4079                         down_write(&pobj->oo_ext_idx_sem);
4080                 }
4081
4082                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4083                         struct lu_fid *tfid = &info->oti_fid;
4084
4085                         *tfid = *fid;
4086                         tfid->f_ver = ~0;
4087                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4088                                               tfid, hlock, th);
4089                 } else {
4090                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4091                                               hlock, th);
4092                 }
4093         }
4094         if (hlock != NULL)
4095                 ldiskfs_htree_unlock(hlock);
4096         else
4097                 up_write(&pobj->oo_ext_idx_sem);
4098
4099         return rc;
4100 }
4101
4102 static void
4103 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4104                       struct osd_idmap_cache *oic)
4105 {
4106         struct osd_scrub    *scrub = &dev->od_scrub;
4107         struct lu_fid       *fid   = &oic->oic_fid;
4108         struct osd_inode_id *id    = &oti->oti_id;
4109         int                  once  = 0;
4110         int                  rc;
4111         ENTRY;
4112
4113         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4114                 RETURN_EXIT;
4115
4116 again:
4117         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4118         if (rc != 0 && rc != -ENOENT)
4119                 RETURN_EXIT;
4120
4121         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4122                 RETURN_EXIT;
4123
4124         if (thread_is_running(&scrub->os_thread)) {
4125                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4126                 /* There is race condition between osd_oi_lookup and OI scrub.
4127                  * The OI scrub finished just after osd_oi_lookup() failure.
4128                  * Under such case, it is unnecessary to trigger OI scrub again,
4129                  * but try to call osd_oi_lookup() again. */
4130                 if (unlikely(rc == -EAGAIN))
4131                         goto again;
4132
4133                 RETURN_EXIT;
4134         }
4135
4136         if (!dev->od_noscrub && ++once == 1) {
4137                 rc = osd_scrub_start(dev);
4138                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4139                               ", rc = %d [2]\n",
4140                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4141                               PFID(fid), rc);
4142                 if (rc == 0)
4143                         goto again;
4144         }
4145
4146         EXIT;
4147 }
4148
4149 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4150                                struct osd_device *dev,
4151                                struct osd_idmap_cache *oic,
4152                                struct lu_fid *fid, __u32 ino)
4153 {
4154         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4155         struct inode            *inode;
4156         int                      rc;
4157
4158         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4159         inode = osd_iget(oti, dev, &oic->oic_lid);
4160         if (IS_ERR(inode)) {
4161                 fid_zero(&oic->oic_fid);
4162                 return PTR_ERR(inode);
4163         }
4164
4165         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4166         iput(inode);
4167         if (rc != 0)
4168                 fid_zero(&oic->oic_fid);
4169         else
4170                 *fid = oic->oic_fid = lma->lma_self_fid;
4171         return rc;
4172 }
4173
4174 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4175                      struct osd_inode_id *id, const struct lu_fid *fid)
4176 {
4177         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4178                id->oii_ino, id->oii_gen, info);
4179         info->oti_cache.oic_lid = *id;
4180         info->oti_cache.oic_fid = *fid;
4181         info->oti_cache.oic_dev = osd;
4182
4183         return 0;
4184 }
4185
4186 /**
4187  * Get parent FID from the linkEA.
4188  *
4189  * For a directory which parent resides on remote MDT, to satisfy the
4190  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4191  * the other hand, to make the lookup(..) on the directory can return
4192  * the real parent FID, we append the real parent FID after its ".."
4193  * name entry in the /REMOTE_PARENT_DIR.
4194  *
4195  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4196  * backup. So after the restore, we cannot get the right parent FID from
4197  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4198  * we have stored the real parent FID in the directory object's linkEA,
4199  * we can parse the linkEA for the real parent FID.
4200  *
4201  * \param[in] env       pointer to the thread context
4202  * \param[in] obj       pointer to the object to be handled
4203  * \param[out]fid       pointer to the buffer to hold the parent FID
4204  *
4205  * \retval              0 for getting the real parent FID successfully
4206  * \retval              negative error number on failure
4207  */
4208 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4209                                     struct osd_object *obj,
4210                                     struct lu_fid *fid)
4211 {
4212         struct osd_thread_info  *oti    = osd_oti_get(env);
4213         struct lu_buf           *buf    = &oti->oti_big_buf;
4214         struct dentry           *dentry = &oti->oti_obj_dentry;
4215         struct inode            *inode  = obj->oo_inode;
4216         struct linkea_data       ldata  = { 0 };
4217         int                      rc;
4218         ENTRY;
4219
4220         fid_zero(fid);
4221         if (!S_ISDIR(inode->i_mode))
4222                 RETURN(-EIO);
4223
4224 again:
4225         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4226                              buf->lb_buf, buf->lb_len);
4227         if (rc == -ERANGE) {
4228                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4229                                      NULL, 0);
4230                 if (rc > 0) {
4231                         lu_buf_realloc(buf, rc);
4232                         if (buf->lb_buf == NULL)
4233                                 RETURN(-ENOMEM);
4234
4235                         goto again;
4236                 }
4237         }
4238
4239         if (unlikely(rc == 0))
4240                 RETURN(-ENODATA);
4241
4242         if (rc < 0)
4243                 RETURN(rc);
4244
4245         if (unlikely(buf->lb_buf == NULL)) {
4246                 lu_buf_realloc(buf, rc);
4247                 if (buf->lb_buf == NULL)
4248                         RETURN(-ENOMEM);
4249
4250                 goto again;
4251         }
4252
4253         ldata.ld_buf = buf;
4254         rc = linkea_init(&ldata);
4255         if (rc == 0) {
4256                 linkea_first_entry(&ldata);
4257                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4258         }
4259
4260         RETURN(rc);
4261 }
4262
4263 /**
4264  * Calls ->lookup() to find dentry. From dentry get inode and
4265  * read inode's ea to get fid. This is required for  interoperability
4266  * mode (b11826)
4267  *
4268  * \retval   0, on success
4269  * \retval -ve, on error
4270  */
4271 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4272                              struct dt_rec *rec, const struct dt_key *key)
4273 {
4274         struct inode               *dir    = obj->oo_inode;
4275         struct dentry              *dentry;
4276         struct ldiskfs_dir_entry_2 *de;
4277         struct buffer_head         *bh;
4278         struct lu_fid              *fid = (struct lu_fid *) rec;
4279         struct htree_lock          *hlock = NULL;
4280         int                         ino;
4281         int                         rc;
4282         ENTRY;
4283
4284         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4285
4286         dentry = osd_child_dentry_get(env, obj,
4287                                       (char *)key, strlen((char *)key));
4288
4289         if (obj->oo_hl_head != NULL) {
4290                 hlock = osd_oti_get(env)->oti_hlock;
4291                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4292                                    dir, LDISKFS_HLOCK_LOOKUP);
4293         } else {
4294                 down_read(&obj->oo_ext_idx_sem);
4295         }
4296
4297         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4298         if (bh) {
4299                 struct osd_thread_info *oti = osd_oti_get(env);
4300                 struct osd_inode_id *id = &oti->oti_id;
4301                 struct osd_idmap_cache *oic = &oti->oti_cache;
4302                 struct osd_device *dev = osd_obj2dev(obj);
4303                 struct osd_scrub *scrub = &dev->od_scrub;
4304                 struct scrub_file *sf = &scrub->os_file;
4305
4306                 ino = le32_to_cpu(de->inode);
4307                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4308                         brelse(bh);
4309                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4310                         GOTO(out, rc);
4311                 }
4312
4313                 rc = osd_get_fid_from_dentry(de, rec);
4314
4315                 /* done with de, release bh */
4316                 brelse(bh);
4317                 if (rc != 0) {
4318                         if (unlikely(ino == osd_remote_parent_ino(dev)))
4319                                 /* If the parent is on remote MDT, and there
4320                                  * is no FID-in-dirent, then we have to get
4321                                  * the parent FID from the linkEA.  */
4322                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
4323                         else
4324                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4325                 } else {
4326                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4327                 }
4328
4329                 if (rc != 0) {
4330                         fid_zero(&oic->oic_fid);
4331                         GOTO(out, rc);
4332                 }
4333
4334                 if (osd_remote_fid(env, dev, fid))
4335                         GOTO(out, rc = 0);
4336
4337                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4338                                       fid);
4339                 if (rc != 0)
4340                         GOTO(out, rc);
4341                 if ((scrub->os_pos_current <= ino) &&
4342                     ((sf->sf_flags & SF_INCONSISTENT) ||
4343                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4344                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4345                                       sf->sf_oi_bitmap)))
4346                         osd_consistency_check(oti, dev, oic);
4347         } else {
4348                 rc = -ENOENT;
4349         }
4350
4351         GOTO(out, rc);
4352
4353 out:
4354         if (hlock != NULL)
4355                 ldiskfs_htree_unlock(hlock);
4356         else
4357                 up_read(&obj->oo_ext_idx_sem);
4358         return rc;
4359 }
4360
4361 /**
4362  * Find the osd object for given fid.
4363  *
4364  * \param fid need to find the osd object having this fid
4365  *
4366  * \retval osd_object on success
4367  * \retval        -ve on error
4368  */
4369 struct osd_object *osd_object_find(const struct lu_env *env,
4370                                    struct dt_object *dt,
4371                                    const struct lu_fid *fid)
4372 {
4373         struct lu_device  *ludev = dt->do_lu.lo_dev;
4374         struct osd_object *child = NULL;
4375         struct lu_object  *luch;
4376         struct lu_object  *lo;
4377
4378         /*
4379          * at this point topdev might not exist yet
4380          * (i.e. MGS is preparing profiles). so we can
4381          * not rely on topdev and instead lookup with
4382          * our device passed as topdev. this can't work
4383          * if the object isn't cached yet (as osd doesn't
4384          * allocate lu_header). IOW, the object must be
4385          * in the cache, otherwise lu_object_alloc() crashes
4386          * -bzzz
4387          */
4388         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4389                                  ludev : ludev->ld_site->ls_top_dev,
4390                                  fid, NULL);
4391         if (!IS_ERR(luch)) {
4392                 if (lu_object_exists(luch)) {
4393                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4394                         if (lo != NULL)
4395                                 child = osd_obj(lo);
4396                         else
4397                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4398                                                 "lu_object can't be located"
4399                                                 DFID"\n", PFID(fid));
4400
4401                         if (child == NULL) {
4402                                 lu_object_put(env, luch);
4403                                 CERROR("Unable to get osd_object\n");
4404                                 child = ERR_PTR(-ENOENT);
4405                         }
4406                 } else {
4407                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4408                                         "lu_object does not exists "DFID"\n",
4409                                         PFID(fid));
4410                         lu_object_put(env, luch);
4411                         child = ERR_PTR(-ENOENT);
4412                 }
4413         } else {
4414                 child = ERR_CAST(luch);
4415         }
4416
4417         return child;
4418 }
4419
4420 /**
4421  * Put the osd object once done with it.
4422  *
4423  * \param obj osd object that needs to be put
4424  */
4425 static inline void osd_object_put(const struct lu_env *env,
4426                                   struct osd_object *obj)
4427 {
4428         lu_object_put(env, &obj->oo_dt.do_lu);
4429 }
4430
4431 static int osd_index_declare_ea_insert(const struct lu_env *env,
4432                                        struct dt_object *dt,
4433                                        const struct dt_rec *rec,
4434                                        const struct dt_key *key,
4435                                        struct thandle *handle)
4436 {
4437         struct osd_thandle      *oh;
4438         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4439         struct lu_fid           *fid = (struct lu_fid *)rec;
4440         int                     rc;
4441         ENTRY;
4442
4443         LASSERT(!dt_object_remote(dt));
4444         LASSERT(handle != NULL);
4445
4446         oh = container_of0(handle, struct osd_thandle, ot_super);
4447         LASSERT(oh->ot_handle == NULL);
4448
4449         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4450                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4451
4452         if (osd_dt_obj(dt)->oo_inode != NULL) {
4453                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4454
4455                 /* We ignore block quota on meta pool (MDTs), so needn't
4456                  * calculate how many blocks will be consumed by this index
4457                  * insert */
4458                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4459                                            i_gid_read(inode), 0, oh,
4460                                            osd_dt_obj(dt), true, NULL, false);
4461         }
4462
4463         if (fid == NULL)
4464                 RETURN(0);
4465
4466         rc = osd_remote_fid(env, osd, fid);
4467         if (rc <= 0)
4468                 RETURN(rc);
4469
4470         rc = 0;
4471
4472         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4473                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4474         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4475                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4476         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4477                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4478
4479         RETURN(rc);
4480 }
4481
4482 /**
4483  * Index add function for interoperability mode (b11826).
4484  * It will add the directory entry.This entry is needed to
4485  * maintain name->fid mapping.
4486  *
4487  * \param key it is key i.e. file entry to be inserted
4488  * \param rec it is value of given key i.e. fid
4489  *
4490  * \retval   0, on success
4491  * \retval -ve, on error
4492  */
4493 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4494                                const struct dt_rec *rec,
4495                                const struct dt_key *key, struct thandle *th,
4496                                struct lustre_capa *capa, int ignore_quota)
4497 {
4498         struct osd_object       *obj = osd_dt_obj(dt);
4499         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4500         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4501         const struct lu_fid     *fid    = rec1->rec_fid;
4502         const char              *name = (const char *)key;
4503         struct osd_thread_info  *oti   = osd_oti_get(env);
4504         struct osd_inode_id     *id    = &oti->oti_id;
4505         struct inode            *child_inode = NULL;
4506         struct osd_object       *child = NULL;
4507         int                     rc;
4508         ENTRY;
4509
4510         LASSERT(osd_invariant(obj));
4511         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4512         LASSERT(th != NULL);
4513
4514         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4515
4516         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4517                 RETURN(-EACCES);
4518
4519         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4520
4521         rc = osd_remote_fid(env, osd, fid);
4522         if (rc < 0) {
4523                 CERROR("%s: Can not find object "DFID" rc %d\n",
4524                        osd_name(osd), PFID(fid), rc);
4525                 RETURN(rc);
4526         }
4527
4528         if (rc == 1) {
4529                 /* Insert remote entry */
4530                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4531                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4532                         struct osd_thandle      *oh;
4533
4534                         /* If parent on remote MDT, we need put this object
4535                          * under AGENT */
4536                         oh = container_of(th, typeof(*oh), ot_super);
4537                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4538                         if (rc != 0) {
4539                                 CERROR("%s: add "DFID" error: rc = %d\n",
4540                                        osd_name(osd),
4541                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4542                                 RETURN(rc);
4543                         }
4544
4545                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4546                 } else {
4547                         child_inode = osd_create_local_agent_inode(env, osd,
4548                                         obj, fid, rec1->rec_type & S_IFMT, th);
4549                         if (IS_ERR(child_inode))
4550                                 RETURN(PTR_ERR(child_inode));
4551                 }
4552         } else {
4553                 /* Insert local entry */
4554                 child = osd_object_find(env, dt, fid);
4555                 if (IS_ERR(child)) {
4556                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4557                                osd_name(osd), PFID(fid),
4558                                id->oii_ino, id->oii_gen,
4559                                (int)PTR_ERR(child));
4560                         RETURN(PTR_ERR(child));
4561                 }
4562                 child_inode = igrab(child->oo_inode);
4563         }
4564
4565         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4566
4567         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4568                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4569
4570         iput(child_inode);
4571         if (child != NULL)
4572                 osd_object_put(env, child);
4573         LASSERT(osd_invariant(obj));
4574         RETURN(rc);
4575 }
4576
4577 /**
4578  *  Initialize osd Iterator for given osd index object.
4579  *
4580  *  \param  dt      osd index object
4581  */
4582
4583 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4584                                      struct dt_object *dt,
4585                                      __u32 unused,
4586                                      struct lustre_capa *capa)
4587 {
4588         struct osd_it_iam      *it;
4589         struct osd_thread_info *oti = osd_oti_get(env);
4590         struct osd_object      *obj = osd_dt_obj(dt);
4591         struct lu_object       *lo  = &dt->do_lu;
4592         struct iam_path_descr  *ipd;
4593         struct iam_container   *bag = &obj->oo_dir->od_container;
4594
4595         LASSERT(lu_object_exists(lo));
4596
4597         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4598                 return ERR_PTR(-EACCES);
4599
4600         it = &oti->oti_it;
4601         ipd = osd_it_ipd_get(env, bag);
4602         if (likely(ipd != NULL)) {
4603                 it->oi_obj = obj;
4604                 it->oi_ipd = ipd;
4605                 lu_object_get(lo);
4606                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4607                 return (struct dt_it *)it;
4608         }
4609         return ERR_PTR(-ENOMEM);
4610 }
4611
4612 /**
4613  * free given Iterator.
4614  */
4615
4616 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4617 {
4618         struct osd_it_iam *it = (struct osd_it_iam *)di;
4619         struct osd_object *obj = it->oi_obj;
4620
4621         iam_it_fini(&it->oi_it);
4622         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4623         lu_object_put(env, &obj->oo_dt.do_lu);
4624 }
4625
4626 /**
4627  *  Move Iterator to record specified by \a key
4628  *
4629  *  \param  di      osd iterator
4630  *  \param  key     key for index
4631  *
4632  *  \retval +ve  di points to record with least key not larger than key
4633  *  \retval  0   di points to exact matched key
4634  *  \retval -ve  failure
4635  */
4636
4637 static int osd_it_iam_get(const struct lu_env *env,
4638                           struct dt_it *di, const struct dt_key *key)
4639 {
4640         struct osd_thread_info  *oti = osd_oti_get(env);
4641         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4642
4643         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4644                 /* swab quota uid/gid */
4645                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4646                 key = (struct dt_key *)&oti->oti_quota_id;
4647         }
4648
4649         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4650 }
4651
4652 /**
4653  *  Release Iterator
4654  *
4655  *  \param  di      osd iterator
4656  */
4657 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4658 {
4659         struct osd_it_iam *it = (struct osd_it_iam *)di;
4660
4661         iam_it_put(&it->oi_it);
4662 }
4663
4664 /**
4665  *  Move iterator by one record
4666  *
4667  *  \param  di      osd iterator
4668  *
4669  *  \retval +1   end of container reached
4670  *  \retval  0   success
4671  *  \retval -ve  failure
4672  */
4673
4674 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4675 {
4676         struct osd_it_iam *it = (struct osd_it_iam *)di;
4677
4678         return iam_it_next(&it->oi_it);
4679 }
4680
4681 /**
4682  * Return pointer to the key under iterator.
4683  */
4684
4685 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4686                                  const struct dt_it *di)
4687 {
4688         struct osd_thread_info *oti = osd_oti_get(env);
4689         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4690         struct osd_object      *obj = it->oi_obj;
4691         struct dt_key          *key;
4692
4693         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4694
4695         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4696                 /* swab quota uid/gid */
4697                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4698                 key = (struct dt_key *)&oti->oti_quota_id;
4699         }
4700
4701         return key;
4702 }
4703
4704 /**
4705  * Return size of key under iterator (in bytes)
4706  */
4707
4708 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4709 {
4710         struct osd_it_iam *it = (struct osd_it_iam *)di;
4711
4712         return iam_it_key_size(&it->oi_it);
4713 }
4714
4715 static inline void
4716 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4717 {
4718         /* check if file type is required */
4719         if (ent->lde_attrs & LUDA_TYPE) {
4720                 struct luda_type *lt;
4721                 int align = sizeof(*lt) - 1;
4722
4723                 len = (len + align) & ~align;
4724                 lt = (struct luda_type *)(ent->lde_name + len);
4725                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4726         }
4727
4728         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4729 }
4730
4731 /**
4732  * build lu direct from backend fs dirent.
4733  */
4734
4735 static inline void
4736 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4737                    char *name, __u16 namelen, __u16 type, __u32 attr)
4738 {
4739         ent->lde_attrs = attr | LUDA_FID;
4740         fid_cpu_to_le(&ent->lde_fid, fid);
4741
4742         ent->lde_hash = cpu_to_le64(offset);
4743         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4744
4745         strncpy(ent->lde_name, name, namelen);
4746         ent->lde_name[namelen] = '\0';
4747         ent->lde_namelen = cpu_to_le16(namelen);
4748
4749         /* append lustre attributes */
4750         osd_it_append_attrs(ent, namelen, type);
4751 }
4752
4753 /**
4754  * Return pointer to the record under iterator.
4755  */
4756 static int osd_it_iam_rec(const struct lu_env *env,
4757                           const struct dt_it *di,
4758                           struct dt_rec *dtrec, __u32 attr)
4759 {
4760         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4761         struct osd_thread_info *info = osd_oti_get(env);
4762         ENTRY;
4763
4764         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4765                 const struct osd_fid_pack *rec;
4766                 struct lu_fid             *fid = &info->oti_fid;
4767                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4768                 char                      *name;
4769                 int                        namelen;
4770                 __u64                      hash;
4771                 int                        rc;
4772
4773                 name = (char *)iam_it_key_get(&it->oi_it);
4774                 if (IS_ERR(name))
4775                         RETURN(PTR_ERR(name));
4776
4777                 namelen = iam_it_key_size(&it->oi_it);
4778
4779                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4780                 if (IS_ERR(rec))
4781                         RETURN(PTR_ERR(rec));
4782
4783                 rc = osd_fid_unpack(fid, rec);
4784                 if (rc)
4785                         RETURN(rc);
4786
4787                 hash = iam_it_store(&it->oi_it);
4788
4789                 /* IAM does not store object type in IAM index (dir) */
4790                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4791                                    0, LUDA_FID);
4792         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4793                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4794                            (struct iam_rec *)dtrec);
4795                 osd_quota_unpack(it->oi_obj, dtrec);
4796         } else {
4797                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4798                            (struct iam_rec *)dtrec);
4799         }
4800
4801         RETURN(0);
4802 }
4803
4804 /**
4805  * Returns cookie for current Iterator position.
4806  */
4807 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4808 {
4809         struct osd_it_iam *it = (struct osd_it_iam *)di;
4810
4811         return iam_it_store(&it->oi_it);
4812 }
4813
4814 /**
4815  * Restore iterator from cookie.
4816  *
4817  * \param  di      osd iterator
4818  * \param  hash    Iterator location cookie
4819  *
4820  * \retval +ve  di points to record with least key not larger than key.
4821  * \retval  0   di points to exact matched key
4822  * \retval -ve  failure
4823  */
4824
4825 static int osd_it_iam_load(const struct lu_env *env,
4826                            const struct dt_it *di, __u64 hash)
4827 {
4828         struct osd_it_iam *it = (struct osd_it_iam *)di;
4829
4830         return iam_it_load(&it->oi_it, hash);
4831 }
4832
4833 static const struct dt_index_operations osd_index_iam_ops = {
4834         .dio_lookup         = osd_index_iam_lookup,
4835         .dio_declare_insert = osd_index_declare_iam_insert,
4836         .dio_insert         = osd_index_iam_insert,
4837         .dio_declare_delete = osd_index_declare_iam_delete,
4838         .dio_delete         = osd_index_iam_delete,
4839         .dio_it     = {
4840                 .init     = osd_it_iam_init,
4841                 .fini     = osd_it_iam_fini,
4842                 .get      = osd_it_iam_get,
4843                 .put      = osd_it_iam_put,
4844                 .next     = osd_it_iam_next,
4845                 .key      = osd_it_iam_key,
4846                 .key_size = osd_it_iam_key_size,
4847                 .rec      = osd_it_iam_rec,
4848                 .store    = osd_it_iam_store,
4849                 .load     = osd_it_iam_load
4850         }
4851 };
4852
4853
4854 /**
4855  * Creates or initializes iterator context.
4856  *
4857  * \retval struct osd_it_ea, iterator structure on success
4858  *
4859  */
4860 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4861                                     struct dt_object *dt,
4862                                     __u32 attr,
4863                                     struct lustre_capa *capa)
4864 {
4865         struct osd_object       *obj  = osd_dt_obj(dt);
4866         struct osd_thread_info  *info = osd_oti_get(env);
4867         struct osd_it_ea        *it   = &info->oti_it_ea;
4868         struct file             *file = &it->oie_file;
4869         struct lu_object        *lo   = &dt->do_lu;
4870         struct dentry           *obj_dentry = &info->oti_it_dentry;
4871         ENTRY;
4872         LASSERT(lu_object_exists(lo));
4873
4874         obj_dentry->d_inode = obj->oo_inode;
4875         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4876         obj_dentry->d_name.hash = 0;
4877
4878         it->oie_rd_dirent       = 0;
4879         it->oie_it_dirent       = 0;
4880         it->oie_dirent          = NULL;
4881         it->oie_buf             = info->oti_it_ea_buf;
4882         it->oie_obj             = obj;
4883
4884         /* Reset the "file" totally to avoid to reuse any old value from
4885          * former readdir handling, the "file->f_pos" should be zero. */
4886         memset(file, 0, sizeof(*file));
4887         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4888         if (attr & LUDA_64BITHASH)
4889                 file->f_mode    = FMODE_64BITHASH;
4890         else
4891                 file->f_mode    = FMODE_32BITHASH;
4892         file->f_dentry          = obj_dentry;
4893         file->f_mapping         = obj->oo_inode->i_mapping;
4894         file->f_op              = obj->oo_inode->i_fop;
4895         set_file_inode(file, obj->oo_inode);
4896
4897         lu_object_get(lo);
4898         RETURN((struct dt_it *) it);
4899 }
4900
4901 /**
4902  * Destroy or finishes iterator context.
4903  *
4904  * \param di iterator structure to be destroyed
4905  */
4906 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4907 {
4908         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4909         struct osd_object    *obj  = it->oie_obj;
4910         struct inode       *inode  = obj->oo_inode;
4911
4912         ENTRY;
4913         it->oie_file.f_op->release(inode, &it->oie_file);
4914         lu_object_put(env, &obj->oo_dt.do_lu);
4915         EXIT;
4916 }
4917
4918 /**
4919  * It position the iterator at given key, so that next lookup continues from
4920  * that key Or it is similar to dio_it->load() but based on a key,
4921  * rather than file position.
4922  *
4923  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4924  * to the beginning.
4925  *
4926  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4927  */
4928 static int osd_it_ea_get(const struct lu_env *env,
4929                          struct dt_it *di, const struct dt_key *key)
4930 {
4931         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4932
4933         ENTRY;
4934         LASSERT(((const char *)key)[0] == '\0');
4935         it->oie_file.f_pos      = 0;
4936         it->oie_rd_dirent       = 0;
4937         it->oie_it_dirent       = 0;
4938         it->oie_dirent          = NULL;
4939
4940         RETURN(+1);
4941 }
4942
4943 /**
4944  * Does nothing
4945  */
4946 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4947 {
4948 }
4949
4950 struct osd_filldir_cbs {
4951 #ifdef HAVE_DIR_CONTEXT
4952         struct dir_context ctx;
4953 #endif
4954         struct osd_it_ea  *it;
4955 };
4956 /**
4957  * It is called internally by ->readdir(). It fills the
4958  * iterator's in-memory data structure with required
4959  * information i.e. name, namelen, rec_size etc.
4960  *
4961  * \param buf in which information to be filled in.
4962  * \param name name of the file in given dir
4963  *
4964  * \retval 0 on success
4965  * \retval 1 on buffer full
4966  */
4967 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4968                                loff_t offset, __u64 ino,
4969                                unsigned d_type)
4970 {
4971         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4972         struct osd_object       *obj  = it->oie_obj;
4973         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4974         struct lu_fid           *fid  = &ent->oied_fid;
4975         struct osd_fid_pack     *rec;
4976         ENTRY;
4977
4978         /* this should never happen */
4979         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4980                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4981                 RETURN(-EIO);
4982         }
4983
4984         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4985             OSD_IT_EA_BUFSIZE)
4986                 RETURN(1);
4987
4988         /* "." is just the object itself. */
4989         if (namelen == 1 && name[0] == '.') {
4990                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4991         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4992                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4993                 if (osd_fid_unpack(fid, rec) != 0)
4994                         fid_zero(fid);
4995         } else {
4996                 fid_zero(fid);
4997         }
4998         d_type &= ~LDISKFS_DIRENT_LUFID;
4999
5000         /* NOT export local root. */
5001         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5002                 ino = obj->oo_inode->i_ino;
5003                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5004         }
5005
5006         ent->oied_ino     = ino;
5007         ent->oied_off     = offset;
5008         ent->oied_namelen = namelen;
5009         ent->oied_type    = d_type;
5010
5011         memcpy(ent->oied_name, name, namelen);
5012
5013         it->oie_rd_dirent++;
5014         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5015         RETURN(0);
5016 }
5017
5018 /**
5019  * Calls ->readdir() to load a directory entry at a time
5020  * and stored it in iterator's in-memory data structure.
5021  *
5022  * \param di iterator's in memory structure
5023  *
5024  * \retval   0 on success
5025  * \retval -ve on error
5026  * \retval +1 reach the end of entry
5027  */
5028 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5029                                const struct dt_it *di)
5030 {
5031         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5032         struct osd_object  *obj   = it->oie_obj;
5033         struct inode       *inode = obj->oo_inode;
5034         struct htree_lock  *hlock = NULL;
5035         struct file        *filp  = &it->oie_file;
5036         int                 rc = 0;
5037         struct osd_filldir_cbs buf = {
5038 #ifdef HAVE_DIR_CONTEXT
5039                 .ctx.actor = osd_ldiskfs_filldir,
5040 #endif
5041                 .it = it
5042         };
5043
5044         ENTRY;
5045         it->oie_dirent = it->oie_buf;
5046         it->oie_rd_dirent = 0;
5047
5048         if (obj->oo_hl_head != NULL) {
5049                 hlock = osd_oti_get(env)->oti_hlock;
5050                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5051                                    inode, LDISKFS_HLOCK_READDIR);
5052         } else {
5053                 down_read(&obj->oo_ext_idx_sem);
5054         }
5055
5056 #ifdef HAVE_DIR_CONTEXT
5057         buf.ctx.pos = filp->f_pos;
5058         rc = inode->i_fop->iterate(filp, &buf.ctx);
5059         filp->f_pos = buf.ctx.pos;
5060 #else
5061         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5062 #endif
5063
5064         if (hlock != NULL)
5065                 ldiskfs_htree_unlock(hlock);
5066         else
5067                 up_read(&obj->oo_ext_idx_sem);
5068
5069         if (it->oie_rd_dirent == 0) {
5070                 /*If it does not get any dirent, it means it has been reached
5071                  *to the end of the dir */
5072                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5073                 if (rc == 0)
5074                         rc = 1;
5075         } else {
5076                 it->oie_dirent = it->oie_buf;
5077                 it->oie_it_dirent = 1;
5078         }
5079
5080         RETURN(rc);
5081 }
5082
5083 /**
5084  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5085  * to load a directory entry at a time and stored it in
5086  * iterator's in-memory data structure.
5087  *
5088  * \param di iterator's in memory structure
5089  *
5090  * \retval +ve iterator reached to end
5091  * \retval   0 iterator not reached to end
5092  * \retval -ve on error
5093  */
5094 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5095 {
5096         struct osd_it_ea *it = (struct osd_it_ea *)di;
5097         int rc;
5098
5099         ENTRY;
5100
5101         if (it->oie_it_dirent < it->oie_rd_dirent) {
5102                 it->oie_dirent =
5103                         (void *) it->oie_dirent +
5104                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5105                                        it->oie_dirent->oied_namelen);
5106                 it->oie_it_dirent++;
5107                 RETURN(0);
5108         } else {
5109                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5110                         rc = +1;
5111                 else
5112                         rc = osd_ldiskfs_it_fill(env, di);
5113         }
5114
5115         RETURN(rc);
5116 }
5117
5118 /**
5119  * Returns the key at current position from iterator's in memory structure.
5120  *
5121  * \param di iterator's in memory structure
5122  *
5123  * \retval key i.e. struct dt_key on success
5124  */
5125 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5126                                     const struct dt_it *di)
5127 {
5128         struct osd_it_ea *it = (struct osd_it_ea *)di;
5129
5130         return (struct dt_key *)it->oie_dirent->oied_name;
5131 }
5132
5133 /**
5134  * Returns the key's size at current position from iterator's in memory structure.
5135  *
5136  * \param di iterator's in memory structure
5137  *
5138  * \retval key_size i.e. struct dt_key on success
5139  */
5140 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5141 {
5142         struct osd_it_ea *it = (struct osd_it_ea *)di;
5143
5144         return it->oie_dirent->oied_namelen;
5145 }
5146
5147 static int
5148 osd_dirent_update(handle_t *jh, struct super_block *sb,
5149                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5150                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5151 {
5152         struct osd_fid_pack *rec;
5153         int                  rc;
5154         ENTRY;
5155
5156         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5157         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5158
5159         rc = ldiskfs_journal_get_write_access(jh, bh);
5160         if (rc != 0)
5161                 RETURN(rc);
5162
5163         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5164         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5165         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5166
5167         RETURN(rc);
5168 }
5169
5170 static inline int
5171 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5172 {
5173         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5174             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5175                 return 1;
5176         else
5177                 return 0;
5178 }
5179
5180 static inline int
5181 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5182 {
5183         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5184                  "dot_dotdot = %d\n", dot_dotdot);
5185
5186         if (LDISKFS_DIR_REC_LEN(de) >=
5187             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5188                 return 1;
5189         else
5190                 return 0;
5191 }
5192
5193 static int
5194 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5195                     struct inode *dir, struct inode *inode,
5196                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5197                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5198                     struct htree_lock *hlock)
5199 {
5200         struct dentry               *dentry;
5201         struct osd_fid_pack         *rec;
5202         struct ldiskfs_dentry_param *ldp;
5203         int                          rc;
5204         ENTRY;
5205
5206         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5207                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5208                 RETURN(0);
5209
5210         /* There is enough space to hold the FID-in-dirent. */
5211         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5212                                  dir->i_sb->s_blocksize)) {
5213                 rc = ldiskfs_journal_get_write_access(jh, bh);
5214                 if (rc != 0)
5215                         RETURN(rc);
5216
5217                 de->name[de->name_len] = 0;
5218                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5219                 rec->fp_len = sizeof(struct lu_fid) + 1;
5220                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5221                 de->file_type |= LDISKFS_DIRENT_LUFID;
5222
5223                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5224
5225                 RETURN(rc);
5226         }
5227
5228         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5229         if (rc != 0)
5230                 RETURN(rc);
5231
5232         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5233                                            ent->oied_namelen);
5234         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5235         osd_get_ldiskfs_dirent_param(ldp, fid);
5236         dentry->d_fsdata = (void *)ldp;
5237         ll_vfs_dq_init(dir);
5238         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5239         /* It is too bad, we cannot reinsert the name entry back.
5240          * That means we lose it! */
5241         if (rc != 0)
5242                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5243                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5244                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5245                        dir->i_ino, dir->i_generation,
5246                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5247
5248         RETURN(rc);
5249 }
5250
5251 static int
5252 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5253                         struct osd_it_ea *it, struct lu_fid *fid,
5254                         struct osd_inode_id *id, __u32 *attr)
5255 {
5256         struct osd_thread_info     *info        = osd_oti_get(env);
5257         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5258         struct osd_device          *dev         = osd_obj2dev(obj);
5259         struct super_block         *sb          = osd_sb(dev);
5260         const char                 *devname     =
5261                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5262         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5263         struct inode               *dir         = obj->oo_inode;
5264         struct htree_lock          *hlock       = NULL;
5265         struct buffer_head         *bh          = NULL;
5266         handle_t                   *jh          = NULL;
5267         struct ldiskfs_dir_entry_2 *de;
5268         struct dentry              *dentry;
5269         struct inode               *inode;
5270         int                         credits;
5271         int                         rc;
5272         int                         dot_dotdot  = 0;
5273         bool                        dirty       = false;
5274         ENTRY;
5275
5276         if (ent->oied_name[0] == '.') {
5277                 if (ent->oied_namelen == 1)
5278                         dot_dotdot = 1;
5279                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5280                         dot_dotdot = 2;
5281         }
5282
5283         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5284                                       ent->oied_namelen);
5285
5286         /* We need to ensure that the name entry is still valid.
5287          * Because it may be removed or renamed by other already.
5288          *
5289          * The unlink or rename operation will start journal before PDO lock,
5290          * so to avoid deadlock, here we need to start journal handle before
5291          * related PDO lock also. But because we do not know whether there
5292          * will be something to be repaired before PDO lock, we just start
5293          * journal without conditions.
5294          *
5295          * We may need to remove the name entry firstly, then insert back.
5296          * One credit is for user quota file update.
5297          * One credit is for group quota file update.
5298          * Two credits are for dirty inode. */
5299         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5300                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5301
5302 again:
5303         if (dev->od_dirent_journal) {
5304                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5305                 if (IS_ERR(jh)) {
5306                         rc = PTR_ERR(jh);
5307                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5308                                "check_repair, dir = %lu/%u, credits = %d, "
5309                                "name = %.*s: rc = %d\n",
5310                                devname, dir->i_ino, dir->i_generation, credits,
5311                                ent->oied_namelen, ent->oied_name, rc);
5312                         RETURN(rc);
5313                 }
5314
5315                 if (obj->oo_hl_head != NULL) {
5316                         hlock = osd_oti_get(env)->oti_hlock;
5317                         /* "0" means exclusive lock for the whole directory.
5318                          * We need to prevent others access such name entry
5319                          * during the delete + insert. Neither HLOCK_ADD nor
5320                          * HLOCK_DEL cannot guarantee the atomicity. */
5321                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5322                 } else {
5323                         down_write(&obj->oo_ext_idx_sem);
5324                 }
5325         } else {
5326                 if (obj->oo_hl_head != NULL) {
5327                         hlock = osd_oti_get(env)->oti_hlock;
5328                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5329                                            LDISKFS_HLOCK_LOOKUP);
5330                 } else {
5331                         down_read(&obj->oo_ext_idx_sem);
5332                 }
5333         }
5334
5335         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5336         /* For dot/dotdot entry, if there is not enough space to hold the
5337          * FID-in-dirent, just keep them there. It only happens when the
5338          * device upgraded from 1.8 or restored from MDT file-level backup.
5339          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5340          * and needs to get FID from LMA when readdir, it will not affect the
5341          * performance much. */
5342         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5343             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5344                 *attr |= LUDA_IGNORE;
5345                 GOTO(out_journal, rc = 0);
5346         }
5347
5348         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5349         inode = osd_iget(info, dev, id);
5350         if (IS_ERR(inode)) {
5351                 rc = PTR_ERR(inode);
5352                 if (rc == -ENOENT || rc == -ESTALE)
5353                         rc = 1;
5354                 else
5355                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5356                                "check_repair, dir = %lu/%u, name = %.*s: "
5357                                "rc = %d\n",
5358                                devname, dir->i_ino, dir->i_generation,
5359                                ent->oied_namelen, ent->oied_name, rc);
5360
5361                 GOTO(out_journal, rc);
5362         }
5363
5364         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5365         if (rc == 0) {
5366                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5367
5368                 if (fid_is_sane(fid)) {
5369                         /* FID-in-dirent is valid. */
5370                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5371                                 GOTO(out_inode, rc = 0);
5372
5373                         /* Do not repair under dryrun mode. */
5374                         if (*attr & LUDA_VERIFY_DRYRUN) {
5375                                 *attr |= LUDA_REPAIR;
5376                                 GOTO(out_inode, rc = 0);
5377                         }
5378
5379                         if (!dev->od_dirent_journal) {
5380                                 iput(inode);
5381                                 brelse(bh);
5382                                 if (hlock != NULL)
5383                                         ldiskfs_htree_unlock(hlock);
5384                                 else
5385                                         up_read(&obj->oo_ext_idx_sem);
5386                                 dev->od_dirent_journal = 1;
5387                                 goto again;
5388                         }
5389
5390                         *fid = lma->lma_self_fid;
5391                         dirty = true;
5392                         /* Update the FID-in-dirent. */
5393                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5394                         if (rc == 0)
5395                                 *attr |= LUDA_REPAIR;
5396                         else
5397                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5398                                        "in the dirent, dir = %lu/%u, "
5399                                        "name = %.*s, "DFID": rc = %d\n",
5400                                        devname, dir->i_ino, dir->i_generation,
5401                                        ent->oied_namelen, ent->oied_name,
5402                                        PFID(fid), rc);
5403                 } else {
5404                         /* Do not repair under dryrun mode. */
5405                         if (*attr & LUDA_VERIFY_DRYRUN) {
5406                                 *fid = lma->lma_self_fid;
5407                                 *attr |= LUDA_REPAIR;
5408                                 GOTO(out_inode, rc = 0);
5409                         }
5410
5411                         if (!dev->od_dirent_journal) {
5412                                 iput(inode);
5413                                 brelse(bh);
5414                                 if (hlock != NULL)
5415                                         ldiskfs_htree_unlock(hlock);
5416                                 else
5417                                         up_read(&obj->oo_ext_idx_sem);
5418                                 dev->od_dirent_journal = 1;
5419                                 goto again;
5420                         }
5421
5422                         *fid = lma->lma_self_fid;
5423                         dirty = true;
5424                         /* Append the FID-in-dirent. */
5425                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5426                                                  fid, bh, de, hlock);
5427                         if (rc == 0)
5428                                 *attr |= LUDA_REPAIR;
5429                         else
5430                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5431                                        "after the dirent, dir = %lu/%u, "
5432                                        "name = %.*s, "DFID": rc = %d\n",
5433                                        devname, dir->i_ino, dir->i_generation,
5434                                        ent->oied_namelen, ent->oied_name,
5435                                        PFID(fid), rc);
5436                 }
5437         } else if (rc == -ENODATA) {
5438                 /* Do not repair under dryrun mode. */
5439                 if (*attr & LUDA_VERIFY_DRYRUN) {
5440                         if (fid_is_sane(fid)) {
5441                                 *attr |= LUDA_REPAIR;
5442                         } else {
5443                                 lu_igif_build(fid, inode->i_ino,
5444                                               inode->i_generation);
5445                                 *attr |= LUDA_UPGRADE;
5446                         }
5447                         GOTO(out_inode, rc = 0);
5448                 }
5449
5450                 if (!dev->od_dirent_journal) {
5451                         iput(inode);
5452                         brelse(bh);
5453                         if (hlock != NULL)
5454                                 ldiskfs_htree_unlock(hlock);
5455                         else
5456                                 up_read(&obj->oo_ext_idx_sem);
5457                         dev->od_dirent_journal = 1;
5458                         goto again;
5459                 }
5460
5461                 dirty = true;
5462                 if (unlikely(fid_is_sane(fid))) {
5463                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5464                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5465                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5466                         if (rc == 0)
5467                                 *attr |= LUDA_REPAIR;
5468                         else
5469                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5470                                        "update dirent, dir = %lu/%u, "
5471                                        "name = %.*s, "DFID": rc = %d\n",
5472                                        devname, dir->i_ino, dir->i_generation,
5473                                        ent->oied_namelen, ent->oied_name,
5474                                        PFID(fid), rc);
5475                 } else {
5476                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5477                         /* It is probably IGIF object. Only aappend the
5478                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5479                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5480                                                  fid, bh, de, hlock);
5481                         if (rc == 0)
5482                                 *attr |= LUDA_UPGRADE;
5483                         else
5484                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5485                                        "after the dirent, dir = %lu/%u, "
5486                                        "name = %.*s, "DFID": rc = %d\n",
5487                                        devname, dir->i_ino, dir->i_generation,
5488                                        ent->oied_namelen, ent->oied_name,
5489                                        PFID(fid), rc);
5490                 }
5491         }
5492
5493         GOTO(out_inode, rc);
5494
5495 out_inode:
5496         iput(inode);
5497
5498 out_journal:
5499         brelse(bh);
5500         if (hlock != NULL) {
5501                 ldiskfs_htree_unlock(hlock);
5502         } else {
5503                 if (dev->od_dirent_journal)
5504                         up_write(&obj->oo_ext_idx_sem);
5505                 else
5506                         up_read(&obj->oo_ext_idx_sem);
5507         }
5508         if (jh != NULL)
5509                 ldiskfs_journal_stop(jh);
5510         if (rc >= 0 && !dirty)
5511                 dev->od_dirent_journal = 0;
5512         return rc;
5513 }
5514
5515 /**
5516  * Returns the value at current position from iterator's in memory structure.
5517  *
5518  * \param di struct osd_it_ea, iterator's in memory structure
5519  * \param attr attr requested for dirent.
5520  * \param lde lustre dirent
5521  *
5522  * \retval   0 no error and \param lde has correct lustre dirent.
5523  * \retval -ve on error
5524  */
5525 static inline int osd_it_ea_rec(const struct lu_env *env,
5526                                 const struct dt_it *di,
5527                                 struct dt_rec *dtrec, __u32 attr)
5528 {
5529         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5530         struct osd_object      *obj   = it->oie_obj;
5531         struct osd_device      *dev   = osd_obj2dev(obj);
5532         struct osd_scrub       *scrub = &dev->od_scrub;
5533         struct scrub_file      *sf    = &scrub->os_file;
5534         struct osd_thread_info *oti   = osd_oti_get(env);
5535         struct osd_inode_id    *id    = &oti->oti_id;
5536         struct osd_idmap_cache *oic   = &oti->oti_cache;
5537         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5538         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5539         __u32                   ino   = it->oie_dirent->oied_ino;
5540         int                     rc    = 0;
5541         ENTRY;
5542
5543         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5544
5545         if (attr & LUDA_VERIFY) {
5546                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5547                         attr |= LUDA_IGNORE;
5548                         /* If the parent is on remote MDT, and there
5549                          * is no FID-in-dirent, then we have to get
5550                          * the parent FID from the linkEA.  */
5551                         osd_get_pfid_from_linkea(env, obj, fid);
5552                 } else {
5553                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5554                                                      &attr);
5555                 }
5556         } else {
5557                 attr &= ~LU_DIRENT_ATTRS_MASK;
5558                 if (!fid_is_sane(fid)) {
5559                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5560                             likely(it->oie_dirent->oied_namelen != 2 ||
5561                                    it->oie_dirent->oied_name[0] != '.' ||
5562                                    it->oie_dirent->oied_name[1] != '.'))
5563                                 RETURN(-ENOENT);
5564
5565                         if (unlikely(ino == osd_remote_parent_ino(dev)))
5566                                 /* If the parent is on remote MDT, and there
5567                                  * is no FID-in-dirent, then we have to get
5568                                  * the parent FID from the linkEA.  */
5569                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5570                         else
5571                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5572                 } else {
5573                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5574                 }
5575         }
5576
5577         /* Pack the entry anyway, at least the offset is right. */
5578         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5579                            it->oie_dirent->oied_name,
5580                            it->oie_dirent->oied_namelen,
5581                            it->oie_dirent->oied_type, attr);
5582
5583         if (rc < 0)
5584                 RETURN(rc);
5585
5586         if (osd_remote_fid(env, dev, fid))
5587                 RETURN(0);
5588
5589         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5590                 rc = osd_add_oi_cache(oti, dev, id, fid);
5591
5592         if (!(attr & LUDA_VERIFY) &&
5593             (scrub->os_pos_current <= ino) &&
5594             ((sf->sf_flags & SF_INCONSISTENT) ||
5595              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5596              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5597                 osd_consistency_check(oti, dev, oic);
5598
5599         RETURN(rc > 0 ? 0 : rc);
5600 }
5601
5602 /**
5603  * Returns the record size size at current position.
5604  *
5605  * This function will return record(lu_dirent) size in bytes.
5606  *
5607  * \param[in] env       execution environment
5608  * \param[in] di        iterator's in memory structure
5609  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5610  *                      calculate the lu_dirent size.
5611  *
5612  * \retval      record size(in bytes & in memory) of the current lu_dirent
5613  *              entry.
5614  */
5615 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5616                               __u32 attr)
5617 {
5618         struct osd_it_ea *it = (struct osd_it_ea *)di;
5619
5620         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5621 }
5622
5623 /**
5624  * Returns a cookie for current position of the iterator head, so that
5625  * user can use this cookie to load/start the iterator next time.
5626  *
5627  * \param di iterator's in memory structure
5628  *
5629  * \retval cookie for current position, on success
5630  */
5631 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5632 {
5633         struct osd_it_ea *it = (struct osd_it_ea *)di;
5634
5635         return it->oie_dirent->oied_off;
5636 }
5637
5638 /**
5639  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5640  * to load a directory entry at a time and stored it i inn,
5641  * in iterator's in-memory data structure.
5642  *
5643  * \param di struct osd_it_ea, iterator's in memory structure
5644  *
5645  * \retval +ve on success
5646  * \retval -ve on error
5647  */
5648 static int osd_it_ea_load(const struct lu_env *env,
5649                           const struct dt_it *di, __u64 hash)
5650 {
5651         struct osd_it_ea *it = (struct osd_it_ea *)di;
5652         int rc;
5653
5654         ENTRY;
5655         it->oie_file.f_pos = hash;
5656
5657         rc =  osd_ldiskfs_it_fill(env, di);
5658         if (rc > 0)
5659                 rc = -ENODATA;
5660
5661         if (rc == 0)
5662                 rc = +1;
5663
5664         RETURN(rc);
5665 }
5666
5667 /**
5668  * Index lookup function for interoperability mode (b11826).
5669  *
5670  * \param key,  key i.e. file name to be searched
5671  *
5672  * \retval +ve, on success
5673  * \retval -ve, on error
5674  */
5675 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5676                                struct dt_rec *rec, const struct dt_key *key,
5677                                struct lustre_capa *capa)
5678 {
5679         struct osd_object *obj = osd_dt_obj(dt);
5680         int rc = 0;
5681
5682         ENTRY;
5683
5684         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5685         LINVRNT(osd_invariant(obj));
5686
5687         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5688                 return -EACCES;
5689
5690         rc = osd_ea_lookup_rec(env, obj, rec, key);
5691         if (rc == 0)
5692                 rc = +1;
5693         RETURN(rc);
5694 }
5695
5696 /**
5697  * Index and Iterator operations for interoperability
5698  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5699  */
5700 static const struct dt_index_operations osd_index_ea_ops = {
5701         .dio_lookup         = osd_index_ea_lookup,
5702         .dio_declare_insert = osd_index_declare_ea_insert,
5703         .dio_insert         = osd_index_ea_insert,
5704         .dio_declare_delete = osd_index_declare_ea_delete,
5705         .dio_delete         = osd_index_ea_delete,
5706         .dio_it     = {
5707                 .init     = osd_it_ea_init,
5708                 .fini     = osd_it_ea_fini,
5709                 .get      = osd_it_ea_get,
5710                 .put      = osd_it_ea_put,
5711                 .next     = osd_it_ea_next,
5712                 .key      = osd_it_ea_key,
5713                 .key_size = osd_it_ea_key_size,
5714                 .rec      = osd_it_ea_rec,
5715                 .rec_size = osd_it_ea_rec_size,
5716                 .store    = osd_it_ea_store,
5717                 .load     = osd_it_ea_load
5718         }
5719 };
5720
5721 static void *osd_key_init(const struct lu_context *ctx,
5722                           struct lu_context_key *key)
5723 {
5724         struct osd_thread_info *info;
5725
5726         OBD_ALLOC_PTR(info);
5727         if (info == NULL)
5728                 return ERR_PTR(-ENOMEM);
5729
5730         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5731         if (info->oti_it_ea_buf == NULL)
5732                 goto out_free_info;
5733
5734         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5735
5736         info->oti_hlock = ldiskfs_htree_lock_alloc();
5737         if (info->oti_hlock == NULL)
5738                 goto out_free_ea;
5739
5740         return info;
5741
5742  out_free_ea:
5743         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5744  out_free_info:
5745         OBD_FREE_PTR(info);
5746         return ERR_PTR(-ENOMEM);
5747 }
5748
5749 static void osd_key_fini(const struct lu_context *ctx,
5750                          struct lu_context_key *key, void* data)
5751 {
5752         struct osd_thread_info *info = data;
5753
5754         if (info->oti_hlock != NULL)
5755                 ldiskfs_htree_lock_free(info->oti_hlock);
5756         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5757         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5758         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5759         lu_buf_free(&info->oti_big_buf);
5760         OBD_FREE_PTR(info);
5761 }
5762
5763 static void osd_key_exit(const struct lu_context *ctx,
5764                          struct lu_context_key *key, void *data)
5765 {
5766         struct osd_thread_info *info = data;
5767
5768         LASSERT(info->oti_r_locks == 0);
5769         LASSERT(info->oti_w_locks == 0);
5770         LASSERT(info->oti_txns    == 0);
5771 }
5772
5773 /* type constructor/destructor: osd_type_init, osd_type_fini */
5774 LU_TYPE_INIT_FINI(osd, &osd_key);
5775
5776 struct lu_context_key osd_key = {
5777         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5778         .lct_init = osd_key_init,
5779         .lct_fini = osd_key_fini,
5780         .lct_exit = osd_key_exit
5781 };
5782
5783
5784 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5785                            const char *name, struct lu_device *next)
5786 {
5787         struct osd_device *osd = osd_dev(d);
5788
5789         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5790             >= sizeof(osd->od_svname))
5791                 return -E2BIG;
5792         return osd_procfs_init(osd, name);
5793 }
5794
5795 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5796 {
5797         struct seq_server_site  *ss = osd_seq_site(osd);
5798         int                     rc;
5799         ENTRY;
5800
5801         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5802                 RETURN(0);
5803
5804         if (unlikely(ss == NULL))
5805                 RETURN(-ENODEV);
5806
5807         OBD_ALLOC_PTR(osd->od_cl_seq);
5808         if (osd->od_cl_seq == NULL)
5809                 RETURN(-ENOMEM);
5810
5811         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5812                              osd->od_svname, ss->ss_server_seq);
5813
5814         if (rc != 0) {
5815                 OBD_FREE_PTR(osd->od_cl_seq);
5816                 osd->od_cl_seq = NULL;
5817         }
5818
5819         RETURN(rc);
5820 }
5821
5822 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5823 {
5824         if (osd->od_cl_seq == NULL)
5825                 return;
5826
5827         seq_client_fini(osd->od_cl_seq);
5828         OBD_FREE_PTR(osd->od_cl_seq);
5829         osd->od_cl_seq = NULL;
5830 }
5831
5832 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5833 {
5834         ENTRY;
5835
5836         /* shutdown quota slave instance associated with the device */
5837         if (o->od_quota_slave != NULL) {
5838                 qsd_fini(env, o->od_quota_slave);
5839                 o->od_quota_slave = NULL;
5840         }
5841
5842         osd_fid_fini(env, o);
5843
5844         RETURN(0);
5845 }
5846
5847 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5848 {
5849         ENTRY;
5850
5851         if (o->od_mnt != NULL) {
5852                 shrink_dcache_sb(osd_sb(o));
5853                 osd_sync(env, &o->od_dt_dev);
5854
5855                 mntput(o->od_mnt);
5856                 o->od_mnt = NULL;
5857         }
5858
5859         EXIT;
5860 }
5861
5862 static int osd_mount(const struct lu_env *env,
5863                      struct osd_device *o, struct lustre_cfg *cfg)
5864 {
5865         const char              *name  = lustre_cfg_string(cfg, 0);
5866         const char              *dev  = lustre_cfg_string(cfg, 1);
5867         const char              *opts;
5868         unsigned long            page, s_flags, lmd_flags = 0;
5869         struct page             *__page;
5870         struct file_system_type *type;
5871         char                    *options = NULL;
5872         char                    *str;
5873         struct osd_thread_info  *info = osd_oti_get(env);
5874         struct lu_fid           *fid = &info->oti_fid;
5875         struct inode            *inode;
5876         int                      rc = 0;
5877         ENTRY;
5878
5879         if (o->od_mnt != NULL)
5880                 RETURN(0);
5881
5882         if (strlen(dev) >= sizeof(o->od_mntdev))
5883                 RETURN(-E2BIG);
5884         strcpy(o->od_mntdev, dev);
5885
5886         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5887         if (__page == NULL)
5888                 GOTO(out, rc = -ENOMEM);
5889
5890         str = lustre_cfg_string(cfg, 2);
5891         s_flags = simple_strtoul(str, NULL, 0);
5892         str = strstr(str, ":");
5893         if (str)
5894                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5895         opts = lustre_cfg_string(cfg, 3);
5896         page = (unsigned long)page_address(__page);
5897         options = (char *)page;
5898         *options = '\0';
5899         if (opts == NULL)
5900                 strcat(options, "user_xattr,acl");
5901         else
5902                 strcat(options, opts);
5903
5904         /* Glom up mount options */
5905         if (*options != '\0')
5906                 strcat(options, ",");
5907         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5908
5909         type = get_fs_type("ldiskfs");
5910         if (!type) {
5911                 CERROR("%s: cannot find ldiskfs module\n", name);
5912                 GOTO(out, rc = -ENODEV);
5913         }
5914
5915         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5916         module_put(type->owner);
5917
5918         if (IS_ERR(o->od_mnt)) {
5919                 rc = PTR_ERR(o->od_mnt);
5920                 o->od_mnt = NULL;
5921                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5922                 GOTO(out, rc);
5923         }
5924
5925 #ifdef HAVE_DEV_SET_RDONLY
5926         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5927                 CERROR("%s: underlying device %s is marked as read-only. "
5928                        "Setup failed\n", name, dev);
5929                 GOTO(out_mnt, rc = -EROFS);
5930         }
5931 #endif
5932
5933         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5934                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5935                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5936                 GOTO(out_mnt, rc = -EINVAL);
5937         }
5938
5939 #ifdef LDISKFS_MOUNT_DIRDATA
5940         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5941                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5942                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5943         else if (!o->od_is_ost)
5944                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5945                       "enabling the dirdata feature. If you do not want to "
5946                       "downgrade to Lustre-1.x again, you can enable it via "
5947                       "'tune2fs -O dirdata device'\n", name, dev);
5948 #endif
5949         inode = osd_sb(o)->s_root->d_inode;
5950         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5951         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5952         if (rc != 0) {
5953                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5954                 GOTO(out_mnt, rc);
5955         }
5956
5957         if (lmd_flags & LMD_FLG_NOSCRUB)
5958                 o->od_noscrub = 1;
5959
5960         GOTO(out, rc = 0);
5961
5962 out_mnt:
5963         mntput(o->od_mnt);
5964         o->od_mnt = NULL;
5965
5966 out:
5967         if (__page)
5968                 OBD_PAGE_FREE(__page);
5969
5970         return rc;
5971 }
5972
5973 static struct lu_device *osd_device_fini(const struct lu_env *env,
5974                                          struct lu_device *d)
5975 {
5976         struct osd_device *o = osd_dev(d);
5977         ENTRY;
5978
5979         osd_shutdown(env, o);
5980         osd_procfs_fini(o);
5981         osd_scrub_cleanup(env, o);
5982         osd_obj_map_fini(o);
5983         osd_umount(env, o);
5984
5985         RETURN(NULL);
5986 }
5987
5988 static int osd_device_init0(const struct lu_env *env,
5989                             struct osd_device *o,
5990                             struct lustre_cfg *cfg)
5991 {
5992         struct lu_device        *l = osd2lu_dev(o);
5993         struct osd_thread_info *info;
5994         int                     rc;
5995         int                     cplen = 0;
5996
5997         /* if the module was re-loaded, env can loose its keys */
5998         rc = lu_env_refill((struct lu_env *) env);
5999         if (rc)
6000                 GOTO(out, rc);
6001         info = osd_oti_get(env);
6002         LASSERT(info);
6003
6004         l->ld_ops = &osd_lu_ops;
6005         o->od_dt_dev.dd_ops = &osd_dt_ops;
6006
6007         spin_lock_init(&o->od_osfs_lock);
6008         mutex_init(&o->od_otable_mutex);
6009
6010         o->od_capa_hash = init_capa_hash();
6011         if (o->od_capa_hash == NULL)
6012                 GOTO(out, rc = -ENOMEM);
6013
6014         o->od_read_cache = 1;
6015         o->od_writethrough_cache = 1;
6016         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6017
6018         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6019                         sizeof(o->od_svname));
6020         if (cplen >= sizeof(o->od_svname)) {
6021                 rc = -E2BIG;
6022                 GOTO(out_capa, rc);
6023         }
6024
6025         if (server_name_is_ost(o->od_svname))
6026                 o->od_is_ost = 1;
6027
6028         rc = osd_mount(env, o, cfg);
6029         if (rc != 0)
6030                 GOTO(out_capa, rc);
6031
6032         rc = osd_obj_map_init(env, o);
6033         if (rc != 0)
6034                 GOTO(out_mnt, rc);
6035
6036         rc = lu_site_init(&o->od_site, l);
6037         if (rc != 0)
6038                 GOTO(out_compat, rc);
6039         o->od_site.ls_bottom_dev = l;
6040
6041         rc = lu_site_init_finish(&o->od_site);
6042         if (rc != 0)
6043                 GOTO(out_site, rc);
6044
6045         /* self-repair LMA by default */
6046         o->od_lma_self_repair = 1;
6047
6048         INIT_LIST_HEAD(&o->od_ios_list);
6049         /* setup scrub, including OI files initialization */
6050         rc = osd_scrub_setup(env, o);
6051         if (rc < 0)
6052                 GOTO(out_site, rc);
6053
6054         rc = osd_procfs_init(o, o->od_svname);
6055         if (rc != 0) {
6056                 CERROR("%s: can't initialize procfs: rc = %d\n",
6057                        o->od_svname, rc);
6058                 GOTO(out_scrub, rc);
6059         }
6060
6061         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
6062
6063         /* initialize quota slave instance */
6064         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6065                                      o->od_proc_entry);
6066         if (IS_ERR(o->od_quota_slave)) {
6067                 rc = PTR_ERR(o->od_quota_slave);
6068                 o->od_quota_slave = NULL;
6069                 GOTO(out_procfs, rc);
6070         }
6071
6072         RETURN(0);
6073
6074 out_procfs:
6075         osd_procfs_fini(o);
6076 out_scrub:
6077         osd_scrub_cleanup(env, o);
6078 out_site:
6079         lu_site_fini(&o->od_site);
6080 out_compat:
6081         osd_obj_map_fini(o);
6082 out_mnt:
6083         osd_umount(env, o);
6084 out_capa:
6085         cleanup_capa_hash(o->od_capa_hash);
6086 out:
6087         return rc;
6088 }
6089
6090 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6091                                           struct lu_device_type *t,
6092                                           struct lustre_cfg *cfg)
6093 {
6094         struct osd_device *o;
6095         int                rc;
6096
6097         OBD_ALLOC_PTR(o);
6098         if (o == NULL)
6099                 return ERR_PTR(-ENOMEM);
6100
6101         rc = dt_device_init(&o->od_dt_dev, t);
6102         if (rc == 0) {
6103                 /* Because the ctx might be revived in dt_device_init,
6104                  * refill the env here */
6105                 lu_env_refill((struct lu_env *)env);
6106                 rc = osd_device_init0(env, o, cfg);
6107                 if (rc)
6108                         dt_device_fini(&o->od_dt_dev);
6109         }
6110
6111         if (unlikely(rc != 0))
6112                 OBD_FREE_PTR(o);
6113
6114         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6115 }
6116
6117 static struct lu_device *osd_device_free(const struct lu_env *env,
6118                                          struct lu_device *d)
6119 {
6120         struct osd_device *o = osd_dev(d);
6121         ENTRY;
6122
6123         cleanup_capa_hash(o->od_capa_hash);
6124         /* XXX: make osd top device in order to release reference */
6125         d->ld_site->ls_top_dev = d;
6126         lu_site_purge(env, d->ld_site, -1);
6127         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6128                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6129                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6130         }
6131         lu_site_fini(&o->od_site);
6132         dt_device_fini(&o->od_dt_dev);
6133         OBD_FREE_PTR(o);
6134         RETURN(NULL);
6135 }
6136
6137 static int osd_process_config(const struct lu_env *env,
6138                               struct lu_device *d, struct lustre_cfg *cfg)
6139 {
6140         struct osd_device               *o = osd_dev(d);
6141         int                             rc;
6142         ENTRY;
6143
6144         switch (cfg->lcfg_command) {
6145         case LCFG_SETUP:
6146                 rc = osd_mount(env, o, cfg);
6147                 break;
6148         case LCFG_CLEANUP:
6149                 lu_dev_del_linkage(d->ld_site, d);
6150                 rc = osd_shutdown(env, o);
6151                 break;
6152         case LCFG_PARAM:
6153                 LASSERT(&o->od_dt_dev);
6154                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6155                                               cfg, &o->od_dt_dev);
6156                 if (rc > 0 || rc == -ENOSYS)
6157                         rc = class_process_proc_param(PARAM_OST,
6158                                                       lprocfs_osd_obd_vars,
6159                                                       cfg, &o->od_dt_dev);
6160                 break;
6161         default:
6162                 rc = -ENOSYS;
6163         }
6164
6165         RETURN(rc);
6166 }
6167
6168 static int osd_recovery_complete(const struct lu_env *env,
6169                                  struct lu_device *d)
6170 {
6171         struct osd_device       *osd = osd_dev(d);
6172         int                      rc = 0;
6173         ENTRY;
6174
6175         if (osd->od_quota_slave == NULL)
6176                 RETURN(0);
6177
6178         /* start qsd instance on recovery completion, this notifies the quota
6179          * slave code that we are about to process new requests now */
6180         rc = qsd_start(env, osd->od_quota_slave);
6181         RETURN(rc);
6182 }
6183
6184 /*
6185  * we use exports to track all osd users
6186  */
6187 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6188                            struct obd_device *obd, struct obd_uuid *cluuid,
6189                            struct obd_connect_data *data, void *localdata)
6190 {
6191         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6192         struct lustre_handle  conn;
6193         int                   rc;
6194         ENTRY;
6195
6196         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6197
6198         rc = class_connect(&conn, obd, cluuid);
6199         if (rc)
6200                 RETURN(rc);
6201
6202         *exp = class_conn2export(&conn);
6203
6204         spin_lock(&osd->od_osfs_lock);
6205         osd->od_connects++;
6206         spin_unlock(&osd->od_osfs_lock);
6207
6208         RETURN(0);
6209 }
6210
6211 /*
6212  * once last export (we don't count self-export) disappeared
6213  * osd can be released
6214  */
6215 static int osd_obd_disconnect(struct obd_export *exp)
6216 {
6217         struct obd_device *obd = exp->exp_obd;
6218         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6219         int                rc, release = 0;
6220         ENTRY;
6221
6222         /* Only disconnect the underlying layers on the final disconnect. */
6223         spin_lock(&osd->od_osfs_lock);
6224         osd->od_connects--;
6225         if (osd->od_connects == 0)
6226                 release = 1;
6227         spin_unlock(&osd->od_osfs_lock);
6228
6229         rc = class_disconnect(exp); /* bz 9811 */
6230
6231         if (rc == 0 && release)
6232                 class_manual_cleanup(obd);
6233         RETURN(rc);
6234 }
6235
6236 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6237                        struct lu_device *dev)
6238 {
6239         struct osd_device *osd = osd_dev(dev);
6240         int                result = 0;
6241         ENTRY;
6242
6243         if (osd->od_quota_slave != NULL) {
6244                 /* set up quota slave objects */
6245                 result = qsd_prepare(env, osd->od_quota_slave);
6246                 if (result != 0)
6247                         RETURN(result);
6248         }
6249
6250         result = osd_fid_init(env, osd);
6251
6252         RETURN(result);
6253 }
6254
6255 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6256                   struct lu_fid *fid, struct md_op_data *op_data)
6257 {
6258         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6259
6260         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6261 }
6262
6263 static const struct lu_object_operations osd_lu_obj_ops = {
6264         .loo_object_init      = osd_object_init,
6265         .loo_object_delete    = osd_object_delete,
6266         .loo_object_release   = osd_object_release,
6267         .loo_object_free      = osd_object_free,
6268         .loo_object_print     = osd_object_print,
6269         .loo_object_invariant = osd_object_invariant
6270 };
6271
6272 const struct lu_device_operations osd_lu_ops = {
6273         .ldo_object_alloc      = osd_object_alloc,
6274         .ldo_process_config    = osd_process_config,
6275         .ldo_recovery_complete = osd_recovery_complete,
6276         .ldo_prepare           = osd_prepare,
6277 };
6278
6279 static const struct lu_device_type_operations osd_device_type_ops = {
6280         .ldto_init = osd_type_init,
6281         .ldto_fini = osd_type_fini,
6282
6283         .ldto_start = osd_type_start,
6284         .ldto_stop  = osd_type_stop,
6285
6286         .ldto_device_alloc = osd_device_alloc,
6287         .ldto_device_free  = osd_device_free,
6288
6289         .ldto_device_init    = osd_device_init,
6290         .ldto_device_fini    = osd_device_fini
6291 };
6292
6293 struct lu_device_type osd_device_type = {
6294         .ldt_tags     = LU_DEVICE_DT,
6295         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6296         .ldt_ops      = &osd_device_type_ops,
6297         .ldt_ctx_tags = LCT_LOCAL,
6298 };
6299
6300 /*
6301  * lprocfs legacy support.
6302  */
6303 static struct obd_ops osd_obd_device_ops = {
6304         .o_owner = THIS_MODULE,
6305         .o_connect      = osd_obd_connect,
6306         .o_disconnect   = osd_obd_disconnect,
6307         .o_fid_alloc    = osd_fid_alloc,
6308 };
6309
6310 static int __init osd_mod_init(void)
6311 {
6312         int rc;
6313
6314         osd_oi_mod_init();
6315
6316         rc = lu_kmem_init(ldiskfs_caches);
6317         if (rc)
6318                 return rc;
6319
6320         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6321                                  lprocfs_osd_module_vars,
6322                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6323         if (rc)
6324                 lu_kmem_fini(ldiskfs_caches);
6325         return rc;
6326 }
6327
6328 static void __exit osd_mod_exit(void)
6329 {
6330         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6331         lu_kmem_fini(ldiskfs_caches);
6332 }
6333
6334 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6335 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6336 MODULE_LICENSE("GPL");
6337
6338 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);