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