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, 2014, 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 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51
  52 /* LUSTRE_VERSION_CODE */
  53 #include <lustre_ver.h>
  54 /* prerequisite for linux/xattr.h */
  55 #include <linux/types.h>
  56 /* prerequisite for linux/xattr.h */
  57 #include <linux/fs.h>
  58 /* XATTR_{REPLACE,CREATE} */
  59 #include <linux/xattr.h>
  60
  61 #include <ldiskfs/ldiskfs.h>
  62 #include <ldiskfs/xattr.h>
  63 #undef ENTRY
  64 /*
  65  * struct OBD_{ALLOC,FREE}*()
  66  * OBD_FAIL_CHECK
  67  */
  68 #include <obd_support.h>
  69 /* struct ptlrpc_thread */
  70 #include <lustre_net.h>
  71 #include <lustre_fid.h>
  72 /* process_config */
  73 #include <lustre_param.h>
  74
  75 #include "osd_internal.h"
  76 #include "osd_dynlocks.h"
  77
  78 /* llo_* api support */
  79 #include <md_object.h>
  80 #include <lustre_quota.h>
  81
  82 #include <lustre_linkea.h>
  83
  84 int ldiskfs_pdo = 1;
  85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  86                 "ldiskfs with parallel directory operations");
  87
  88 int ldiskfs_track_declares_assert;
  89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  90                 "LBUG during tracking of declares");
  91
  92 /* Slab to allocate dynlocks */
  93 struct kmem_cache *dynlock_cachep;
  94
  95 /* Slab to allocate osd_it_ea */
  96 struct kmem_cache *osd_itea_cachep;
  97
  98 static struct lu_kmem_descr ldiskfs_caches[] = {
  99         {
 100                 .ckd_cache = &dynlock_cachep,
 101                 .ckd_name  = "dynlock_cache",
 102                 .ckd_size  = sizeof(struct dynlock_handle)
 103         },
 104         {
 105                 .ckd_cache = &osd_itea_cachep,
 106                 .ckd_name  = "osd_itea_cache",
 107                 .ckd_size  = sizeof(struct osd_it_ea)
 108         },
 109         {
 110                 .ckd_cache = NULL
 111         }
 112 };
 113
 114 static const char dot[] = ".";
 115 static const char dotdot[] = "..";
 116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 117
 118 static const struct lu_object_operations      osd_lu_obj_ops;
 119 static const struct dt_object_operations      osd_obj_ops;
 120 static const struct dt_object_operations      osd_obj_ea_ops;
 121 static const struct dt_object_operations      osd_obj_otable_it_ops;
 122 static const struct dt_index_operations       osd_index_iam_ops;
 123 static const struct dt_index_operations       osd_index_ea_ops;
 124
 125 int osd_trans_declare_op2rb[] = {
 126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 137 };
 138
 139 static int osd_has_index(const struct osd_object *obj)
 140 {
 141         return obj->oo_dt.do_index_ops != NULL;
 142 }
 143
 144 static int osd_object_invariant(const struct lu_object *l)
 145 {
 146         return osd_invariant(osd_obj(l));
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152
 153 /*
 154  * Concurrency: doesn't matter
 155  */
 156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 157 {
 158         struct osd_thread_info *oti = osd_oti_get(env);
 159         return oti->oti_w_locks > 0 && o->oo_owner == env;
 160 }
 161
 162 /*
 163  * Concurrency: doesn't access mutable data
 164  */
 165 static int osd_root_get(const struct lu_env *env,
 166                         struct dt_device *dev, struct lu_fid *f)
 167 {
 168         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 169         return 0;
 170 }
 171
 172 /*
 173  * OSD object methods.
 174  */
 175
 176 /*
 177  * Concurrency: no concurrent access is possible that early in object
 178  * life-cycle.
 179  */
 180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 181                                           const struct lu_object_header *hdr,
 182                                           struct lu_device *d)
 183 {
 184         struct osd_object *mo;
 185
 186         OBD_ALLOC_PTR(mo);
 187         if (mo != NULL) {
 188                 struct lu_object *l;
 189
 190                 l = &mo->oo_dt.do_lu;
 191                 dt_object_init(&mo->oo_dt, NULL, d);
 192                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 193                 l->lo_ops = &osd_lu_obj_ops;
 194                 init_rwsem(&mo->oo_sem);
 195                 init_rwsem(&mo->oo_ext_idx_sem);
 196                 spin_lock_init(&mo->oo_guard);
 197                 return l;
 198         } else {
 199                 return NULL;
 200         }
 201 }
 202
 203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 204                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 205 {
 206         int rc;
 207
 208         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 209         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 210                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 211         if (rc > 0) {
 212                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 213                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 214                 rc = 0;
 215                 lustre_lma_swab(lma);
 216                 /* Check LMA compatibility */
 217                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 218                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 219                               "for fid = "DFID", ino = %lu\n",
 220                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 221                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 222                               PFID(&lma->lma_self_fid), inode->i_ino);
 223                         rc = -EOPNOTSUPP;
 224                 }
 225         } else if (rc == 0) {
 226                 rc = -ENODATA;
 227         }
 228
 229         return rc;
 230 }
 231
 232 /*
 233  * retrieve object from backend ext fs.
 234  **/
 235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 236                        struct osd_inode_id *id)
 237 {
 238         struct inode *inode = NULL;
 239
 240         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 241         if (IS_ERR(inode)) {
 242                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 243                        id->oii_ino, PTR_ERR(inode));
 244         } else if (id->oii_gen != OSD_OII_NOGEN &&
 245                    inode->i_generation != id->oii_gen) {
 246                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 247                        "i_generation = %u\n",
 248                        id->oii_ino, id->oii_gen, inode->i_generation);
 249                 iput(inode);
 250                 inode = ERR_PTR(-ESTALE);
 251         } else if (inode->i_nlink == 0) {
 252                 /* due to parallel readdir and unlink,
 253                 * we can have dead inode here. */
 254                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 255                 iput(inode);
 256                 inode = ERR_PTR(-ESTALE);
 257         } else if (is_bad_inode(inode)) {
 258                 CWARN("%.16s: bad inode: ino = %u\n",
 259                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 260                 iput(inode);
 261                 inode = ERR_PTR(-ENOENT);
 262         } else {
 263                 if (id->oii_gen == OSD_OII_NOGEN)
 264                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 265
 266                 /* Do not update file c/mtime in ldiskfs.
 267                  * NB: we don't have any lock to protect this because we don't
 268                  * have reference on osd_object now, but contention with
 269                  * another lookup + attr_set can't happen in the tiny window
 270                  * between if (...) and set S_NOCMTIME. */
 271                 if (!(inode->i_flags & S_NOCMTIME))
 272                         inode->i_flags |= S_NOCMTIME;
 273         }
 274         return inode;
 275 }
 276
 277 static struct inode *
 278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 279              struct osd_inode_id *id, struct lu_fid *fid)
 280 {
 281         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 282         struct inode            *inode;
 283         int                      rc;
 284
 285         inode = osd_iget(info, dev, id);
 286         if (IS_ERR(inode))
 287                 return inode;
 288
 289         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 290         if (rc == 0) {
 291                 *fid = lma->lma_self_fid;
 292         } else if (rc == -ENODATA) {
 293                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 294                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 295                 else
 296                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 297         } else {
 298                 iput(inode);
 299                 inode = ERR_PTR(rc);
 300         }
 301         return inode;
 302 }
 303
 304 static struct inode *osd_iget_check(struct osd_thread_info *info,
 305                                     struct osd_device *dev,
 306                                     const struct lu_fid *fid,
 307                                     struct osd_inode_id *id,
 308                                     bool in_oi)
 309 {
 310         struct inode    *inode;
 311         int              rc     = 0;
 312         ENTRY;
 313
 314         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 315         if (IS_ERR(inode)) {
 316                 rc = PTR_ERR(inode);
 317                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 318                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 319                                id->oii_ino, rc);
 320
 321                         GOTO(put, rc);
 322                 }
 323
 324                 goto check_oi;
 325         }
 326
 327         if (is_bad_inode(inode)) {
 328                 rc = -ENOENT;
 329                 if (!in_oi) {
 330                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 331
 332                         GOTO(put, rc);
 333                 }
 334
 335                 goto check_oi;
 336         }
 337
 338         if (id->oii_gen != OSD_OII_NOGEN &&
 339             inode->i_generation != id->oii_gen) {
 340                 rc = -ESTALE;
 341                 if (!in_oi) {
 342                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 343                                "oii_gen = %u, i_generation = %u\n",
 344                                id->oii_ino, id->oii_gen, inode->i_generation);
 345
 346                         GOTO(put, rc);
 347                 }
 348
 349                 goto check_oi;
 350         }
 351
 352         if (inode->i_nlink == 0) {
 353                 rc = -ENOENT;
 354                 if (!in_oi) {
 355                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 356
 357                         GOTO(put, rc);
 358                 }
 359
 360                 goto check_oi;
 361         }
 362
 363 check_oi:
 364         if (rc != 0) {
 365                 struct osd_inode_id saved_id = *id;
 366
 367                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 368
 369                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 370                 /* XXX: There are some possible cases:
 371                  *      1. rc = 0.
 372                  *         Backup/restore caused the OI invalid.
 373                  *      2. rc = 0.
 374                  *         Someone unlinked the object but NOT removed
 375                  *         the OI mapping, such as mount target device
 376                  *         as ldiskfs, and modify something directly.
 377                  *      3. rc = -ENOENT.
 378                  *         Someone just removed the object between the
 379                  *         former oi_lookup and the iget. It is normal.
 380                  *      4. Other failure cases.
 381                  *
 382                  *      Generally, when the device is mounted, it will
 383                  *      auto check whether the system is restored from
 384                  *      file-level backup or not. We trust such detect
 385                  *      to distinguish the 1st case from the 2nd case. */
 386                 if (rc == 0) {
 387                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 388                             inode->i_generation == id->oii_gen) {
 389                                 rc = -ENOENT;
 390                         } else {
 391                                 __u32 level = D_LFSCK;
 392
 393                                 rc = -EREMCHG;
 394                                 if (!thread_is_running(&dev->od_scrub.os_thread))
 395                                         level |= D_CONSOLE;
 396
 397                                 CDEBUG(level, "%s: the OI mapping for the FID "
 398                                        DFID" become inconsistent, the given ID "
 399                                        "%u/%u, the ID in OI mapping %u/%u\n",
 400                                        osd_name(dev), PFID(fid),
 401                                        saved_id.oii_ino, saved_id.oii_gen,
 402                                        id->oii_ino, id->oii_ino);
 403                         }
 404                 }
 405         } else {
 406                 if (id->oii_gen == OSD_OII_NOGEN)
 407                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 408
 409                 /* Do not update file c/mtime in ldiskfs.
 410                  * NB: we don't have any lock to protect this because we don't
 411                  * have reference on osd_object now, but contention with
 412                  * another lookup + attr_set can't happen in the tiny window
 413                  * between if (...) and set S_NOCMTIME. */
 414                 if (!(inode->i_flags & S_NOCMTIME))
 415                         inode->i_flags |= S_NOCMTIME;
 416         }
 417
 418         GOTO(put, rc);
 419
 420 put:
 421         if (rc != 0) {
 422                 if (!IS_ERR(inode))
 423                         iput(inode);
 424
 425                 inode = ERR_PTR(rc);
 426         }
 427
 428         return inode;
 429 }
 430
 431 /**
 432  * \retval +v: new filter_fid, does not contain self-fid
 433  * \retval 0:  filter_fid_old, contains self-fid
 434  * \retval -v: other failure cases
 435  */
 436 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 437                  struct dentry *dentry, struct lu_fid *fid)
 438 {
 439         struct filter_fid_old   *ff     = &info->oti_ff;
 440         struct ost_id           *ostid  = &info->oti_ostid;
 441         int                      rc;
 442
 443         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 444         if (rc == sizeof(*ff)) {
 445                 rc = 0;
 446                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 447                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 448                 /* XXX: should use real OST index in the future. LU-3569 */
 449                 ostid_to_fid(fid, ostid, 0);
 450         } else if (rc == sizeof(struct filter_fid)) {
 451                 rc = 1;
 452         } else if (rc >= 0) {
 453                 rc = -EINVAL;
 454         }
 455
 456         return rc;
 457 }
 458
 459 static int osd_lma_self_repair(struct osd_thread_info *info,
 460                                struct osd_device *osd, struct inode *inode,
 461                                const struct lu_fid *fid, __u32 compat)
 462 {
 463         handle_t *jh;
 464         int       rc;
 465
 466         LASSERT(current->journal_info == NULL);
 467
 468         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 469                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 470         if (IS_ERR(jh)) {
 471                 rc = PTR_ERR(jh);
 472                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 473                       osd_name(osd), rc);
 474                 return rc;
 475         }
 476
 477         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 478         if (rc != 0)
 479                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 480                       osd_name(osd), rc);
 481         ldiskfs_journal_stop(jh);
 482         return rc;
 483 }
 484
 485 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 486 {
 487         struct osd_thread_info  *info   = osd_oti_get(env);
 488         struct osd_device       *osd    = osd_obj2dev(obj);
 489         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 490         struct inode            *inode  = obj->oo_inode;
 491         struct dentry           *dentry = &info->oti_obj_dentry;
 492         struct lu_fid           *fid    = NULL;
 493         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 494         int                      rc;
 495         ENTRY;
 496
 497         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 498         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 499                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 500         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 501                 fid = &lma->lma_self_fid;
 502                 rc = osd_get_idif(info, inode, dentry, fid);
 503                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 504                         /* For the given OST-object, if it has neither LMA nor
 505                          * FID in XATTR_NAME_FID, then the given FID (which is
 506                          * contained in the @obj, from client RPC for locating
 507                          * the OST-object) is trusted. We use it to generate
 508                          * the LMA. */
 509                         osd_lma_self_repair(info, osd, inode, rfid,
 510                                             LMAC_FID_ON_OST);
 511                         RETURN(0);
 512                 }
 513         }
 514
 515         if (unlikely(rc == -ENODATA))
 516                 RETURN(0);
 517
 518         if (rc < 0)
 519                 RETURN(rc);
 520
 521         if (rc > 0) {
 522                 rc = 0;
 523                 lustre_lma_swab(lma);
 524                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 525                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 526                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 527                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 528                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 529                               PFID(rfid), inode->i_ino);
 530                         rc = -EOPNOTSUPP;
 531                 } else {
 532                         fid = &lma->lma_self_fid;
 533                 }
 534         }
 535
 536         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 537                 __u32 level = D_LFSCK;
 538
 539                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 540                         struct ost_id   *oi   = &info->oti_ostid;
 541                         struct lu_fid   *fid1 = &info->oti_fid3;
 542                         __u32            idx  = fid_idif_ost_idx(rfid);
 543
 544                         /* For old IDIF, the OST index is not part of the IDIF,
 545                          * Means that different OSTs may have the same IDIFs.
 546                          * Under such case, we need to make some compatible
 547                          * check to make sure to trigger OI scrub properly. */
 548                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 549                                 /* Given @rfid is new, LMA is old. */
 550                                 fid_to_ostid(fid, oi);
 551                                 ostid_to_fid(fid1, oi, idx);
 552                                 if (lu_fid_eq(fid1, rfid)) {
 553                                         if (osd->od_index_in_idif)
 554                                                 osd_lma_self_repair(info, osd,
 555                                                         inode, rfid,
 556                                                         LMAC_FID_ON_OST);
 557                                         RETURN(0);
 558                                 }
 559                         }
 560                 }
 561
 562
 563                 rc = -EREMCHG;
 564                 if (!thread_is_running(&osd->od_scrub.os_thread))
 565                         level |= D_CONSOLE;
 566
 567                 CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
 568                        osd_name(osd), PFID(rfid), PFID(fid));
 569         }
 570
 571         RETURN(rc);
 572 }
 573
 574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 575                           const struct lu_fid *fid,
 576                           const struct lu_object_conf *conf)
 577 {
 578         struct osd_thread_info *info;
 579         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 580         struct osd_device      *dev;
 581         struct osd_idmap_cache *oic;
 582         struct osd_inode_id    *id;
 583         struct inode           *inode;
 584         struct osd_scrub       *scrub;
 585         struct scrub_file      *sf;
 586         int                     result;
 587         int                     saved  = 0;
 588         bool                    in_oi  = false;
 589         bool                    triggered = false;
 590         ENTRY;
 591
 592         LINVRNT(osd_invariant(obj));
 593         LASSERT(obj->oo_inode == NULL);
 594         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 595
 596         dev = osd_dev(ldev);
 597         scrub = &dev->od_scrub;
 598         sf = &scrub->os_file;
 599         info = osd_oti_get(env);
 600         LASSERT(info);
 601         oic = &info->oti_cache;
 602
 603         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 604                 RETURN(-ENOENT);
 605
 606         /* For the object is created as locking anchor, or for the object to
 607          * be created on disk. No need to osd_oi_lookup() at here because FID
 608          * shouldn't never be re-used, if it's really a duplicate FID from
 609          * unexpected reason, we should be able to detect it later by calling
 610          * do_create->osd_oi_insert(). */
 611         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 612                 GOTO(out, result = 0);
 613
 614         /* Search order: 1. per-thread cache. */
 615         if (lu_fid_eq(fid, &oic->oic_fid) &&
 616             likely(oic->oic_dev == dev)) {
 617                 id = &oic->oic_lid;
 618                 goto iget;
 619         }
 620
 621         id = &info->oti_id;
 622         if (!list_empty(&scrub->os_inconsistent_items)) {
 623                 /* Search order: 2. OI scrub pending list. */
 624                 result = osd_oii_lookup(dev, fid, id);
 625                 if (result == 0)
 626                         goto iget;
 627         }
 628
 629         /* Search order: 3. OI files. */
 630         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 631         if (result == -ENOENT) {
 632                 if (!fid_is_norm(fid) ||
 633                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 634                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 635                                       sf->sf_oi_bitmap))
 636                         GOTO(out, result = 0);
 637
 638                 goto trigger;
 639         }
 640
 641         if (result != 0)
 642                 GOTO(out, result);
 643
 644         in_oi = true;
 645
 646 iget:
 647         inode = osd_iget_check(info, dev, fid, id, in_oi);
 648         if (IS_ERR(inode)) {
 649                 result = PTR_ERR(inode);
 650                 if (result == -ENOENT || result == -ESTALE) {
 651                         if (!in_oi)
 652                                 fid_zero(&oic->oic_fid);
 653
 654                         GOTO(out, result = -ENOENT);
 655                 } else if (result == -EREMCHG) {
 656
 657 trigger:
 658                         if (!in_oi)
 659                                 fid_zero(&oic->oic_fid);
 660
 661                         if (unlikely(triggered))
 662                                 GOTO(out, result = saved);
 663
 664                         triggered = true;
 665                         if (thread_is_running(&scrub->os_thread)) {
 666                                 result = -EINPROGRESS;
 667                         } else if (!dev->od_noscrub) {
 668                                 /* Since we do not know the right OI mapping,
 669                                  * we have to trigger OI scrub to scan the
 670                                  * whole device. */
 671                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
 672                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 673                                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
 674                                        "scrub by RPC for "DFID", rc = %d [1]\n",
 675                                        osd_name(dev), PFID(fid),result);
 676                                 if (result == 0 || result == -EALREADY)
 677                                         result = -EINPROGRESS;
 678                                 else
 679                                         result = -EREMCHG;
 680                         }
 681
 682                         /* We still have chance to get the valid inode: for the
 683                          * object which is referenced by remote name entry, the
 684                          * object on the local MDT will be linked under the dir
 685                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 686                          *
 687                          * We do not know whether the object for the given FID
 688                          * is referenced by some remote name entry or not, and
 689                          * especially for DNE II, a multiple-linked object may
 690                          * have many name entries reside on many MDTs.
 691                          *
 692                          * To simplify the operation, OSD will not distinguish
 693                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 694                          * only happened for the RPC from other MDT during the
 695                          * OI scrub, or for the client side RPC with FID only,
 696                          * such as FID to path, or from old connected client. */
 697                         saved = result;
 698                         result = osd_lookup_in_remote_parent(info, dev,
 699                                                              fid, id);
 700                         if (result == 0) {
 701                                 in_oi = false;
 702                                 goto iget;
 703                         }
 704
 705                         result = saved;
 706                 }
 707
 708                 GOTO(out, result);
 709         }
 710
 711         obj->oo_inode = inode;
 712         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 713
 714         result = osd_check_lma(env, obj);
 715         if (result != 0) {
 716                 iput(inode);
 717                 obj->oo_inode = NULL;
 718                 if (result == -EREMCHG) {
 719                         if (!in_oi) {
 720                                 result = osd_oi_lookup(info, dev, fid, id,
 721                                                        OI_CHECK_FLD);
 722                                 if (result != 0) {
 723                                         fid_zero(&oic->oic_fid);
 724                                         GOTO(out, result);
 725                                 }
 726                         }
 727
 728                         goto trigger;
 729                 }
 730
 731                 GOTO(out, result);
 732         }
 733
 734         obj->oo_compat_dot_created = 1;
 735         obj->oo_compat_dotdot_created = 1;
 736
 737         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 738                 GOTO(out, result = 0);
 739
 740         LASSERT(obj->oo_hl_head == NULL);
 741         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 742         if (obj->oo_hl_head == NULL) {
 743                 obj->oo_inode = NULL;
 744                 iput(inode);
 745                 GOTO(out, result = -ENOMEM);
 746         }
 747         GOTO(out, result = 0);
 748
 749 out:
 750         LINVRNT(osd_invariant(obj));
 751         return result;
 752 }
 753
 754 /*
 755  * Concurrency: shouldn't matter.
 756  */
 757 static void osd_object_init0(struct osd_object *obj)
 758 {
 759         LASSERT(obj->oo_inode != NULL);
 760         obj->oo_dt.do_body_ops = &osd_body_ops;
 761         obj->oo_dt.do_lu.lo_header->loh_attr |=
 762                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 763 }
 764
 765 /*
 766  * Concurrency: no concurrent access is possible that early in object
 767  * life-cycle.
 768  */
 769 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 770                            const struct lu_object_conf *conf)
 771 {
 772         struct osd_object *obj = osd_obj(l);
 773         int result;
 774
 775         LINVRNT(osd_invariant(obj));
 776
 777         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 778                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 779                 l->lo_header->loh_attr |= LOHA_EXISTS;
 780                 return 0;
 781         }
 782
 783         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 784         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 785         if (result == 0 && obj->oo_inode != NULL)
 786                 osd_object_init0(obj);
 787
 788         LINVRNT(osd_invariant(obj));
 789         return result;
 790 }
 791
 792 /*
 793  * Concurrency: no concurrent access is possible that late in object
 794  * life-cycle.
 795  */
 796 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 797 {
 798         struct osd_object *obj = osd_obj(l);
 799
 800         LINVRNT(osd_invariant(obj));
 801
 802         dt_object_fini(&obj->oo_dt);
 803         if (obj->oo_hl_head != NULL)
 804                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 805         OBD_FREE_PTR(obj);
 806 }
 807
 808 /*
 809  * Concurrency: no concurrent access is possible that late in object
 810  * life-cycle.
 811  */
 812 static void osd_index_fini(struct osd_object *o)
 813 {
 814         struct iam_container *bag;
 815
 816         if (o->oo_dir != NULL) {
 817                 bag = &o->oo_dir->od_container;
 818                 if (o->oo_inode != NULL) {
 819                         if (bag->ic_object == o->oo_inode)
 820                                 iam_container_fini(bag);
 821                 }
 822                 OBD_FREE_PTR(o->oo_dir);
 823                 o->oo_dir = NULL;
 824         }
 825 }
 826
 827 /*
 828  * Concurrency: no concurrent access is possible that late in object
 829  * life-cycle (for all existing callers, that is. New callers have to provide
 830  * their own locking.)
 831  */
 832 static int osd_inode_unlinked(const struct inode *inode)
 833 {
 834         return inode->i_nlink == 0;
 835 }
 836
 837 enum {
 838         OSD_TXN_OI_DELETE_CREDITS    = 20,
 839         OSD_TXN_INODE_DELETE_CREDITS = 20
 840 };
 841
 842 /*
 843  * Journal
 844  */
 845
 846 #if OSD_THANDLE_STATS
 847 /**
 848  * Set time when the handle is allocated
 849  */
 850 static void osd_th_alloced(struct osd_thandle *oth)
 851 {
 852         oth->oth_alloced = cfs_time_current();
 853 }
 854
 855 /**
 856  * Set time when the handle started
 857  */
 858 static void osd_th_started(struct osd_thandle *oth)
 859 {
 860         oth->oth_started = cfs_time_current();
 861 }
 862
 863 /**
 864  * Helper function to convert time interval to microseconds packed in
 865  * long int.
 866  */
 867 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 868 {
 869         struct timeval val;
 870
 871         cfs_duration_usec(cfs_time_sub(end, start), &val);
 872         return val.tv_sec * 1000000 + val.tv_usec;
 873 }
 874
 875 /**
 876  * Check whether the we deal with this handle for too long.
 877  */
 878 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 879                                 cfs_time_t alloced, cfs_time_t started,
 880                                 cfs_time_t closed)
 881 {
 882         cfs_time_t now = cfs_time_current();
 883
 884         LASSERT(dev != NULL);
 885
 886         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 887                             interval_to_usec(alloced, started));
 888         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 889                             interval_to_usec(started, closed));
 890         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 891                             interval_to_usec(closed, now));
 892
 893         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 894                 CWARN("transaction handle %p was open for too long: "
 895                       "now "CFS_TIME_T" ,"
 896                       "alloced "CFS_TIME_T" ,"
 897                       "started "CFS_TIME_T" ,"
 898                       "closed "CFS_TIME_T"\n",
 899                       oth, now, alloced, started, closed);
 900                 libcfs_debug_dumpstack(NULL);
 901         }
 902 }
 903
 904 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 905 {                                                                       \
 906         cfs_time_t __closed = cfs_time_current();                       \
 907         cfs_time_t __alloced = oth->oth_alloced;                        \
 908         cfs_time_t __started = oth->oth_started;                        \
 909                                                                         \
 910         expr;                                                           \
 911         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 912 }
 913
 914 #else /* OSD_THANDLE_STATS */
 915
 916 #define osd_th_alloced(h)                  do {} while(0)
 917 #define osd_th_started(h)                  do {} while(0)
 918 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 919
 920 #endif /* OSD_THANDLE_STATS */
 921
 922 /*
 923  * Concurrency: doesn't access mutable data.
 924  */
 925 static int osd_param_is_not_sane(const struct osd_device *dev,
 926                                  const struct thandle *th)
 927 {
 928         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 929
 930         return oh->ot_credits > osd_transaction_size(dev);
 931 }
 932
 933 /*
 934  * Concurrency: shouldn't matter.
 935  */
 936 static void osd_trans_commit_cb(struct super_block *sb,
 937                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 938 {
 939         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 940         struct thandle     *th  = &oh->ot_super;
 941         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 942         struct dt_txn_commit_cb *dcb, *tmp;
 943
 944         LASSERT(oh->ot_handle == NULL);
 945
 946         if (error)
 947                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 948
 949         dt_txn_hook_commit(th);
 950
 951         /* call per-transaction callbacks if any */
 952         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 954                          "commit callback entry: magic=%x name='%s'\n",
 955                          dcb->dcb_magic, dcb->dcb_name);
 956                 list_del_init(&dcb->dcb_linkage);
 957                 dcb->dcb_func(NULL, th, dcb, error);
 958         }
 959
 960         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 961         lu_device_put(lud);
 962         th->th_dev = NULL;
 963
 964         lu_context_exit(&th->th_ctx);
 965         lu_context_fini(&th->th_ctx);
 966         OBD_FREE_PTR(oh);
 967 }
 968
 969 static struct thandle *osd_trans_create(const struct lu_env *env,
 970                                         struct dt_device *d)
 971 {
 972         struct osd_thread_info  *oti = osd_oti_get(env);
 973         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 974         struct osd_thandle      *oh;
 975         struct thandle          *th;
 976         ENTRY;
 977
 978         /* on pending IO in this thread should left from prev. request */
 979         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 980
 981         th = ERR_PTR(-ENOMEM);
 982         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 983         if (oh != NULL) {
 984                 oh->ot_quota_trans = &oti->oti_quota_trans;
 985                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 986                 th = &oh->ot_super;
 987                 th->th_dev = d;
 988                 th->th_result = 0;
 989                 th->th_tags = LCT_TX_HANDLE;
 990                 oh->ot_credits = 0;
 991                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 992                 osd_th_alloced(oh);
 993
 994                 memset(oti->oti_declare_ops, 0,
 995                                         sizeof(oti->oti_declare_ops));
 996                 memset(oti->oti_declare_ops_rb, 0,
 997                                         sizeof(oti->oti_declare_ops_rb));
 998                 memset(oti->oti_declare_ops_cred, 0,
 999                                         sizeof(oti->oti_declare_ops_cred));
1000                 oti->oti_rollback = false;
1001         }
1002         RETURN(th);
1003 }
1004
1005 /*
1006  * Concurrency: shouldn't matter.
1007  */
1008 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1009                            struct thandle *th)
1010 {
1011         struct osd_thread_info *oti = osd_oti_get(env);
1012         struct osd_device  *dev = osd_dt_dev(d);
1013         handle_t           *jh;
1014         struct osd_thandle *oh;
1015         int rc;
1016
1017         ENTRY;
1018
1019         LASSERT(current->journal_info == NULL);
1020
1021         oh = container_of0(th, struct osd_thandle, ot_super);
1022         LASSERT(oh != NULL);
1023         LASSERT(oh->ot_handle == NULL);
1024
1025         rc = dt_txn_hook_start(env, d, th);
1026         if (rc != 0)
1027                 GOTO(out, rc);
1028
1029         if (unlikely(osd_param_is_not_sane(dev, th))) {
1030                 static unsigned long last_printed;
1031                 static int last_credits;
1032
1033                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1034                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1035                       oh->ot_credits,
1036                       osd_journal(dev)->j_max_transaction_buffers);
1037                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1038                       oti->oti_declare_ops[OSD_OT_CREATE],
1039                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1040                       oti->oti_declare_ops[OSD_OT_DESTROY],
1041                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1042                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1043                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1044                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1045                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1046                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1047                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1048                       oti->oti_declare_ops[OSD_OT_WRITE],
1049                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1050                       oti->oti_declare_ops[OSD_OT_PUNCH],
1051                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1052                       oti->oti_declare_ops[OSD_OT_QUOTA],
1053                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1054                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1055                       oti->oti_declare_ops[OSD_OT_INSERT],
1056                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1057                       oti->oti_declare_ops[OSD_OT_DELETE],
1058                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1059                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1060                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1061                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1062                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1063                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1064
1065                 if (last_credits != oh->ot_credits &&
1066                     time_after(jiffies, last_printed +
1067                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1068                         libcfs_debug_dumpstack(NULL);
1069                         last_credits = oh->ot_credits;
1070                         last_printed = jiffies;
1071                 }
1072                 /* XXX Limit the credits to 'max_transaction_buffers', and
1073                  *     let the underlying filesystem to catch the error if
1074                  *     we really need so many credits.
1075                  *
1076                  *     This should be removed when we can calculate the
1077                  *     credits precisely. */
1078                 oh->ot_credits = osd_transaction_size(dev);
1079         }
1080
1081         /*
1082          * XXX temporary stuff. Some abstraction layer should
1083          * be used.
1084          */
1085         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1086         osd_th_started(oh);
1087         if (!IS_ERR(jh)) {
1088                 oh->ot_handle = jh;
1089                 LASSERT(oti->oti_txns == 0);
1090                 lu_context_init(&th->th_ctx, th->th_tags);
1091                 lu_context_enter(&th->th_ctx);
1092
1093                 lu_device_get(&d->dd_lu_dev);
1094                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1095                               "osd-tx", th);
1096                 oti->oti_txns++;
1097                 rc = 0;
1098         } else {
1099                 rc = PTR_ERR(jh);
1100         }
1101 out:
1102         RETURN(rc);
1103 }
1104
1105 static int osd_seq_exists(const struct lu_env *env,
1106                           struct osd_device *osd, u64 seq)
1107 {
1108         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1109         struct seq_server_site  *ss = osd_seq_site(osd);
1110         int                     rc;
1111         ENTRY;
1112
1113         LASSERT(ss != NULL);
1114         LASSERT(ss->ss_server_fld != NULL);
1115
1116         rc = osd_fld_lookup(env, osd, seq, range);
1117         if (rc != 0) {
1118                 if (rc != -ENOENT)
1119                         CERROR("%s: can't lookup FLD sequence "LPX64
1120                                ": rc = %d\n", osd_name(osd), seq, rc);
1121                 RETURN(0);
1122         }
1123
1124         RETURN(ss->ss_node_id == range->lsr_index);
1125 }
1126
1127 /*
1128  * Concurrency: shouldn't matter.
1129  */
1130 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1131                           struct thandle *th)
1132 {
1133         int                     rc = 0;
1134         struct osd_thandle     *oh;
1135         struct osd_thread_info *oti = osd_oti_get(env);
1136         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1137         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1138         struct qsd_instance    *qsd = osd->od_quota_slave;
1139         struct lquota_trans    *qtrans;
1140         ENTRY;
1141
1142         oh = container_of0(th, struct osd_thandle, ot_super);
1143
1144         qtrans = oh->ot_quota_trans;
1145         oh->ot_quota_trans = NULL;
1146
1147         if (oh->ot_handle != NULL) {
1148                 handle_t *hdl = oh->ot_handle;
1149
1150                 /*
1151                  * add commit callback
1152                  * notice we don't do this in osd_trans_start()
1153                  * as underlying transaction can change during truncate
1154                  */
1155                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1156                                          &oh->ot_jcb);
1157
1158                 LASSERT(oti->oti_txns == 1);
1159                 oti->oti_txns--;
1160                 rc = dt_txn_hook_stop(env, th);
1161                 if (rc != 0)
1162                         CERROR("%s: failed in transaction hook: rc = %d\n",
1163                                osd_name(osd), rc);
1164
1165                 /* hook functions might modify th_sync */
1166                 hdl->h_sync = th->th_sync;
1167
1168                 oh->ot_handle = NULL;
1169                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1170                 if (rc != 0)
1171                         CERROR("%s: failed to stop transaction: rc = %d\n",
1172                                osd_name(osd), rc);
1173         } else {
1174                 OBD_FREE_PTR(oh);
1175         }
1176
1177         /* inform the quota slave device that the transaction is stopping */
1178         qsd_op_end(env, qsd, qtrans);
1179
1180         /* as we want IO to journal and data IO be concurrent, we don't block
1181          * awaiting data IO completion in osd_do_bio(), instead we wait here
1182          * once transaction is submitted to the journal. all reqular requests
1183          * don't do direct IO (except read/write), thus this wait_event becomes
1184          * no-op for them.
1185          *
1186          * IMPORTANT: we have to wait till any IO submited by the thread is
1187          * completed otherwise iobuf may be corrupted by different request
1188          */
1189         wait_event(iobuf->dr_wait,
1190                        atomic_read(&iobuf->dr_numreqs) == 0);
1191         osd_fini_iobuf(osd, iobuf);
1192         if (!rc)
1193                 rc = iobuf->dr_error;
1194
1195         RETURN(rc);
1196 }
1197
1198 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1199 {
1200         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1201                                                ot_super);
1202
1203         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1204         LASSERT(&dcb->dcb_func != NULL);
1205         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * Called just before object is freed. Releases all resources except for
1212  * object itself (that is released by osd_object_free()).
1213  *
1214  * Concurrency: no concurrent access is possible that late in object
1215  * life-cycle.
1216  */
1217 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1218 {
1219         struct osd_object *obj   = osd_obj(l);
1220         struct inode      *inode = obj->oo_inode;
1221
1222         LINVRNT(osd_invariant(obj));
1223
1224         /*
1225          * If object is unlinked remove fid->ino mapping from object index.
1226          */
1227
1228         osd_index_fini(obj);
1229         if (inode != NULL) {
1230                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1231                 qid_t                    uid = i_uid_read(inode);
1232                 qid_t                    gid = i_gid_read(inode);
1233
1234                 iput(inode);
1235                 obj->oo_inode = NULL;
1236
1237                 if (qsd != NULL) {
1238                         struct osd_thread_info  *info = osd_oti_get(env);
1239                         struct lquota_id_info   *qi = &info->oti_qi;
1240
1241                         /* Release granted quota to master if necessary */
1242                         qi->lqi_id.qid_uid = uid;
1243                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1244
1245                         qi->lqi_id.qid_uid = gid;
1246                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1247                 }
1248         }
1249 }
1250
1251 /*
1252  * Concurrency: ->loo_object_release() is called under site spin-lock.
1253  */
1254 static void osd_object_release(const struct lu_env *env,
1255                                struct lu_object *l)
1256 {
1257 }
1258
1259 /*
1260  * Concurrency: shouldn't matter.
1261  */
1262 static int osd_object_print(const struct lu_env *env, void *cookie,
1263                             lu_printer_t p, const struct lu_object *l)
1264 {
1265         struct osd_object *o = osd_obj(l);
1266         struct iam_descr  *d;
1267
1268         if (o->oo_dir != NULL)
1269                 d = o->oo_dir->od_container.ic_descr;
1270         else
1271                 d = NULL;
1272         return (*p)(env, cookie,
1273                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1274                     o, o->oo_inode,
1275                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1276                     o->oo_inode ? o->oo_inode->i_generation : 0,
1277                     d ? d->id_ops->id_name : "plain");
1278 }
1279
1280 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1281
1282 /*
1283  * Concurrency: shouldn't matter.
1284  */
1285 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1286                struct obd_statfs *sfs)
1287 {
1288         struct osd_device  *osd = osd_dt_dev(d);
1289         struct super_block *sb = osd_sb(osd);
1290         struct kstatfs     *ksfs;
1291         int result = 0;
1292
1293         if (unlikely(osd->od_mnt == NULL))
1294                 return -EINPROGRESS;
1295
1296         /* osd_lproc.c call this without env, allocate ksfs for that case */
1297         if (unlikely(env == NULL)) {
1298                 OBD_ALLOC_PTR(ksfs);
1299                 if (ksfs == NULL)
1300                         return -ENOMEM;
1301         } else {
1302                 ksfs = &osd_oti_get(env)->oti_ksfs;
1303         }
1304
1305         spin_lock(&osd->od_osfs_lock);
1306         result = sb->s_op->statfs(sb->s_root, ksfs);
1307         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1308                 statfs_pack(sfs, ksfs);
1309                 if (sb->s_flags & MS_RDONLY)
1310                         sfs->os_state = OS_STATE_READONLY;
1311         }
1312
1313         spin_unlock(&osd->od_osfs_lock);
1314
1315         if (unlikely(env == NULL))
1316                 OBD_FREE_PTR(ksfs);
1317
1318         /* Reserve a small amount of space for local objects like last_rcvd,
1319          * llog, quota files, ... */
1320         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1321                 sfs->os_bavail = 0;
1322         } else {
1323                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1324                 /** Take out metadata overhead for indirect blocks */
1325                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1326         }
1327
1328         return result;
1329 }
1330
1331 /**
1332  * Estimate space needed for file creations. We assume the largest filename
1333  * which is 2^64 - 1, hence a filename of 20 chars.
1334  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1335  */
1336 #ifdef __LDISKFS_DIR_REC_LEN
1337 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1338 #else
1339 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1340 #endif
1341
1342 /*
1343  * Concurrency: doesn't access mutable data.
1344  */
1345 static void osd_conf_get(const struct lu_env *env,
1346                          const struct dt_device *dev,
1347                          struct dt_device_param *param)
1348 {
1349         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1350         int                ea_overhead;
1351
1352         /*
1353          * XXX should be taken from not-yet-existing fs abstraction layer.
1354          */
1355         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1356         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1357         param->ddp_block_shift  = sb->s_blocksize_bits;
1358         param->ddp_mount_type     = LDD_MT_LDISKFS;
1359         param->ddp_maxbytes       = sb->s_maxbytes;
1360         /* Overhead estimate should be fairly accurate, so we really take a tiny
1361          * error margin which also avoids fragmenting the filesystem too much */
1362         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1363         /* inode are statically allocated, so per-inode space consumption
1364          * is the space consumed by the directory entry */
1365         param->ddp_inodespace     = PER_OBJ_USAGE;
1366         /* per-fragment overhead to be used by the client code */
1367         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1368         param->ddp_mntopts      = 0;
1369         if (test_opt(sb, XATTR_USER))
1370                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1371         if (test_opt(sb, POSIX_ACL))
1372                 param->ddp_mntopts |= MNTOPT_ACL;
1373
1374         /* LOD might calculate the max stripe count based on max_ea_size,
1375          * so we need take account in the overhead as well,
1376          * xattr_header + magic + xattr_entry_head */
1377         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1378                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1379
1380 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1381         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1382                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1383                                                                 ea_overhead;
1384         else
1385 #endif
1386                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1387 }
1388
1389 /*
1390  * Concurrency: shouldn't matter.
1391  */
1392 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1393 {
1394         int rc;
1395
1396         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1397
1398         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1399
1400         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1401                LUSTRE_OSD_LDISKFS_NAME, rc);
1402
1403         return rc;
1404 }
1405
1406 /**
1407  * Start commit for OSD device.
1408  *
1409  * An implementation of dt_commit_async method for OSD device.
1410  * Asychronously starts underlayng fs sync and thereby a transaction
1411  * commit.
1412  *
1413  * \param env environment
1414  * \param d dt device
1415  *
1416  * \see dt_device_operations
1417  */
1418 static int osd_commit_async(const struct lu_env *env,
1419                             struct dt_device *d)
1420 {
1421         struct super_block *s = osd_sb(osd_dt_dev(d));
1422         ENTRY;
1423
1424         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1425         RETURN(s->s_op->sync_fs(s, 0));
1426 }
1427
1428 /*
1429  * Concurrency: shouldn't matter.
1430  */
1431
1432 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1433 {
1434         struct super_block *sb = osd_sb(osd_dt_dev(d));
1435         struct block_device *dev = sb->s_bdev;
1436 #ifdef HAVE_DEV_SET_RDONLY
1437         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1438         int rc = 0;
1439 #else
1440         int rc = -EOPNOTSUPP;
1441 #endif
1442         ENTRY;
1443
1444 #ifdef HAVE_DEV_SET_RDONLY
1445         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1446
1447         if (jdev && (jdev != dev)) {
1448                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1449                        (long)jdev);
1450                 dev_set_rdonly(jdev);
1451         }
1452         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1453         dev_set_rdonly(dev);
1454 #else
1455         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1456                osd_dt_dev(d)->od_svname, (long)dev, rc);
1457 #endif
1458         RETURN(rc);
1459 }
1460
1461 /**
1462  * Note: we do not count into QUOTA here.
1463  * If we mount with --data_journal we may need more.
1464  */
1465 const int osd_dto_credits_noquota[DTO_NR] = {
1466         /**
1467          * Insert.
1468          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1469          * SINGLEDATA_TRANS_BLOCKS(8)
1470          * XXX Note: maybe iam need more, since iam have more level than
1471          *           EXT3 htree.
1472          */
1473         [DTO_INDEX_INSERT]  = 16,
1474         /**
1475          * Delete
1476          * just modify a single entry, probably merge few within a block
1477          */
1478         [DTO_INDEX_DELETE]  = 1,
1479         /**
1480          * Used for OI scrub
1481          */
1482         [DTO_INDEX_UPDATE]  = 16,
1483         /**
1484          * 4(inode, inode bits, groups, GDT)
1485          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1486          */
1487         [DTO_OBJECT_CREATE] = 4,
1488         /**
1489          * 4(inode, inode bits, groups, GDT)
1490          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1491          */
1492         [DTO_OBJECT_DELETE] = 4,
1493         /**
1494          * Attr set credits (inode)
1495          */
1496         [DTO_ATTR_SET_BASE] = 1,
1497         /**
1498          * Xattr set. The same as xattr of EXT3.
1499          * DATA_TRANS_BLOCKS(14)
1500          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1501          * are also counted in. Do not know why?
1502          */
1503         [DTO_XATTR_SET]     = 14,
1504         /**
1505          * credits for inode change during write.
1506          */
1507         [DTO_WRITE_BASE]    = 3,
1508         /**
1509          * credits for single block write.
1510          */
1511         [DTO_WRITE_BLOCK]   = 14,
1512         /**
1513          * Attr set credits for chown.
1514          * This is extra credits for setattr, and it is null without quota
1515          */
1516         [DTO_ATTR_SET_CHOWN] = 0
1517 };
1518
1519 static const struct dt_device_operations osd_dt_ops = {
1520         .dt_root_get       = osd_root_get,
1521         .dt_statfs         = osd_statfs,
1522         .dt_trans_create   = osd_trans_create,
1523         .dt_trans_start    = osd_trans_start,
1524         .dt_trans_stop     = osd_trans_stop,
1525         .dt_trans_cb_add   = osd_trans_cb_add,
1526         .dt_conf_get       = osd_conf_get,
1527         .dt_sync           = osd_sync,
1528         .dt_ro             = osd_ro,
1529         .dt_commit_async   = osd_commit_async,
1530 };
1531
1532 static void osd_object_read_lock(const struct lu_env *env,
1533                                  struct dt_object *dt, unsigned role)
1534 {
1535         struct osd_object *obj = osd_dt_obj(dt);
1536         struct osd_thread_info *oti = osd_oti_get(env);
1537
1538         LINVRNT(osd_invariant(obj));
1539
1540         LASSERT(obj->oo_owner != env);
1541         down_read_nested(&obj->oo_sem, role);
1542
1543         LASSERT(obj->oo_owner == NULL);
1544         oti->oti_r_locks++;
1545 }
1546
1547 static void osd_object_write_lock(const struct lu_env *env,
1548                                   struct dt_object *dt, unsigned role)
1549 {
1550         struct osd_object *obj = osd_dt_obj(dt);
1551         struct osd_thread_info *oti = osd_oti_get(env);
1552
1553         LINVRNT(osd_invariant(obj));
1554
1555         LASSERT(obj->oo_owner != env);
1556         down_write_nested(&obj->oo_sem, role);
1557
1558         LASSERT(obj->oo_owner == NULL);
1559         obj->oo_owner = env;
1560         oti->oti_w_locks++;
1561 }
1562
1563 static void osd_object_read_unlock(const struct lu_env *env,
1564                                    struct dt_object *dt)
1565 {
1566         struct osd_object *obj = osd_dt_obj(dt);
1567         struct osd_thread_info *oti = osd_oti_get(env);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         LASSERT(oti->oti_r_locks > 0);
1572         oti->oti_r_locks--;
1573         up_read(&obj->oo_sem);
1574 }
1575
1576 static void osd_object_write_unlock(const struct lu_env *env,
1577                                     struct dt_object *dt)
1578 {
1579         struct osd_object *obj = osd_dt_obj(dt);
1580         struct osd_thread_info *oti = osd_oti_get(env);
1581
1582         LINVRNT(osd_invariant(obj));
1583
1584         LASSERT(obj->oo_owner == env);
1585         LASSERT(oti->oti_w_locks > 0);
1586         oti->oti_w_locks--;
1587         obj->oo_owner = NULL;
1588         up_write(&obj->oo_sem);
1589 }
1590
1591 static int osd_object_write_locked(const struct lu_env *env,
1592                                    struct dt_object *dt)
1593 {
1594         struct osd_object *obj = osd_dt_obj(dt);
1595
1596         LINVRNT(osd_invariant(obj));
1597
1598         return obj->oo_owner == env;
1599 }
1600
1601 static struct timespec *osd_inode_time(const struct lu_env *env,
1602                                        struct inode *inode, __u64 seconds)
1603 {
1604         struct osd_thread_info  *oti = osd_oti_get(env);
1605         struct timespec         *t   = &oti->oti_time;
1606
1607         t->tv_sec = seconds;
1608         t->tv_nsec = 0;
1609         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1610         return t;
1611 }
1612
1613
1614 static void osd_inode_getattr(const struct lu_env *env,
1615                               struct inode *inode, struct lu_attr *attr)
1616 {
1617         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1618                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1619                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1620                            LA_TYPE;
1621
1622         attr->la_atime   = LTIME_S(inode->i_atime);
1623         attr->la_mtime   = LTIME_S(inode->i_mtime);
1624         attr->la_ctime   = LTIME_S(inode->i_ctime);
1625         attr->la_mode    = inode->i_mode;
1626         attr->la_size    = i_size_read(inode);
1627         attr->la_blocks  = inode->i_blocks;
1628         attr->la_uid     = i_uid_read(inode);
1629         attr->la_gid     = i_gid_read(inode);
1630         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1631         attr->la_nlink   = inode->i_nlink;
1632         attr->la_rdev    = inode->i_rdev;
1633         attr->la_blksize = 1 << inode->i_blkbits;
1634         attr->la_blkbits = inode->i_blkbits;
1635 }
1636
1637 static int osd_attr_get(const struct lu_env *env,
1638                         struct dt_object *dt,
1639                         struct lu_attr *attr)
1640 {
1641         struct osd_object *obj = osd_dt_obj(dt);
1642
1643         if (!dt_object_exists(dt))
1644                 return -ENOENT;
1645
1646         LASSERT(!dt_object_remote(dt));
1647         LINVRNT(osd_invariant(obj));
1648
1649         spin_lock(&obj->oo_guard);
1650         osd_inode_getattr(env, obj->oo_inode, attr);
1651         spin_unlock(&obj->oo_guard);
1652         return 0;
1653 }
1654
1655 static int osd_declare_attr_set(const struct lu_env *env,
1656                                 struct dt_object *dt,
1657                                 const struct lu_attr *attr,
1658                                 struct thandle *handle)
1659 {
1660         struct osd_thandle     *oh;
1661         struct osd_object      *obj;
1662         struct osd_thread_info *info = osd_oti_get(env);
1663         struct lquota_id_info  *qi = &info->oti_qi;
1664         qid_t                   uid;
1665         qid_t                   gid;
1666         long long               bspace;
1667         int                     rc = 0;
1668         bool                    enforce;
1669         ENTRY;
1670
1671         LASSERT(dt != NULL);
1672         LASSERT(handle != NULL);
1673
1674         obj = osd_dt_obj(dt);
1675         LASSERT(osd_invariant(obj));
1676
1677         oh = container_of0(handle, struct osd_thandle, ot_super);
1678         LASSERT(oh->ot_handle == NULL);
1679
1680         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1681                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1682
1683         if (attr == NULL || obj->oo_inode == NULL)
1684                 RETURN(rc);
1685
1686         bspace   = obj->oo_inode->i_blocks;
1687         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1688         bspace   = toqb(bspace);
1689
1690         /* Changing ownership is always preformed by super user, it should not
1691          * fail with EDQUOT.
1692          *
1693          * We still need to call the osd_declare_qid() to calculate the journal
1694          * credits for updating quota accounting files and to trigger quota
1695          * space adjustment once the operation is completed.*/
1696         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1697                 /* USERQUOTA */
1698                 uid = i_uid_read(obj->oo_inode);
1699                 qi->lqi_type = USRQUOTA;
1700                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1701                 /* inode accounting */
1702                 qi->lqi_is_blk = false;
1703
1704                 /* one more inode for the new uid ... */
1705                 qi->lqi_id.qid_uid = attr->la_uid;
1706                 qi->lqi_space      = 1;
1707                 /* Reserve credits for the new uid */
1708                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1709                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1710                         rc = 0;
1711                 if (rc)
1712                         RETURN(rc);
1713
1714                 /* and one less inode for the current uid */
1715                 qi->lqi_id.qid_uid = uid;
1716                 qi->lqi_space      = -1;
1717                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1718                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1719                         rc = 0;
1720                 if (rc)
1721                         RETURN(rc);
1722
1723                 /* block accounting */
1724                 qi->lqi_is_blk = true;
1725
1726                 /* more blocks for the new uid ... */
1727                 qi->lqi_id.qid_uid = attr->la_uid;
1728                 qi->lqi_space      = bspace;
1729                 /*
1730                  * Credits for the new uid has been reserved, re-use "obj"
1731                  * to save credit reservation.
1732                  */
1733                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1734                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1735                         rc = 0;
1736                 if (rc)
1737                         RETURN(rc);
1738
1739                 /* and finally less blocks for the current uid */
1740                 qi->lqi_id.qid_uid = uid;
1741                 qi->lqi_space      = -bspace;
1742                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1743                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1744                         rc = 0;
1745                 if (rc)
1746                         RETURN(rc);
1747
1748                 /* GROUP QUOTA */
1749                 gid = i_gid_read(obj->oo_inode);
1750                 qi->lqi_type = GRPQUOTA;
1751                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1752
1753                 /* inode accounting */
1754                 qi->lqi_is_blk = false;
1755
1756                 /* one more inode for the new gid ... */
1757                 qi->lqi_id.qid_gid = attr->la_gid;
1758                 qi->lqi_space      = 1;
1759                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1760                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1761                         rc = 0;
1762                 if (rc)
1763                         RETURN(rc);
1764
1765                 /* and one less inode for the current gid */
1766                 qi->lqi_id.qid_gid = gid;
1767                 qi->lqi_space      = -1;
1768                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1769                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1770                         rc = 0;
1771                 if (rc)
1772                         RETURN(rc);
1773
1774                 /* block accounting */
1775                 qi->lqi_is_blk = true;
1776
1777                 /* more blocks for the new gid ... */
1778                 qi->lqi_id.qid_gid = attr->la_gid;
1779                 qi->lqi_space      = bspace;
1780                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1781                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1782                         rc = 0;
1783                 if (rc)
1784                         RETURN(rc);
1785
1786                 /* and finally less blocks for the current gid */
1787                 qi->lqi_id.qid_gid = gid;
1788                 qi->lqi_space      = -bspace;
1789                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1790                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1791                         rc = 0;
1792                 if (rc)
1793                         RETURN(rc);
1794         }
1795
1796         RETURN(rc);
1797 }
1798
1799 static int osd_inode_setattr(const struct lu_env *env,
1800                              struct inode *inode, const struct lu_attr *attr)
1801 {
1802         __u64 bits = attr->la_valid;
1803
1804         /* Only allow set size for regular file */
1805         if (!S_ISREG(inode->i_mode))
1806                 bits &= ~(LA_SIZE | LA_BLOCKS);
1807
1808         if (bits == 0)
1809                 return 0;
1810
1811         if (bits & LA_ATIME)
1812                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1813         if (bits & LA_CTIME)
1814                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1815         if (bits & LA_MTIME)
1816                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1817         if (bits & LA_SIZE) {
1818                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1819                 i_size_write(inode, attr->la_size);
1820         }
1821
1822 #if 0
1823         /* OSD should not change "i_blocks" which is used by quota.
1824          * "i_blocks" should be changed by ldiskfs only. */
1825         if (bits & LA_BLOCKS)
1826                 inode->i_blocks = attr->la_blocks;
1827 #endif
1828         if (bits & LA_MODE)
1829                 inode->i_mode = (inode->i_mode & S_IFMT) |
1830                                 (attr->la_mode & ~S_IFMT);
1831         if (bits & LA_UID)
1832                 i_uid_write(inode, attr->la_uid);
1833         if (bits & LA_GID)
1834                 i_gid_write(inode, attr->la_gid);
1835         if (bits & LA_NLINK)
1836                 set_nlink(inode, attr->la_nlink);
1837         if (bits & LA_RDEV)
1838                 inode->i_rdev = attr->la_rdev;
1839
1840         if (bits & LA_FLAGS) {
1841                 /* always keep S_NOCMTIME */
1842                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1843                                  S_NOCMTIME;
1844         }
1845         return 0;
1846 }
1847
1848 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1849 {
1850         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1851             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1852                 struct iattr    iattr;
1853                 int             rc;
1854
1855                 ll_vfs_dq_init(inode);
1856                 iattr.ia_valid = 0;
1857                 if (attr->la_valid & LA_UID)
1858                         iattr.ia_valid |= ATTR_UID;
1859                 if (attr->la_valid & LA_GID)
1860                         iattr.ia_valid |= ATTR_GID;
1861                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1862                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1863
1864                 rc = ll_vfs_dq_transfer(inode, &iattr);
1865                 if (rc) {
1866                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1867                                "enforcement enabled on the ldiskfs "
1868                                "filesystem?\n", inode->i_sb->s_id, rc);
1869                         return rc;
1870                 }
1871         }
1872         return 0;
1873 }
1874
1875 static int osd_attr_set(const struct lu_env *env,
1876                         struct dt_object *dt,
1877                         const struct lu_attr *attr,
1878                         struct thandle *handle)
1879 {
1880         struct osd_object *obj = osd_dt_obj(dt);
1881         struct inode      *inode;
1882         int rc;
1883
1884         if (!dt_object_exists(dt))
1885                 return -ENOENT;
1886
1887         LASSERT(handle != NULL);
1888         LASSERT(!dt_object_remote(dt));
1889         LASSERT(osd_invariant(obj));
1890
1891         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1892
1893         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1894                 struct osd_thread_info  *oti  = osd_oti_get(env);
1895                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1896                 struct lu_fid           *fid1 = &oti->oti_fid;
1897                 struct osd_inode_id     *id   = &oti->oti_id;
1898                 struct iam_path_descr   *ipd;
1899                 struct iam_container    *bag;
1900                 struct osd_thandle      *oh;
1901                 int                      rc;
1902
1903                 fid_cpu_to_be(fid1, fid0);
1904                 memset(id, 1, sizeof(*id));
1905                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1906                                   fid0)->oi_dir.od_container;
1907                 ipd = osd_idx_ipd_get(env, bag);
1908                 if (unlikely(ipd == NULL))
1909                         RETURN(-ENOMEM);
1910
1911                 oh = container_of0(handle, struct osd_thandle, ot_super);
1912                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1913                                 (const struct iam_rec *)id, ipd);
1914                 osd_ipd_put(env, bag, ipd);
1915                 return(rc > 0 ? 0 : rc);
1916         }
1917
1918         inode = obj->oo_inode;
1919
1920         rc = osd_quota_transfer(inode, attr);
1921         if (rc)
1922                 return rc;
1923
1924         spin_lock(&obj->oo_guard);
1925         rc = osd_inode_setattr(env, inode, attr);
1926         spin_unlock(&obj->oo_guard);
1927
1928         if (!rc)
1929                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1930         return rc;
1931 }
1932
1933 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1934                                            struct osd_object *obj,
1935                                            const char *name, const int namelen)
1936 {
1937         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1938 }
1939
1940 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1941                       umode_t mode, struct dt_allocation_hint *hint,
1942                       struct thandle *th)
1943 {
1944         int result;
1945         struct osd_device  *osd = osd_obj2dev(obj);
1946         struct osd_thandle *oth;
1947         struct dt_object   *parent = NULL;
1948         struct inode       *inode;
1949
1950         LINVRNT(osd_invariant(obj));
1951         LASSERT(obj->oo_inode == NULL);
1952         LASSERT(obj->oo_hl_head == NULL);
1953
1954         if (S_ISDIR(mode) && ldiskfs_pdo) {
1955                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1956                 if (obj->oo_hl_head == NULL)
1957                         return -ENOMEM;
1958         }
1959
1960         oth = container_of(th, struct osd_thandle, ot_super);
1961         LASSERT(oth->ot_handle->h_transaction != NULL);
1962
1963         if (hint && hint->dah_parent)
1964                 parent = hint->dah_parent;
1965
1966         inode = ldiskfs_create_inode(oth->ot_handle,
1967                                      parent ? osd_dt_obj(parent)->oo_inode :
1968                                               osd_sb(osd)->s_root->d_inode,
1969                                      mode);
1970         if (!IS_ERR(inode)) {
1971                 /* Do not update file c/mtime in ldiskfs. */
1972                 inode->i_flags |= S_NOCMTIME;
1973
1974                 /* For new created object, it must be consistent,
1975                  * and it is unnecessary to scrub against it. */
1976                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1977
1978                 obj->oo_inode = inode;
1979                 result = 0;
1980         } else {
1981                 if (obj->oo_hl_head != NULL) {
1982                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1983                         obj->oo_hl_head = NULL;
1984                 }
1985                 result = PTR_ERR(inode);
1986         }
1987         LINVRNT(osd_invariant(obj));
1988         return result;
1989 }
1990
1991 enum {
1992         OSD_NAME_LEN = 255
1993 };
1994
1995 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1996                      struct lu_attr *attr,
1997                      struct dt_allocation_hint *hint,
1998                      struct dt_object_format *dof,
1999                      struct thandle *th)
2000 {
2001         int result;
2002         struct osd_thandle *oth;
2003         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2004
2005         LASSERT(S_ISDIR(attr->la_mode));
2006
2007         oth = container_of(th, struct osd_thandle, ot_super);
2008         LASSERT(oth->ot_handle->h_transaction != NULL);
2009         result = osd_mkfile(info, obj, mode, hint, th);
2010
2011         return result;
2012 }
2013
2014 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2015                         struct lu_attr *attr,
2016                         struct dt_allocation_hint *hint,
2017                         struct dt_object_format *dof,
2018                         struct thandle *th)
2019 {
2020         int result;
2021         struct osd_thandle *oth;
2022         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2023
2024         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2025
2026         LASSERT(S_ISREG(attr->la_mode));
2027
2028         oth = container_of(th, struct osd_thandle, ot_super);
2029         LASSERT(oth->ot_handle->h_transaction != NULL);
2030
2031         result = osd_mkfile(info, obj, mode, hint, th);
2032         if (result == 0) {
2033                 LASSERT(obj->oo_inode != NULL);
2034                 if (feat->dif_flags & DT_IND_VARKEY)
2035                         result = iam_lvar_create(obj->oo_inode,
2036                                                  feat->dif_keysize_max,
2037                                                  feat->dif_ptrsize,
2038                                                  feat->dif_recsize_max,
2039                                                  oth->ot_handle);
2040                 else
2041                         result = iam_lfix_create(obj->oo_inode,
2042                                                  feat->dif_keysize_max,
2043                                                  feat->dif_ptrsize,
2044                                                  feat->dif_recsize_max,
2045                                                  oth->ot_handle);
2046
2047         }
2048         return result;
2049 }
2050
2051 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2052                      struct lu_attr *attr,
2053                      struct dt_allocation_hint *hint,
2054                      struct dt_object_format *dof,
2055                      struct thandle *th)
2056 {
2057         LASSERT(S_ISREG(attr->la_mode));
2058         return osd_mkfile(info, obj, (attr->la_mode &
2059                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2060 }
2061
2062 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2063                      struct lu_attr *attr,
2064                      struct dt_allocation_hint *hint,
2065                      struct dt_object_format *dof,
2066                      struct thandle *th)
2067 {
2068         LASSERT(S_ISLNK(attr->la_mode));
2069         return osd_mkfile(info, obj, (attr->la_mode &
2070                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2071 }
2072
2073 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2074                      struct lu_attr *attr,
2075                      struct dt_allocation_hint *hint,
2076                      struct dt_object_format *dof,
2077                      struct thandle *th)
2078 {
2079         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2080         int result;
2081
2082         LINVRNT(osd_invariant(obj));
2083         LASSERT(obj->oo_inode == NULL);
2084         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2085                 S_ISFIFO(mode) || S_ISSOCK(mode));
2086
2087         result = osd_mkfile(info, obj, mode, hint, th);
2088         if (result == 0) {
2089                 LASSERT(obj->oo_inode != NULL);
2090                 /*
2091                  * This inode should be marked dirty for i_rdev.  Currently
2092                  * that is done in the osd_attr_init().
2093                  */
2094                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2095                                    attr->la_rdev);
2096         }
2097         LINVRNT(osd_invariant(obj));
2098         return result;
2099 }
2100
2101 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2102                               struct lu_attr *,
2103                               struct dt_allocation_hint *hint,
2104                               struct dt_object_format *dof,
2105                               struct thandle *);
2106
2107 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2108 {
2109         osd_obj_type_f result;
2110
2111         switch (type) {
2112         case DFT_DIR:
2113                 result = osd_mkdir;
2114                 break;
2115         case DFT_REGULAR:
2116                 result = osd_mkreg;
2117                 break;
2118         case DFT_SYM:
2119                 result = osd_mksym;
2120                 break;
2121         case DFT_NODE:
2122                 result = osd_mknod;
2123                 break;
2124         case DFT_INDEX:
2125                 result = osd_mk_index;
2126                 break;
2127
2128         default:
2129                 LBUG();
2130                 break;
2131         }
2132         return result;
2133 }
2134
2135
2136 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2137                         struct dt_object *parent, struct dt_object *child,
2138                         umode_t child_mode)
2139 {
2140         LASSERT(ah);
2141
2142         ah->dah_parent = parent;
2143         ah->dah_mode = child_mode;
2144 }
2145
2146 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2147                           struct lu_attr *attr, struct dt_object_format *dof)
2148 {
2149         struct inode   *inode = obj->oo_inode;
2150         __u64           valid = attr->la_valid;
2151         int             result;
2152
2153         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2154
2155         if (dof->dof_type != DFT_NODE)
2156                 attr->la_valid &= ~LA_RDEV;
2157         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2158                 attr->la_valid &= ~LA_ATIME;
2159         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2160                 attr->la_valid &= ~LA_CTIME;
2161         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2162                 attr->la_valid &= ~LA_MTIME;
2163
2164         result = osd_quota_transfer(inode, attr);
2165         if (result)
2166                 return;
2167
2168         if (attr->la_valid != 0) {
2169                 result = osd_inode_setattr(info->oti_env, inode, attr);
2170                 /*
2171                  * The osd_inode_setattr() should always succeed here.  The
2172                  * only error that could be returned is EDQUOT when we are
2173                  * trying to change the UID or GID of the inode. However, this
2174                  * should not happen since quota enforcement is no longer
2175                  * enabled on ldiskfs (lquota takes care of it).
2176                  */
2177                 LASSERTF(result == 0, "%d\n", result);
2178                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2179         }
2180
2181         attr->la_valid = valid;
2182 }
2183
2184 /**
2185  * Helper function for osd_object_create()
2186  *
2187  * \retval 0, on success
2188  */
2189 static int __osd_object_create(struct osd_thread_info *info,
2190                                struct osd_object *obj, struct lu_attr *attr,
2191                                struct dt_allocation_hint *hint,
2192                                struct dt_object_format *dof,
2193                                struct thandle *th)
2194 {
2195         int     result;
2196         __u32   umask;
2197
2198         /* we drop umask so that permissions we pass are not affected */
2199         umask = current->fs->umask;
2200         current->fs->umask = 0;
2201
2202         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2203                                                   th);
2204         if (result == 0) {
2205                 osd_attr_init(info, obj, attr, dof);
2206                 osd_object_init0(obj);
2207         }
2208
2209         if (obj->oo_inode != NULL) {
2210                 LASSERT(obj->oo_inode->i_state & I_NEW);
2211
2212                 unlock_new_inode(obj->oo_inode);
2213         }
2214
2215         /* restore previous umask value */
2216         current->fs->umask = umask;
2217
2218         return result;
2219 }
2220
2221 /**
2222  * Helper function for osd_object_create()
2223  *
2224  * \retval 0, on success
2225  */
2226 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2227                            const struct lu_fid *fid, struct thandle *th)
2228 {
2229         struct osd_thread_info *info = osd_oti_get(env);
2230         struct osd_inode_id    *id   = &info->oti_id;
2231         struct osd_device      *osd  = osd_obj2dev(obj);
2232         struct osd_thandle     *oh;
2233
2234         LASSERT(obj->oo_inode != NULL);
2235
2236         oh = container_of0(th, struct osd_thandle, ot_super);
2237         LASSERT(oh->ot_handle);
2238
2239         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2240         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2241 }
2242
2243 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2244                    u64 seq, struct lu_seq_range *range)
2245 {
2246         struct seq_server_site  *ss = osd_seq_site(osd);
2247
2248         if (fid_seq_is_idif(seq)) {
2249                 fld_range_set_ost(range);
2250                 range->lsr_index = idif_ost_idx(seq);
2251                 return 0;
2252         }
2253
2254         if (!fid_seq_in_fldb(seq)) {
2255                 fld_range_set_mdt(range);
2256                 if (ss != NULL)
2257                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2258                          * at all */
2259                         range->lsr_index = ss->ss_node_id;
2260                 return 0;
2261         }
2262
2263         LASSERT(ss != NULL);
2264         fld_range_set_any(range);
2265         /* OSD will only do local fld lookup */
2266         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2267 }
2268
2269 /*
2270  * Concurrency: no external locking is necessary.
2271  */
2272 static int osd_declare_object_create(const struct lu_env *env,
2273                                      struct dt_object *dt,
2274                                      struct lu_attr *attr,
2275                                      struct dt_allocation_hint *hint,
2276                                      struct dt_object_format *dof,
2277                                      struct thandle *handle)
2278 {
2279         struct osd_thandle      *oh;
2280         int                      rc;
2281         ENTRY;
2282
2283         LASSERT(handle != NULL);
2284
2285         oh = container_of0(handle, struct osd_thandle, ot_super);
2286         LASSERT(oh->ot_handle == NULL);
2287
2288         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2289                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2290         /* Reuse idle OI block may cause additional one OI block
2291          * to be changed. */
2292         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2293                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2294
2295         /* If this is directory, then we expect . and .. to be inserted as
2296          * well. The one directory block always needs to be created for the
2297          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2298          * block), there is no danger of needing a tree for the first block.
2299          */
2300         if (attr && S_ISDIR(attr->la_mode)) {
2301                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2302                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2303                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2304         }
2305
2306         if (!attr)
2307                 RETURN(0);
2308
2309         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2310                                    osd_dt_obj(dt), false, NULL, false);
2311         if (rc != 0)
2312                 RETURN(rc);
2313
2314         RETURN(rc);
2315 }
2316
2317 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2318                              struct lu_attr *attr,
2319                              struct dt_allocation_hint *hint,
2320                              struct dt_object_format *dof, struct thandle *th)
2321 {
2322         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2323         struct osd_object       *obj    = osd_dt_obj(dt);
2324         struct osd_thread_info  *info   = osd_oti_get(env);
2325         int result;
2326         ENTRY;
2327
2328         if (dt_object_exists(dt))
2329                 return -EEXIST;
2330
2331         LINVRNT(osd_invariant(obj));
2332         LASSERT(!dt_object_remote(dt));
2333         LASSERT(osd_write_locked(env, obj));
2334         LASSERT(th != NULL);
2335
2336         if (unlikely(fid_is_acct(fid)))
2337                 /* Quota files can't be created from the kernel any more,
2338                  * 'tune2fs -O quota' will take care of creating them */
2339                 RETURN(-EPERM);
2340
2341         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2342         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2343
2344         result = __osd_object_create(info, obj, attr, hint, dof, th);
2345         if (result == 0)
2346                 result = __osd_oi_insert(env, obj, fid, th);
2347
2348         LASSERT(ergo(result == 0,
2349                      dt_object_exists(dt) && !dt_object_remote(dt)));
2350
2351         LASSERT(osd_invariant(obj));
2352         RETURN(result);
2353 }
2354
2355 /**
2356  * Called to destroy on-disk representation of the object
2357  *
2358  * Concurrency: must be locked
2359  */
2360 static int osd_declare_object_destroy(const struct lu_env *env,
2361                                       struct dt_object *dt,
2362                                       struct thandle *th)
2363 {
2364         struct osd_object  *obj = osd_dt_obj(dt);
2365         struct inode       *inode = obj->oo_inode;
2366         struct osd_thandle *oh;
2367         int                 rc;
2368         ENTRY;
2369
2370         oh = container_of0(th, struct osd_thandle, ot_super);
2371         LASSERT(oh->ot_handle == NULL);
2372         LASSERT(inode);
2373
2374         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2375                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2376         /* Recycle idle OI leaf may cause additional three OI blocks
2377          * to be changed. */
2378         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2379                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2380         /* one less inode */
2381         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2382                                    -1, oh, obj, false, NULL, false);
2383         if (rc)
2384                 RETURN(rc);
2385         /* data to be truncated */
2386         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2387                                    0, oh, obj, true, NULL, false);
2388         RETURN(rc);
2389 }
2390
2391 static int osd_object_destroy(const struct lu_env *env,
2392                               struct dt_object *dt,
2393                               struct thandle *th)
2394 {
2395         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2396         struct osd_object      *obj = osd_dt_obj(dt);
2397         struct inode           *inode = obj->oo_inode;
2398         struct osd_device      *osd = osd_obj2dev(obj);
2399         struct osd_thandle     *oh;
2400         int                     result;
2401         ENTRY;
2402
2403         oh = container_of0(th, struct osd_thandle, ot_super);
2404         LASSERT(oh->ot_handle);
2405         LASSERT(inode);
2406         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2407
2408         if (unlikely(fid_is_acct(fid)))
2409                 RETURN(-EPERM);
2410
2411         if (S_ISDIR(inode->i_mode)) {
2412                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2413                         inode->i_nlink == 2);
2414                 /* it will check/delete the inode from remote parent,
2415                  * how to optimize it? unlink performance impaction XXX */
2416                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2417                 if (result != 0 && result != -ENOENT) {
2418                         CERROR("%s: delete inode "DFID": rc = %d\n",
2419                                osd_name(osd), PFID(fid), result);
2420                 }
2421                 spin_lock(&obj->oo_guard);
2422                 clear_nlink(inode);
2423                 spin_unlock(&obj->oo_guard);
2424                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2425         }
2426
2427         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2428
2429         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2430                                OI_CHECK_FLD);
2431
2432         /* XXX: add to ext3 orphan list */
2433         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2434
2435         /* not needed in the cache anymore */
2436         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2437
2438         RETURN(0);
2439 }
2440
2441 /**
2442  * Put the fid into lustre_mdt_attrs, and then place the structure
2443  * inode's ea. This fid should not be altered during the life time
2444  * of the inode.
2445  *
2446  * \retval +ve, on success
2447  * \retval -ve, on error
2448  *
2449  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2450  */
2451 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2452                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2453 {
2454         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2455         int                      rc;
2456         ENTRY;
2457
2458         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2459                 RETURN(0);
2460
2461         lustre_lma_init(lma, fid, compat, incompat);
2462         lustre_lma_swab(lma);
2463
2464         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2465                              XATTR_CREATE);
2466         /* LMA may already exist, but we need to check that all the
2467          * desired compat/incompat flags have been added. */
2468         if (unlikely(rc == -EEXIST)) {
2469                 if (compat == 0 && incompat == 0)
2470                         RETURN(0);
2471
2472                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2473                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2474                                      LMA_OLD_SIZE);
2475                 if (rc <= 0)
2476                         RETURN(-EINVAL);
2477
2478                 lustre_lma_swab(lma);
2479                 if (!(~lma->lma_compat & compat) &&
2480                     !(~lma->lma_incompat & incompat))
2481                         RETURN(0);
2482
2483                 lma->lma_compat |= compat;
2484                 lma->lma_incompat |= incompat;
2485                 lustre_lma_swab(lma);
2486                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2487                                      sizeof(*lma), XATTR_REPLACE);
2488         }
2489
2490         RETURN(rc);
2491 }
2492
2493 /**
2494  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2495  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2496  * To have compatilibility with 1.8 ldiskfs driver we need to have
2497  * magic number at start of fid data.
2498  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2499  * its inmemory API.
2500  */
2501 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2502                                          const struct lu_fid *fid)
2503 {
2504         if (!fid_is_namespace_visible(fid) ||
2505             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2506                 param->edp_magic = 0;
2507                 return;
2508         }
2509
2510         param->edp_magic = LDISKFS_LUFID_MAGIC;
2511         param->edp_len =  sizeof(struct lu_fid) + 1;
2512         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2513 }
2514
2515 /**
2516  * Try to read the fid from inode ea into dt_rec.
2517  *
2518  * \param fid object fid.
2519  *
2520  * \retval 0 on success
2521  */
2522 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2523                           __u32 ino, struct lu_fid *fid,
2524                           struct osd_inode_id *id)
2525 {
2526         struct osd_thread_info *info  = osd_oti_get(env);
2527         struct inode           *inode;
2528         ENTRY;
2529
2530         osd_id_gen(id, ino, OSD_OII_NOGEN);
2531         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2532         if (IS_ERR(inode))
2533                 RETURN(PTR_ERR(inode));
2534
2535         iput(inode);
2536         RETURN(0);
2537 }
2538
2539 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2540                                         struct inode *dir,
2541                                         struct inode *parent_dir,
2542                                         const struct lu_fid *dot_fid,
2543                                         const struct lu_fid *dot_dot_fid,
2544                                         struct osd_thandle *oth)
2545 {
2546         struct ldiskfs_dentry_param *dot_ldp;
2547         struct ldiskfs_dentry_param *dot_dot_ldp;
2548
2549         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2550         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2551
2552         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2553         dot_ldp->edp_magic = 0;
2554         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2555                                         dir, dot_ldp, dot_dot_ldp);
2556 }
2557
2558 /**
2559  * Create an local agent inode for remote entry
2560  */
2561 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2562                                                   struct osd_device *osd,
2563                                                   struct osd_object *pobj,
2564                                                   const struct lu_fid *fid,
2565                                                   __u32 type,
2566                                                   struct thandle *th)
2567 {
2568         struct osd_thread_info  *info = osd_oti_get(env);
2569         struct inode            *local;
2570         struct osd_thandle      *oh;
2571         int                     rc;
2572         ENTRY;
2573
2574         LASSERT(th);
2575         oh = container_of(th, struct osd_thandle, ot_super);
2576         LASSERT(oh->ot_handle->h_transaction != NULL);
2577
2578         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2579         if (IS_ERR(local)) {
2580                 CERROR("%s: create local error %d\n", osd_name(osd),
2581                        (int)PTR_ERR(local));
2582                 RETURN(local);
2583         }
2584
2585         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2586         unlock_new_inode(local);
2587
2588         /* Set special LMA flag for local agent inode */
2589         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2590         if (rc != 0) {
2591                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2592                        osd_name(osd), PFID(fid), rc);
2593                 RETURN(ERR_PTR(rc));
2594         }
2595
2596         if (!S_ISDIR(type))
2597                 RETURN(local);
2598
2599         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2600                                          lu_object_fid(&pobj->oo_dt.do_lu),
2601                                          fid, oh);
2602         if (rc != 0) {
2603                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2604                         osd_name(osd), PFID(fid), rc);
2605                 RETURN(ERR_PTR(rc));
2606         }
2607
2608         RETURN(local);
2609 }
2610
2611 /**
2612  * Delete local agent inode for remote entry
2613  */
2614 static int osd_delete_local_agent_inode(const struct lu_env *env,
2615                                         struct osd_device *osd,
2616                                         const struct lu_fid *fid,
2617                                         __u32 ino, struct osd_thandle *oh)
2618 {
2619         struct osd_thread_info  *oti = osd_oti_get(env);
2620         struct osd_inode_id     *id = &oti->oti_id;
2621         struct inode            *inode;
2622         ENTRY;
2623
2624         id->oii_ino = le32_to_cpu(ino);
2625         id->oii_gen = OSD_OII_NOGEN;
2626         inode = osd_iget(oti, osd, id);
2627         if (IS_ERR(inode)) {
2628                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2629                        PFID(fid), id->oii_ino, id->oii_gen);
2630                 RETURN(PTR_ERR(inode));
2631         }
2632
2633         clear_nlink(inode);
2634         mark_inode_dirty(inode);
2635         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2636                 osd_name(osd), PFID(fid), inode->i_ino);
2637         iput(inode);
2638         RETURN(0);
2639 }
2640
2641 /**
2642  * OSD layer object create function for interoperability mode (b11826).
2643  * This is mostly similar to osd_object_create(). Only difference being, fid is
2644  * inserted into inode ea here.
2645  *
2646  * \retval   0, on success
2647  * \retval -ve, on error
2648  */
2649 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2650                                 struct lu_attr *attr,
2651                                 struct dt_allocation_hint *hint,
2652                                 struct dt_object_format *dof,
2653                                 struct thandle *th)
2654 {
2655         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2656         struct osd_object       *obj    = osd_dt_obj(dt);
2657         struct osd_thread_info  *info   = osd_oti_get(env);
2658         int                      result;
2659
2660         ENTRY;
2661
2662         if (dt_object_exists(dt))
2663                 RETURN(-EEXIST);
2664
2665         LASSERT(osd_invariant(obj));
2666         LASSERT(!dt_object_remote(dt));
2667         LASSERT(osd_write_locked(env, obj));
2668         LASSERT(th != NULL);
2669
2670         if (unlikely(fid_is_acct(fid)))
2671                 /* Quota files can't be created from the kernel any more,
2672                  * 'tune2fs -O quota' will take care of creating them */
2673                 RETURN(-EPERM);
2674
2675         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2676         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2677
2678         result = __osd_object_create(info, obj, attr, hint, dof, th);
2679         if (result == 0) {
2680                 if (fid_is_idif(fid) &&
2681                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2682                         struct lu_fid *tfid = &info->oti_fid;
2683                         struct ost_id *oi   = &info->oti_ostid;
2684
2685                         fid_to_ostid(fid, oi);
2686                         ostid_to_fid(tfid, oi, 0);
2687                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2688                                                 LMAC_FID_ON_OST, 0);
2689                 } else {
2690                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2691                                 fid_is_on_ost(info, osd_obj2dev(obj),
2692                                               fid, OI_CHECK_FLD) ?
2693                                 LMAC_FID_ON_OST : 0, 0);
2694                 }
2695         }
2696
2697         if (result == 0)
2698                 result = __osd_oi_insert(env, obj, fid, th);
2699
2700         LASSERT(ergo(result == 0,
2701                      dt_object_exists(dt) && !dt_object_remote(dt)));
2702         LINVRNT(osd_invariant(obj));
2703         RETURN(result);
2704 }
2705
2706 static int osd_declare_object_ref_add(const struct lu_env *env,
2707                                       struct dt_object *dt,
2708                                       struct thandle *handle)
2709 {
2710         struct osd_thandle       *oh;
2711
2712         /* it's possible that object doesn't exist yet */
2713         LASSERT(handle != NULL);
2714
2715         oh = container_of0(handle, struct osd_thandle, ot_super);
2716         LASSERT(oh->ot_handle == NULL);
2717
2718         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2719                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2720
2721         return 0;
2722 }
2723
2724 /*
2725  * Concurrency: @dt is write locked.
2726  */
2727 static int osd_object_ref_add(const struct lu_env *env,
2728                               struct dt_object *dt, struct thandle *th)
2729 {
2730         struct osd_object  *obj = osd_dt_obj(dt);
2731         struct inode       *inode = obj->oo_inode;
2732         struct osd_thandle *oh;
2733         int                 rc = 0;
2734
2735         if (!dt_object_exists(dt))
2736                 return -ENOENT;
2737
2738         LINVRNT(osd_invariant(obj));
2739         LASSERT(!dt_object_remote(dt));
2740         LASSERT(osd_write_locked(env, obj));
2741         LASSERT(th != NULL);
2742
2743         oh = container_of0(th, struct osd_thandle, ot_super);
2744         LASSERT(oh->ot_handle != NULL);
2745
2746         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2747
2748         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2749                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2750         /*
2751          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2752          * (65000) subdirectories by storing "1" in i_nlink if the link count
2753          * would otherwise overflow. Directory tranversal tools understand
2754          * that (st_nlink == 1) indicates that the filesystem dose not track
2755          * hard links count on the directory, and will not abort subdirectory
2756          * scanning early once (st_nlink - 2) subdirs have been found.
2757          *
2758          * This also has to properly handle the case of inodes with nlink == 0
2759          * in case they are being linked into the PENDING directory
2760          */
2761         spin_lock(&obj->oo_guard);
2762         if (unlikely(inode->i_nlink == 0))
2763                 /* inc_nlink from 0 may cause WARN_ON */
2764                 set_nlink(inode, 1);
2765         else {
2766                 ldiskfs_inc_count(oh->ot_handle, inode);
2767                 if (!S_ISDIR(inode->i_mode))
2768                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2769         }
2770         spin_unlock(&obj->oo_guard);
2771
2772         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2773         LINVRNT(osd_invariant(obj));
2774
2775         return rc;
2776 }
2777
2778 static int osd_declare_object_ref_del(const struct lu_env *env,
2779                                       struct dt_object *dt,
2780                                       struct thandle *handle)
2781 {
2782         struct osd_thandle *oh;
2783
2784         LASSERT(!dt_object_remote(dt));
2785         LASSERT(handle != NULL);
2786
2787         oh = container_of0(handle, struct osd_thandle, ot_super);
2788         LASSERT(oh->ot_handle == NULL);
2789
2790         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2791                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2792
2793         return 0;
2794 }
2795
2796 /*
2797  * Concurrency: @dt is write locked.
2798  */
2799 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2800                               struct thandle *th)
2801 {
2802         struct osd_object       *obj = osd_dt_obj(dt);
2803         struct inode            *inode = obj->oo_inode;
2804         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2805         struct osd_thandle      *oh;
2806
2807         if (!dt_object_exists(dt))
2808                 return -ENOENT;
2809
2810         LINVRNT(osd_invariant(obj));
2811         LASSERT(!dt_object_remote(dt));
2812         LASSERT(osd_write_locked(env, obj));
2813         LASSERT(th != NULL);
2814
2815         oh = container_of0(th, struct osd_thandle, ot_super);
2816         LASSERT(oh->ot_handle != NULL);
2817
2818         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2819
2820         spin_lock(&obj->oo_guard);
2821         /* That can be result of upgrade from old Lustre version and
2822          * applied only to local files.  Just skip this ref_del call.
2823          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2824         if (inode->i_nlink == 0) {
2825                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2826                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2827                              ", maybe an upgraded file? (LU-3915)\n",
2828                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2829                 spin_unlock(&obj->oo_guard);
2830                 return 0;
2831         }
2832
2833         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2834                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2835
2836         ldiskfs_dec_count(oh->ot_handle, inode);
2837         spin_unlock(&obj->oo_guard);
2838
2839         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2840         LINVRNT(osd_invariant(obj));
2841
2842         return 0;
2843 }
2844
2845 /*
2846  * Get the 64-bit version for an inode.
2847  */
2848 static int osd_object_version_get(const struct lu_env *env,
2849                                   struct dt_object *dt, dt_obj_version_t *ver)
2850 {
2851         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2852
2853         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2854                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2855         *ver = LDISKFS_I(inode)->i_fs_version;
2856         return 0;
2857 }
2858
2859 /*
2860  * Concurrency: @dt is read locked.
2861  */
2862 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2863                          struct lu_buf *buf, const char *name)
2864 {
2865         struct osd_object      *obj    = osd_dt_obj(dt);
2866         struct inode           *inode  = obj->oo_inode;
2867         struct osd_thread_info *info   = osd_oti_get(env);
2868         struct dentry          *dentry = &info->oti_obj_dentry;
2869
2870         /* version get is not real XATTR but uses xattr API */
2871         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2872                 /* for version we are just using xattr API but change inode
2873                  * field instead */
2874                 if (buf->lb_len == 0)
2875                         return sizeof(dt_obj_version_t);
2876
2877                 if (buf->lb_len < sizeof(dt_obj_version_t))
2878                         return -ERANGE;
2879
2880                 osd_object_version_get(env, dt, buf->lb_buf);
2881
2882                 return sizeof(dt_obj_version_t);
2883         }
2884
2885         if (!dt_object_exists(dt))
2886                 return -ENOENT;
2887
2888         LASSERT(!dt_object_remote(dt));
2889         LASSERT(inode->i_op != NULL);
2890         LASSERT(inode->i_op->getxattr != NULL);
2891
2892         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2893 }
2894
2895
2896 static int osd_declare_xattr_set(const struct lu_env *env,
2897                                  struct dt_object *dt,
2898                                  const struct lu_buf *buf, const char *name,
2899                                  int fl, struct thandle *handle)
2900 {
2901         struct osd_thandle *oh;
2902         int credits;
2903         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2904
2905         LASSERT(handle != NULL);
2906
2907         oh = container_of0(handle, struct osd_thandle, ot_super);
2908         LASSERT(oh->ot_handle == NULL);
2909
2910         /* optimistic optimization: LMA is set first and usually fit inode */
2911         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2912                 if (dt_object_exists(dt))
2913                         credits = 0;
2914                 else
2915                         credits = 1;
2916         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2917                 credits = 1;
2918         } else {
2919                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2920                 if (buf && buf->lb_len > sb->s_blocksize) {
2921                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2922                                         sb->s_blocksize_bits;
2923                 }
2924                 /*
2925                  * xattr set may involve inode quota change, reserve credits for
2926                  * dquot_initialize()
2927                  */
2928                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2929         }
2930
2931         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2932
2933         return 0;
2934 }
2935
2936 /*
2937  * Set the 64-bit version for object
2938  */
2939 static void osd_object_version_set(const struct lu_env *env,
2940                                    struct dt_object *dt,
2941                                    dt_obj_version_t *new_version)
2942 {
2943         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2944
2945         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2946                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2947
2948         LDISKFS_I(inode)->i_fs_version = *new_version;
2949         /** Version is set after all inode operations are finished,
2950          *  so we should mark it dirty here */
2951         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2952 }
2953
2954 /*
2955  * Concurrency: @dt is write locked.
2956  */
2957 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2958                          const struct lu_buf *buf, const char *name, int fl,
2959                          struct thandle *handle)
2960 {
2961         struct osd_object      *obj      = osd_dt_obj(dt);
2962         struct inode           *inode    = obj->oo_inode;
2963         struct osd_thread_info *info     = osd_oti_get(env);
2964         int                     fs_flags = 0;
2965         ENTRY;
2966
2967         LASSERT(handle != NULL);
2968
2969         /* version set is not real XATTR */
2970         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2971                 /* for version we are just using xattr API but change inode
2972                  * field instead */
2973                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2974                 osd_object_version_set(env, dt, buf->lb_buf);
2975                 return sizeof(dt_obj_version_t);
2976         }
2977
2978         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2979                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2980
2981         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2982         if (fl & LU_XATTR_REPLACE)
2983                 fs_flags |= XATTR_REPLACE;
2984
2985         if (fl & LU_XATTR_CREATE)
2986                 fs_flags |= XATTR_CREATE;
2987
2988         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2989                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2990                 int                      rc;
2991
2992                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2993                 if (rc != 0)
2994                         RETURN(rc);
2995
2996                 lma->lma_incompat |= LMAI_STRIPED;
2997                 lustre_lma_swab(lma);
2998                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2999                                      sizeof(*lma), XATTR_REPLACE);
3000                 if (rc != 0)
3001                         RETURN(rc);
3002         }
3003
3004         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3005             strcmp(name, XATTR_NAME_LINK) == 0)
3006                 return -ENOSPC;
3007
3008         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3009                                fs_flags);
3010 }
3011
3012 /*
3013  * Concurrency: @dt is read locked.
3014  */
3015 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3016                           const struct lu_buf *buf)
3017 {
3018         struct osd_object      *obj    = osd_dt_obj(dt);
3019         struct inode           *inode  = obj->oo_inode;
3020         struct osd_thread_info *info   = osd_oti_get(env);
3021         struct dentry          *dentry = &info->oti_obj_dentry;
3022
3023         if (!dt_object_exists(dt))
3024                 return -ENOENT;
3025
3026         LASSERT(!dt_object_remote(dt));
3027         LASSERT(inode->i_op != NULL);
3028         LASSERT(inode->i_op->listxattr != NULL);
3029
3030         dentry->d_inode = inode;
3031         dentry->d_sb = inode->i_sb;
3032         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3033 }
3034
3035 static int osd_declare_xattr_del(const struct lu_env *env,
3036                                  struct dt_object *dt, const char *name,
3037                                  struct thandle *handle)
3038 {
3039         struct osd_thandle *oh;
3040         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3041
3042         LASSERT(!dt_object_remote(dt));
3043         LASSERT(handle != NULL);
3044
3045         oh = container_of0(handle, struct osd_thandle, ot_super);
3046         LASSERT(oh->ot_handle == NULL);
3047
3048         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3049                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3050         /*
3051          * xattr del may involve inode quota change, reserve credits for
3052          * dquot_initialize()
3053          */
3054         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3055
3056         return 0;
3057 }
3058
3059 /*
3060  * Concurrency: @dt is write locked.
3061  */
3062 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3063                          const char *name, struct thandle *handle)
3064 {
3065         struct osd_object      *obj    = osd_dt_obj(dt);
3066         struct inode           *inode  = obj->oo_inode;
3067         struct osd_thread_info *info   = osd_oti_get(env);
3068         struct dentry          *dentry = &info->oti_obj_dentry;
3069         int                     rc;
3070
3071         if (!dt_object_exists(dt))
3072                 return -ENOENT;
3073
3074         LASSERT(!dt_object_remote(dt));
3075         LASSERT(inode->i_op != NULL);
3076         LASSERT(inode->i_op->removexattr != NULL);
3077         LASSERT(handle != NULL);
3078
3079         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3080
3081         ll_vfs_dq_init(inode);
3082         dentry->d_inode = inode;
3083         dentry->d_sb = inode->i_sb;
3084         rc = inode->i_op->removexattr(dentry, name);
3085         return rc;
3086 }
3087
3088 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3089                            __u64 start, __u64 end)
3090 {
3091         struct osd_object       *obj    = osd_dt_obj(dt);
3092         struct inode            *inode  = obj->oo_inode;
3093         struct osd_thread_info  *info   = osd_oti_get(env);
3094         struct dentry           *dentry = &info->oti_obj_dentry;
3095         struct file             *file   = &info->oti_file;
3096         int                     rc;
3097
3098         ENTRY;
3099
3100         dentry->d_inode = inode;
3101         dentry->d_sb = inode->i_sb;
3102         file->f_dentry = dentry;
3103         file->f_mapping = inode->i_mapping;
3104         file->f_op = inode->i_fop;
3105         set_file_inode(file, inode);
3106
3107         rc = ll_vfs_fsync_range(file, start, end, 0);
3108
3109         RETURN(rc);
3110 }
3111
3112 /*
3113  * Index operations.
3114  */
3115
3116 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3117                            const struct dt_index_features *feat)
3118 {
3119         struct iam_descr *descr;
3120
3121         if (osd_object_is_root(o))
3122                 return feat == &dt_directory_features;
3123
3124         LASSERT(o->oo_dir != NULL);
3125
3126         descr = o->oo_dir->od_container.ic_descr;
3127         if (feat == &dt_directory_features) {
3128                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3129                         return 1;
3130                 else
3131                         return 0;
3132         } else {
3133                 return
3134                         feat->dif_keysize_min <= descr->id_key_size &&
3135                         descr->id_key_size <= feat->dif_keysize_max &&
3136                         feat->dif_recsize_min <= descr->id_rec_size &&
3137                         descr->id_rec_size <= feat->dif_recsize_max &&
3138                         !(feat->dif_flags & (DT_IND_VARKEY |
3139                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3140                         ergo(feat->dif_flags & DT_IND_UPDATE,
3141                              1 /* XXX check that object (and file system) is
3142                                 * writable */);
3143         }
3144 }
3145
3146 static int osd_iam_container_init(const struct lu_env *env,
3147                                   struct osd_object *obj,
3148                                   struct osd_directory *dir)
3149 {
3150         struct iam_container *bag = &dir->od_container;
3151         int result;
3152
3153         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3154         if (result != 0)
3155                 return result;
3156
3157         result = iam_container_setup(bag);
3158         if (result == 0)
3159                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3160         else
3161                 iam_container_fini(bag);
3162
3163         return result;
3164 }
3165
3166
3167 /*
3168  * Concurrency: no external locking is necessary.
3169  */
3170 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3171                          const struct dt_index_features *feat)
3172 {
3173         int                      result;
3174         int                      skip_iam = 0;
3175         struct osd_object       *obj = osd_dt_obj(dt);
3176
3177         LINVRNT(osd_invariant(obj));
3178
3179         if (osd_object_is_root(obj)) {
3180                 dt->do_index_ops = &osd_index_ea_ops;
3181                 result = 0;
3182         } else if (feat == &dt_directory_features) {
3183                 dt->do_index_ops = &osd_index_ea_ops;
3184                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3185                         result = 0;
3186                 else
3187                         result = -ENOTDIR;
3188                 skip_iam = 1;
3189         } else if (unlikely(feat == &dt_otable_features)) {
3190                 dt->do_index_ops = &osd_otable_ops;
3191                 return 0;
3192         } else if (unlikely(feat == &dt_acct_features)) {
3193                 dt->do_index_ops = &osd_acct_index_ops;
3194                 result = 0;
3195                 skip_iam = 1;
3196         } else if (!osd_has_index(obj)) {
3197                 struct osd_directory *dir;
3198
3199                 OBD_ALLOC_PTR(dir);
3200                 if (dir != NULL) {
3201
3202                         spin_lock(&obj->oo_guard);
3203                         if (obj->oo_dir == NULL)
3204                                 obj->oo_dir = dir;
3205                         else
3206                                 /*
3207                                  * Concurrent thread allocated container data.
3208                                  */
3209                                 OBD_FREE_PTR(dir);
3210                         spin_unlock(&obj->oo_guard);
3211                         /*
3212                          * Now, that we have container data, serialize its
3213                          * initialization.
3214                          */
3215                         down_write(&obj->oo_ext_idx_sem);
3216                         /*
3217                          * recheck under lock.
3218                          */
3219                         if (!osd_has_index(obj))
3220                                 result = osd_iam_container_init(env, obj,
3221                                                                 obj->oo_dir);
3222                         else
3223                                 result = 0;
3224                         up_write(&obj->oo_ext_idx_sem);
3225                 } else {
3226                         result = -ENOMEM;
3227                 }
3228         } else {
3229                 result = 0;
3230         }
3231
3232         if (result == 0 && skip_iam == 0) {
3233                 if (!osd_iam_index_probe(env, obj, feat))
3234                         result = -ENOTDIR;
3235         }
3236         LINVRNT(osd_invariant(obj));
3237
3238         if (result == 0 && is_quota_glb_feat(feat) &&
3239             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3240                 result = osd_quota_migration(env, dt, feat);
3241
3242         return result;
3243 }
3244
3245 static int osd_otable_it_attr_get(const struct lu_env *env,
3246                                  struct dt_object *dt,
3247                                  struct lu_attr *attr)
3248 {
3249         attr->la_valid = 0;
3250         return 0;
3251 }
3252
3253 static const struct dt_object_operations osd_obj_ops = {
3254         .do_read_lock         = osd_object_read_lock,
3255         .do_write_lock        = osd_object_write_lock,
3256         .do_read_unlock       = osd_object_read_unlock,
3257         .do_write_unlock      = osd_object_write_unlock,
3258         .do_write_locked      = osd_object_write_locked,
3259         .do_attr_get          = osd_attr_get,
3260         .do_declare_attr_set  = osd_declare_attr_set,
3261         .do_attr_set          = osd_attr_set,
3262         .do_ah_init           = osd_ah_init,
3263         .do_declare_create    = osd_declare_object_create,
3264         .do_create            = osd_object_create,
3265         .do_declare_destroy   = osd_declare_object_destroy,
3266         .do_destroy           = osd_object_destroy,
3267         .do_index_try         = osd_index_try,
3268         .do_declare_ref_add   = osd_declare_object_ref_add,
3269         .do_ref_add           = osd_object_ref_add,
3270         .do_declare_ref_del   = osd_declare_object_ref_del,
3271         .do_ref_del           = osd_object_ref_del,
3272         .do_xattr_get         = osd_xattr_get,
3273         .do_declare_xattr_set = osd_declare_xattr_set,
3274         .do_xattr_set         = osd_xattr_set,
3275         .do_declare_xattr_del = osd_declare_xattr_del,
3276         .do_xattr_del         = osd_xattr_del,
3277         .do_xattr_list        = osd_xattr_list,
3278         .do_object_sync       = osd_object_sync,
3279 };
3280
3281 /**
3282  * dt_object_operations for interoperability mode
3283  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3284  */
3285 static const struct dt_object_operations osd_obj_ea_ops = {
3286         .do_read_lock         = osd_object_read_lock,
3287         .do_write_lock        = osd_object_write_lock,
3288         .do_read_unlock       = osd_object_read_unlock,
3289         .do_write_unlock      = osd_object_write_unlock,
3290         .do_write_locked      = osd_object_write_locked,
3291         .do_attr_get          = osd_attr_get,
3292         .do_declare_attr_set  = osd_declare_attr_set,
3293         .do_attr_set          = osd_attr_set,
3294         .do_ah_init           = osd_ah_init,
3295         .do_declare_create    = osd_declare_object_create,
3296         .do_create            = osd_object_ea_create,
3297         .do_declare_destroy   = osd_declare_object_destroy,
3298         .do_destroy           = osd_object_destroy,
3299         .do_index_try         = osd_index_try,
3300         .do_declare_ref_add   = osd_declare_object_ref_add,
3301         .do_ref_add           = osd_object_ref_add,
3302         .do_declare_ref_del   = osd_declare_object_ref_del,
3303         .do_ref_del           = osd_object_ref_del,
3304         .do_xattr_get         = osd_xattr_get,
3305         .do_declare_xattr_set = osd_declare_xattr_set,
3306         .do_xattr_set         = osd_xattr_set,
3307         .do_declare_xattr_del = osd_declare_xattr_del,
3308         .do_xattr_del         = osd_xattr_del,
3309         .do_xattr_list        = osd_xattr_list,
3310         .do_object_sync       = osd_object_sync,
3311 };
3312
3313 static const struct dt_object_operations osd_obj_otable_it_ops = {
3314         .do_attr_get    = osd_otable_it_attr_get,
3315         .do_index_try   = osd_index_try,
3316 };
3317
3318 static int osd_index_declare_iam_delete(const struct lu_env *env,
3319                                         struct dt_object *dt,
3320                                         const struct dt_key *key,
3321                                         struct thandle *handle)
3322 {
3323         struct osd_thandle    *oh;
3324
3325         oh = container_of0(handle, struct osd_thandle, ot_super);
3326         LASSERT(oh->ot_handle == NULL);
3327
3328         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3329                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3330
3331         return 0;
3332 }
3333
3334 /**
3335  *      delete a (key, value) pair from index \a dt specified by \a key
3336  *
3337  *      \param  dt      osd index object
3338  *      \param  key     key for index
3339  *      \param  rec     record reference
3340  *      \param  handle  transaction handler
3341  *
3342  *      \retval  0  success
3343  *      \retval -ve   failure
3344  */
3345 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3346                                 const struct dt_key *key,
3347                                 struct thandle *handle)
3348 {
3349         struct osd_thread_info *oti = osd_oti_get(env);
3350         struct osd_object      *obj = osd_dt_obj(dt);
3351         struct osd_thandle     *oh;
3352         struct iam_path_descr  *ipd;
3353         struct iam_container   *bag = &obj->oo_dir->od_container;
3354         int                     rc;
3355         ENTRY;
3356
3357         if (!dt_object_exists(dt))
3358                 RETURN(-ENOENT);
3359
3360         LINVRNT(osd_invariant(obj));
3361         LASSERT(!dt_object_remote(dt));
3362         LASSERT(bag->ic_object == obj->oo_inode);
3363         LASSERT(handle != NULL);
3364
3365         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3366
3367         ipd = osd_idx_ipd_get(env, bag);
3368         if (unlikely(ipd == NULL))
3369                 RETURN(-ENOMEM);
3370
3371         oh = container_of0(handle, struct osd_thandle, ot_super);
3372         LASSERT(oh->ot_handle != NULL);
3373         LASSERT(oh->ot_handle->h_transaction != NULL);
3374
3375         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3376                 /* swab quota uid/gid provided by caller */
3377                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3378                 key = (const struct dt_key *)&oti->oti_quota_id;
3379         }
3380
3381         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3382         osd_ipd_put(env, bag, ipd);
3383         LINVRNT(osd_invariant(obj));
3384         RETURN(rc);
3385 }
3386
3387 static int osd_index_declare_ea_delete(const struct lu_env *env,
3388                                        struct dt_object *dt,
3389                                        const struct dt_key *key,
3390                                        struct thandle *handle)
3391 {
3392         struct osd_thandle *oh;
3393         struct inode       *inode;
3394         int                 rc;
3395         ENTRY;
3396
3397         LASSERT(!dt_object_remote(dt));
3398         LASSERT(handle != NULL);
3399
3400         oh = container_of0(handle, struct osd_thandle, ot_super);
3401         LASSERT(oh->ot_handle == NULL);
3402
3403         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3404                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3405
3406         inode = osd_dt_obj(dt)->oo_inode;
3407         LASSERT(inode);
3408
3409         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3410                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3411         RETURN(rc);
3412 }
3413
3414 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3415                                           struct dt_rec *fid)
3416 {
3417         struct osd_fid_pack *rec;
3418         int                  rc = -ENODATA;
3419
3420         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3421                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3422                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3423         }
3424         return rc;
3425 }
3426
3427 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3428                           const struct lu_fid *fid)
3429 {
3430         struct seq_server_site  *ss = osd_seq_site(osd);
3431         ENTRY;
3432
3433         /* FID seqs not in FLDB, must be local seq */
3434         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3435                 RETURN(0);
3436
3437         /* If FLD is not being initialized yet, it only happens during the
3438          * initialization, likely during mgs initialization, and we assume
3439          * this is local FID. */
3440         if (ss == NULL || ss->ss_server_fld == NULL)
3441                 RETURN(0);
3442
3443         /* Only check the local FLDB here */
3444         if (osd_seq_exists(env, osd, fid_seq(fid)))
3445                 RETURN(0);
3446
3447         RETURN(1);
3448 }
3449
3450 /**
3451  * Index delete function for interoperability mode (b11826).
3452  * It will remove the directory entry added by osd_index_ea_insert().
3453  * This entry is needed to maintain name->fid mapping.
3454  *
3455  * \param key,  key i.e. file entry to be deleted
3456  *
3457  * \retval   0, on success
3458  * \retval -ve, on error
3459  */
3460 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3461                                const struct dt_key *key, struct thandle *handle)
3462 {
3463         struct osd_object          *obj = osd_dt_obj(dt);
3464         struct inode               *dir = obj->oo_inode;
3465         struct dentry              *dentry;
3466         struct osd_thandle         *oh;
3467         struct ldiskfs_dir_entry_2 *de = NULL;
3468         struct buffer_head         *bh;
3469         struct htree_lock          *hlock = NULL;
3470         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3471         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3472         int                        rc;
3473         ENTRY;
3474
3475         if (!dt_object_exists(dt))
3476                 RETURN(-ENOENT);
3477
3478         LINVRNT(osd_invariant(obj));
3479         LASSERT(!dt_object_remote(dt));
3480         LASSERT(handle != NULL);
3481
3482         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3483
3484         oh = container_of(handle, struct osd_thandle, ot_super);
3485         LASSERT(oh->ot_handle != NULL);
3486         LASSERT(oh->ot_handle->h_transaction != NULL);
3487
3488         ll_vfs_dq_init(dir);
3489         dentry = osd_child_dentry_get(env, obj,
3490                                       (char *)key, strlen((char *)key));
3491
3492         if (obj->oo_hl_head != NULL) {
3493                 hlock = osd_oti_get(env)->oti_hlock;
3494                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3495                                    dir, LDISKFS_HLOCK_DEL);
3496         } else {
3497                 down_write(&obj->oo_ext_idx_sem);
3498         }
3499
3500         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3501         if (bh) {
3502                 __u32 ino = 0;
3503
3504                 /* If this is not the ".." entry, it might be a remote DNE
3505                  * entry and  we need to check if the FID is for a remote
3506                  * MDT.  If the FID is  not in the directory entry (e.g.
3507                  * upgraded 1.8 filesystem without dirdata enabled) then
3508                  * we need to get the FID from the LMA. For a remote directory
3509                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3510                  * case.
3511                  *
3512                  * Delete the entry before the agent inode in order to
3513                  * simplify error handling.  At worst an error after deleting
3514                  * the entry first might leak the agent inode afterward. The
3515                  * reverse would need filesystem abort in case of error deleting
3516                  * the entry after the agent had been removed, or leave a
3517                  * dangling entry pointing at a random inode. */
3518                 if (strcmp((char *)key, dotdot) != 0) {
3519                         LASSERT(de != NULL);
3520                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3521                         /* If Fid is not in dentry, try to get it from LMA */
3522                         if (rc == -ENODATA) {
3523                                 struct osd_inode_id *id;
3524                                 struct inode *inode;
3525
3526                                 /* Before trying to get fid from the inode,
3527                                  * check whether the inode is valid.
3528                                  *
3529                                  * If the inode has been deleted, do not go
3530                                  * ahead to do osd_ea_fid_get, which will set
3531                                  * the inode to bad inode, which might cause
3532                                  * the inode to be deleted uncorrectly */
3533                                 inode = ldiskfs_iget(osd_sb(osd),
3534                                                      le32_to_cpu(de->inode));
3535                                 if (IS_ERR(inode)) {
3536                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3537                                                "error.\n", osd_name(osd),
3538                                                PFID(fid));
3539                                         rc = PTR_ERR(inode);
3540                                 } else {
3541                                         if (likely(inode->i_nlink != 0)) {
3542                                                 id = &osd_oti_get(env)->oti_id;
3543                                                 rc = osd_ea_fid_get(env, obj,
3544                                                         le32_to_cpu(de->inode),
3545                                                                     fid, id);
3546                                         } else {
3547                                                 CDEBUG(D_INFO, "%s: %u "DFID
3548                                                        "deleted.\n",
3549                                                        osd_name(osd),
3550                                                        le32_to_cpu(de->inode),
3551                                                        PFID(fid));
3552                                                 rc = -ESTALE;
3553                                         }
3554                                         iput(inode);
3555                                 }
3556                         }
3557                         if (rc == 0 &&
3558                             unlikely(osd_remote_fid(env, osd, fid)))
3559                                 /* Need to delete agent inode */
3560                                 ino = le32_to_cpu(de->inode);
3561                 }
3562                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3563                 brelse(bh);
3564                 if (rc == 0 && unlikely(ino != 0)) {
3565                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3566                                                           oh);
3567                         if (rc != 0)
3568                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3569                                        osd_name(osd), PFID(fid), rc);
3570                 }
3571         } else {
3572                 rc = -ENOENT;
3573         }
3574         if (hlock != NULL)
3575                 ldiskfs_htree_unlock(hlock);
3576         else
3577                 up_write(&obj->oo_ext_idx_sem);
3578
3579         if (rc != 0)
3580                 GOTO(out, rc);
3581
3582         /* For inode on the remote MDT, .. will point to
3583          * /Agent directory, Check whether it needs to delete
3584          * from agent directory */
3585         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3586                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3587                                                    oh);
3588                 if (rc != 0 && rc != -ENOENT) {
3589                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3590                                osd_name(osd), PFID(fid), rc);
3591                 }
3592
3593                 if (rc == -ENOENT)
3594                         rc = 0;
3595
3596                 GOTO(out, rc);
3597         }
3598 out:
3599
3600         LASSERT(osd_invariant(obj));
3601         RETURN(rc);
3602 }
3603
3604 /**
3605  *      Lookup index for \a key and copy record to \a rec.
3606  *
3607  *      \param  dt      osd index object
3608  *      \param  key     key for index
3609  *      \param  rec     record reference
3610  *
3611  *      \retval  +ve  success : exact mach
3612  *      \retval  0    return record with key not greater than \a key
3613  *      \retval -ve   failure
3614  */
3615 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3616                                 struct dt_rec *rec, const struct dt_key *key)
3617 {
3618         struct osd_object      *obj = osd_dt_obj(dt);
3619         struct iam_path_descr  *ipd;
3620         struct iam_container   *bag = &obj->oo_dir->od_container;
3621         struct osd_thread_info *oti = osd_oti_get(env);
3622         struct iam_iterator    *it = &oti->oti_idx_it;
3623         struct iam_rec         *iam_rec;
3624         int                     rc;
3625         ENTRY;
3626
3627         if (!dt_object_exists(dt))
3628                 RETURN(-ENOENT);
3629
3630         LASSERT(osd_invariant(obj));
3631         LASSERT(!dt_object_remote(dt));
3632         LASSERT(bag->ic_object == obj->oo_inode);
3633
3634         ipd = osd_idx_ipd_get(env, bag);
3635         if (IS_ERR(ipd))
3636                 RETURN(-ENOMEM);
3637
3638         /* got ipd now we can start iterator. */
3639         iam_it_init(it, bag, 0, ipd);
3640
3641         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3642                 /* swab quota uid/gid provided by caller */
3643                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3644                 key = (const struct dt_key *)&oti->oti_quota_id;
3645         }
3646
3647         rc = iam_it_get(it, (struct iam_key *)key);
3648         if (rc >= 0) {
3649                 if (S_ISDIR(obj->oo_inode->i_mode))
3650                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3651                 else
3652                         iam_rec = (struct iam_rec *) rec;
3653
3654                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3655
3656                 if (S_ISDIR(obj->oo_inode->i_mode))
3657                         osd_fid_unpack((struct lu_fid *) rec,
3658                                        (struct osd_fid_pack *)iam_rec);
3659                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3660                         osd_quota_unpack(obj, rec);
3661         }
3662
3663         iam_it_put(it);
3664         iam_it_fini(it);
3665         osd_ipd_put(env, bag, ipd);
3666
3667         LINVRNT(osd_invariant(obj));
3668
3669         RETURN(rc);
3670 }
3671
3672 static int osd_index_declare_iam_insert(const struct lu_env *env,
3673                                         struct dt_object *dt,
3674                                         const struct dt_rec *rec,
3675                                         const struct dt_key *key,
3676                                         struct thandle *handle)
3677 {
3678         struct osd_thandle *oh;
3679
3680         LASSERT(handle != NULL);
3681
3682         oh = container_of0(handle, struct osd_thandle, ot_super);
3683         LASSERT(oh->ot_handle == NULL);
3684
3685         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3686                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3687
3688         return 0;
3689 }
3690
3691 /**
3692  *      Inserts (key, value) pair in \a dt index object.
3693  *
3694  *      \param  dt      osd index object
3695  *      \param  key     key for index
3696  *      \param  rec     record reference
3697  *      \param  th      transaction handler
3698  *
3699  *      \retval  0  success
3700  *      \retval -ve failure
3701  */
3702 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3703                                 const struct dt_rec *rec,
3704                                 const struct dt_key *key, struct thandle *th,
3705                                 int ignore_quota)
3706 {
3707         struct osd_object     *obj = osd_dt_obj(dt);
3708         struct iam_path_descr *ipd;
3709         struct osd_thandle    *oh;
3710         struct iam_container  *bag;
3711         struct osd_thread_info *oti = osd_oti_get(env);
3712         struct iam_rec         *iam_rec;
3713         int                     rc;
3714         ENTRY;
3715
3716         if (!dt_object_exists(dt))
3717                 RETURN(-ENOENT);
3718
3719         LINVRNT(osd_invariant(obj));
3720         LASSERT(!dt_object_remote(dt));
3721
3722         bag = &obj->oo_dir->od_container;
3723         LASSERT(bag->ic_object == obj->oo_inode);
3724         LASSERT(th != NULL);
3725
3726         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3727
3728         ipd = osd_idx_ipd_get(env, bag);
3729         if (unlikely(ipd == NULL))
3730                 RETURN(-ENOMEM);
3731
3732         oh = container_of0(th, struct osd_thandle, ot_super);
3733         LASSERT(oh->ot_handle != NULL);
3734         LASSERT(oh->ot_handle->h_transaction != NULL);
3735         if (S_ISDIR(obj->oo_inode->i_mode)) {
3736                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3737                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3738         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3739                 /* pack quota uid/gid */
3740                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3741                 key = (const struct dt_key *)&oti->oti_quota_id;
3742                 /* pack quota record */
3743                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3744                 iam_rec = (struct iam_rec *)rec;
3745         } else {
3746                 iam_rec = (struct iam_rec *)rec;
3747         }
3748
3749         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3750                         iam_rec, ipd);
3751         osd_ipd_put(env, bag, ipd);
3752         LINVRNT(osd_invariant(obj));
3753         RETURN(rc);
3754 }
3755
3756 /**
3757  * Calls ldiskfs_add_entry() to add directory entry
3758  * into the directory. This is required for
3759  * interoperability mode (b11826)
3760  *
3761  * \retval   0, on success
3762  * \retval -ve, on error
3763  */
3764 static int __osd_ea_add_rec(struct osd_thread_info *info,
3765                             struct osd_object *pobj, struct inode  *cinode,
3766                             const char *name, const struct lu_fid *fid,
3767                             struct htree_lock *hlock, struct thandle *th)
3768 {
3769         struct ldiskfs_dentry_param *ldp;
3770         struct dentry               *child;
3771         struct osd_thandle          *oth;
3772         int                          rc;
3773
3774         oth = container_of(th, struct osd_thandle, ot_super);
3775         LASSERT(oth->ot_handle != NULL);
3776         LASSERT(oth->ot_handle->h_transaction != NULL);
3777         LASSERT(pobj->oo_inode);
3778
3779         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3780         if (unlikely(pobj->oo_inode ==
3781                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3782                 ldp->edp_magic = 0;
3783         else
3784                 osd_get_ldiskfs_dirent_param(ldp, fid);
3785         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3786         child->d_fsdata = (void *)ldp;
3787         ll_vfs_dq_init(pobj->oo_inode);
3788         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3789         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3790                 struct ldiskfs_dir_entry_2      *de;
3791                 struct buffer_head              *bh;
3792                 int                              rc1;
3793
3794                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3795                                             NULL, hlock);
3796                 if (bh != NULL) {
3797                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3798                                                                bh);
3799                         if (rc1 == 0) {
3800                                 if (S_ISDIR(cinode->i_mode))
3801                                         de->file_type = LDISKFS_DIRENT_LUFID |
3802                                                         LDISKFS_FT_REG_FILE;
3803                                 else
3804                                         de->file_type = LDISKFS_DIRENT_LUFID |
3805                                                         LDISKFS_FT_DIR;
3806                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3807                                                               NULL, bh);
3808                                 brelse(bh);
3809                         }
3810                 }
3811         }
3812
3813         RETURN(rc);
3814 }
3815
3816 /**
3817  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3818  * into the directory.Also sets flags into osd object to
3819  * indicate dot and dotdot are created. This is required for
3820  * interoperability mode (b11826)
3821  *
3822  * \param dir   directory for dot and dotdot fixup.
3823  * \param obj   child object for linking
3824  *
3825  * \retval   0, on success
3826  * \retval -ve, on error
3827  */
3828 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3829                               struct osd_object *dir,
3830                               struct inode *parent_dir, const char *name,
3831                               const struct lu_fid *dot_fid,
3832                               const struct lu_fid *dot_dot_fid,
3833                               struct thandle *th)
3834 {
3835         struct inode                *inode = dir->oo_inode;
3836         struct osd_thandle          *oth;
3837         int result = 0;
3838
3839         oth = container_of(th, struct osd_thandle, ot_super);
3840         LASSERT(oth->ot_handle->h_transaction != NULL);
3841         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3842
3843         if (strcmp(name, dot) == 0) {
3844                 if (dir->oo_compat_dot_created) {
3845                         result = -EEXIST;
3846                 } else {
3847                         LASSERT(inode == parent_dir);
3848                         dir->oo_compat_dot_created = 1;
3849                         result = 0;
3850                 }
3851         } else if (strcmp(name, dotdot) == 0) {
3852                 if (!dir->oo_compat_dot_created)
3853                         return -EINVAL;
3854                 /* in case of rename, dotdot is already created */
3855                 if (dir->oo_compat_dotdot_created) {
3856                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3857                                                 dot_dot_fid, NULL, th);
3858                 }
3859
3860                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
3861                         struct lu_fid tfid = *dot_dot_fid;
3862
3863                         tfid.f_oid--;
3864                         result = osd_add_dot_dotdot_internal(info,
3865                                         dir->oo_inode, parent_dir, dot_fid,
3866                                         &tfid, oth);
3867                 } else {
3868                         result = osd_add_dot_dotdot_internal(info,
3869                                         dir->oo_inode, parent_dir, dot_fid,
3870                                         dot_dot_fid, oth);
3871                 }
3872
3873                 if (result == 0)
3874                         dir->oo_compat_dotdot_created = 1;
3875         }
3876
3877         return result;
3878 }
3879
3880
3881 /**
3882  * It will call the appropriate osd_add* function and return the
3883  * value, return by respective functions.
3884  */
3885 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3886                           struct inode *cinode, const char *name,
3887                           const struct lu_fid *fid, struct thandle *th)
3888 {
3889         struct osd_thread_info *info   = osd_oti_get(env);
3890         struct htree_lock      *hlock;
3891         int                     rc;
3892
3893         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3894
3895         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3896                                                    name[2] =='\0'))) {
3897                 if (hlock != NULL) {
3898                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3899                                            pobj->oo_inode, 0);
3900                 } else {
3901                         down_write(&pobj->oo_ext_idx_sem);
3902                 }
3903
3904                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3905                                         lu_object_fid(&pobj->oo_dt.do_lu),
3906                                         fid, th);
3907         } else {
3908                 if (hlock != NULL) {
3909                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3910                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3911                 } else {
3912                         down_write(&pobj->oo_ext_idx_sem);
3913                 }
3914
3915                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3916                         struct lu_fid *tfid = &info->oti_fid;
3917
3918                         *tfid = *fid;
3919                         tfid->f_ver = ~0;
3920                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3921                                               tfid, hlock, th);
3922                 } else {
3923                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3924                                               hlock, th);
3925                 }
3926         }
3927         if (hlock != NULL)
3928                 ldiskfs_htree_unlock(hlock);
3929         else
3930                 up_write(&pobj->oo_ext_idx_sem);
3931
3932         return rc;
3933 }
3934
3935 static void
3936 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3937                       struct osd_idmap_cache *oic)
3938 {
3939         struct osd_scrub    *scrub = &dev->od_scrub;
3940         struct lu_fid       *fid   = &oic->oic_fid;
3941         struct osd_inode_id *id    = &oti->oti_id;
3942         int                  once  = 0;
3943         int                  rc;
3944         ENTRY;
3945
3946         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3947                 RETURN_EXIT;
3948
3949         if (scrub->os_pos_current > id->oii_ino)
3950                 RETURN_EXIT;
3951
3952 again:
3953         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
3954         if (rc != 0 && rc != -ENOENT)
3955                 RETURN_EXIT;
3956
3957         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3958                 RETURN_EXIT;
3959
3960         if (thread_is_running(&scrub->os_thread)) {
3961                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3962                 /* There is race condition between osd_oi_lookup and OI scrub.
3963                  * The OI scrub finished just after osd_oi_lookup() failure.
3964                  * Under such case, it is unnecessary to trigger OI scrub again,
3965                  * but try to call osd_oi_lookup() again. */
3966                 if (unlikely(rc == -EAGAIN))
3967                         goto again;
3968
3969                 RETURN_EXIT;
3970         }
3971
3972         if (!dev->od_noscrub && ++once == 1) {
3973                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
3974                                      SS_CLEAR_FAILOUT);
3975                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
3976                        "for "DFID", rc = %d [2]\n",
3977                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3978                        PFID(fid), rc);
3979                 if (rc == 0 || rc == -EALREADY)
3980                         goto again;
3981         }
3982
3983         EXIT;
3984 }
3985
3986 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3987                                struct osd_device *dev,
3988                                struct osd_idmap_cache *oic,
3989                                struct lu_fid *fid, __u32 ino)
3990 {
3991         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3992         struct inode            *inode;
3993         int                      rc;
3994
3995         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3996         inode = osd_iget(oti, dev, &oic->oic_lid);
3997         if (IS_ERR(inode)) {
3998                 fid_zero(&oic->oic_fid);
3999                 return PTR_ERR(inode);
4000         }
4001
4002         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4003         iput(inode);
4004         if (rc != 0)
4005                 fid_zero(&oic->oic_fid);
4006         else
4007                 *fid = oic->oic_fid = lma->lma_self_fid;
4008         return rc;
4009 }
4010
4011 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4012                       struct osd_inode_id *id, const struct lu_fid *fid)
4013 {
4014         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4015                id->oii_ino, id->oii_gen, info);
4016         info->oti_cache.oic_lid = *id;
4017         info->oti_cache.oic_fid = *fid;
4018         info->oti_cache.oic_dev = osd;
4019 }
4020
4021 /**
4022  * Get parent FID from the linkEA.
4023  *
4024  * For a directory which parent resides on remote MDT, to satisfy the
4025  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4026  * the other hand, to make the lookup(..) on the directory can return
4027  * the real parent FID, we append the real parent FID after its ".."
4028  * name entry in the /REMOTE_PARENT_DIR.
4029  *
4030  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4031  * backup. So after the restore, we cannot get the right parent FID from
4032  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4033  * we have stored the real parent FID in the directory object's linkEA,
4034  * we can parse the linkEA for the real parent FID.
4035  *
4036  * \param[in] env       pointer to the thread context
4037  * \param[in] obj       pointer to the object to be handled
4038  * \param[out]fid       pointer to the buffer to hold the parent FID
4039  *
4040  * \retval              0 for getting the real parent FID successfully
4041  * \retval              negative error number on failure
4042  */
4043 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4044                                     struct osd_object *obj,
4045                                     struct lu_fid *fid)
4046 {
4047         struct osd_thread_info  *oti    = osd_oti_get(env);
4048         struct lu_buf           *buf    = &oti->oti_big_buf;
4049         struct dentry           *dentry = &oti->oti_obj_dentry;
4050         struct inode            *inode  = obj->oo_inode;
4051         struct linkea_data       ldata  = { NULL };
4052         int                      rc;
4053         ENTRY;
4054
4055         fid_zero(fid);
4056         if (!S_ISDIR(inode->i_mode))
4057                 RETURN(-EIO);
4058
4059 again:
4060         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4061                              buf->lb_buf, buf->lb_len);
4062         if (rc == -ERANGE) {
4063                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4064                                      NULL, 0);
4065                 if (rc > 0) {
4066                         lu_buf_realloc(buf, rc);
4067                         if (buf->lb_buf == NULL)
4068                                 RETURN(-ENOMEM);
4069
4070                         goto again;
4071                 }
4072         }
4073
4074         if (unlikely(rc == 0))
4075                 RETURN(-ENODATA);
4076
4077         if (rc < 0)
4078                 RETURN(rc);
4079
4080         if (unlikely(buf->lb_buf == NULL)) {
4081                 lu_buf_realloc(buf, rc);
4082                 if (buf->lb_buf == NULL)
4083                         RETURN(-ENOMEM);
4084
4085                 goto again;
4086         }
4087
4088         ldata.ld_buf = buf;
4089         rc = linkea_init(&ldata);
4090         if (rc == 0) {
4091                 linkea_first_entry(&ldata);
4092                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4093         }
4094
4095         RETURN(rc);
4096 }
4097
4098 /**
4099  * Calls ->lookup() to find dentry. From dentry get inode and
4100  * read inode's ea to get fid. This is required for  interoperability
4101  * mode (b11826)
4102  *
4103  * \retval   0, on success
4104  * \retval -ve, on error
4105  */
4106 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4107                              struct dt_rec *rec, const struct dt_key *key)
4108 {
4109         struct inode                    *dir    = obj->oo_inode;
4110         struct dentry                   *dentry;
4111         struct ldiskfs_dir_entry_2      *de;
4112         struct buffer_head              *bh;
4113         struct lu_fid                   *fid = (struct lu_fid *) rec;
4114         struct htree_lock               *hlock = NULL;
4115         int                             ino;
4116         int                             rc;
4117         ENTRY;
4118
4119         LASSERT(dir->i_op != NULL);
4120         LASSERT(dir->i_op->lookup != NULL);
4121
4122         dentry = osd_child_dentry_get(env, obj,
4123                                       (char *)key, strlen((char *)key));
4124
4125         if (obj->oo_hl_head != NULL) {
4126                 hlock = osd_oti_get(env)->oti_hlock;
4127                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4128                                    dir, LDISKFS_HLOCK_LOOKUP);
4129         } else {
4130                 down_read(&obj->oo_ext_idx_sem);
4131         }
4132
4133         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4134         if (bh) {
4135                 struct osd_thread_info *oti = osd_oti_get(env);
4136                 struct osd_inode_id *id = &oti->oti_id;
4137                 struct osd_idmap_cache *oic = &oti->oti_cache;
4138                 struct osd_device *dev = osd_obj2dev(obj);
4139
4140                 ino = le32_to_cpu(de->inode);
4141                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4142                         brelse(bh);
4143                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4144                         GOTO(out, rc);
4145                 }
4146
4147                 rc = osd_get_fid_from_dentry(de, rec);
4148
4149                 /* done with de, release bh */
4150                 brelse(bh);
4151                 if (rc != 0) {
4152                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
4153                                 const char *name = (const char *)key;
4154
4155                                 /* If the parent is on remote MDT, and there
4156                                  * is no FID-in-dirent, then we have to get
4157                                  * the parent FID from the linkEA.  */
4158                                 if (likely(strlen(name) == 2 &&
4159                                            name[0] == '.' && name[1] == '.'))
4160                                         rc = osd_get_pfid_from_linkea(env, obj,
4161                                                                       fid);
4162                         } else {
4163                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4164                         }
4165                 } else {
4166                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4167                 }
4168
4169                 if (rc != 0) {
4170                         fid_zero(&oic->oic_fid);
4171                         GOTO(out, rc);
4172                 }
4173
4174                 if (osd_remote_fid(env, dev, fid))
4175                         GOTO(out, rc = 0);
4176
4177                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4178                 osd_consistency_check(oti, dev, oic);
4179         } else {
4180                 rc = -ENOENT;
4181         }
4182
4183         GOTO(out, rc);
4184
4185 out:
4186         if (hlock != NULL)
4187                 ldiskfs_htree_unlock(hlock);
4188         else
4189                 up_read(&obj->oo_ext_idx_sem);
4190         return rc;
4191 }
4192
4193 /**
4194  * Find the osd object for given fid.
4195  *
4196  * \param fid need to find the osd object having this fid
4197  *
4198  * \retval osd_object on success
4199  * \retval        -ve on error
4200  */
4201 static struct osd_object *osd_object_find(const struct lu_env *env,
4202                                           struct dt_object *dt,
4203                                           const struct lu_fid *fid)
4204 {
4205         struct lu_device  *ludev = dt->do_lu.lo_dev;
4206         struct osd_object *child = NULL;
4207         struct lu_object  *luch;
4208         struct lu_object  *lo;
4209
4210         /*
4211          * at this point topdev might not exist yet
4212          * (i.e. MGS is preparing profiles). so we can
4213          * not rely on topdev and instead lookup with
4214          * our device passed as topdev. this can't work
4215          * if the object isn't cached yet (as osd doesn't
4216          * allocate lu_header). IOW, the object must be
4217          * in the cache, otherwise lu_object_alloc() crashes
4218          * -bzzz
4219          */
4220         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4221                                  ludev : ludev->ld_site->ls_top_dev,
4222                                  fid, NULL);
4223         if (!IS_ERR(luch)) {
4224                 if (lu_object_exists(luch)) {
4225                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4226                         if (lo != NULL)
4227                                 child = osd_obj(lo);
4228                         else
4229                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4230                                                 "lu_object can't be located"
4231                                                 DFID"\n", PFID(fid));
4232
4233                         if (child == NULL) {
4234                                 lu_object_put(env, luch);
4235                                 CERROR("Unable to get osd_object\n");
4236                                 child = ERR_PTR(-ENOENT);
4237                         }
4238                 } else {
4239                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4240                                         "lu_object does not exists "DFID"\n",
4241                                         PFID(fid));
4242                         lu_object_put(env, luch);
4243                         child = ERR_PTR(-ENOENT);
4244                 }
4245         } else {
4246                 child = ERR_CAST(luch);
4247         }
4248
4249         return child;
4250 }
4251
4252 /**
4253  * Put the osd object once done with it.
4254  *
4255  * \param obj osd object that needs to be put
4256  */
4257 static inline void osd_object_put(const struct lu_env *env,
4258                                   struct osd_object *obj)
4259 {
4260         lu_object_put(env, &obj->oo_dt.do_lu);
4261 }
4262
4263 static int osd_index_declare_ea_insert(const struct lu_env *env,
4264                                        struct dt_object *dt,
4265                                        const struct dt_rec *rec,
4266                                        const struct dt_key *key,
4267                                        struct thandle *handle)
4268 {
4269         struct osd_thandle      *oh;
4270         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4271         struct lu_fid           *fid = (struct lu_fid *)rec;
4272         int                     rc;
4273         ENTRY;
4274
4275         LASSERT(!dt_object_remote(dt));
4276         LASSERT(handle != NULL);
4277
4278         oh = container_of0(handle, struct osd_thandle, ot_super);
4279         LASSERT(oh->ot_handle == NULL);
4280
4281         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4282                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4283
4284         if (osd_dt_obj(dt)->oo_inode != NULL) {
4285                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4286
4287                 /* We ignore block quota on meta pool (MDTs), so needn't
4288                  * calculate how many blocks will be consumed by this index
4289                  * insert */
4290                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4291                                            i_gid_read(inode), 0, oh,
4292                                            osd_dt_obj(dt), true, NULL, false);
4293         }
4294
4295         if (fid == NULL)
4296                 RETURN(0);
4297
4298         rc = osd_remote_fid(env, osd, fid);
4299         if (rc <= 0)
4300                 RETURN(rc);
4301
4302         rc = 0;
4303
4304         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4305                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4306         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4307                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4308         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4309                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4310
4311         RETURN(rc);
4312 }
4313
4314 /**
4315  * Index add function for interoperability mode (b11826).
4316  * It will add the directory entry.This entry is needed to
4317  * maintain name->fid mapping.
4318  *
4319  * \param key it is key i.e. file entry to be inserted
4320  * \param rec it is value of given key i.e. fid
4321  *
4322  * \retval   0, on success
4323  * \retval -ve, on error
4324  */
4325 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4326                                const struct dt_rec *rec,
4327                                const struct dt_key *key, struct thandle *th,
4328                                int ignore_quota)
4329 {
4330         struct osd_object       *obj = osd_dt_obj(dt);
4331         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4332         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4333         const struct lu_fid     *fid    = rec1->rec_fid;
4334         const char              *name = (const char *)key;
4335         struct osd_thread_info  *oti   = osd_oti_get(env);
4336         struct osd_inode_id     *id    = &oti->oti_id;
4337         struct inode            *child_inode = NULL;
4338         struct osd_object       *child = NULL;
4339         int                     rc;
4340         ENTRY;
4341
4342         if (!dt_object_exists(dt))
4343                 RETURN(-ENOENT);
4344
4345         LASSERT(osd_invariant(obj));
4346         LASSERT(!dt_object_remote(dt));
4347         LASSERT(th != NULL);
4348
4349         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4350
4351         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4352
4353         rc = osd_remote_fid(env, osd, fid);
4354         if (rc < 0) {
4355                 CERROR("%s: Can not find object "DFID" rc %d\n",
4356                        osd_name(osd), PFID(fid), rc);
4357                 RETURN(rc);
4358         }
4359
4360         if (rc == 1) {
4361                 /* Insert remote entry */
4362                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4363                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4364                         struct osd_thandle      *oh;
4365
4366                         /* If parent on remote MDT, we need put this object
4367                          * under AGENT */
4368                         oh = container_of(th, typeof(*oh), ot_super);
4369                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4370                         if (rc != 0) {
4371                                 CERROR("%s: add "DFID" error: rc = %d\n",
4372                                        osd_name(osd),
4373                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4374                                 RETURN(rc);
4375                         }
4376
4377                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4378                 } else {
4379                         child_inode = osd_create_local_agent_inode(env, osd,
4380                                         obj, fid, rec1->rec_type & S_IFMT, th);
4381                         if (IS_ERR(child_inode))
4382                                 RETURN(PTR_ERR(child_inode));
4383                 }
4384         } else {
4385                 /* Insert local entry */
4386                 child = osd_object_find(env, dt, fid);
4387                 if (IS_ERR(child)) {
4388                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4389                                osd_name(osd), PFID(fid),
4390                                id->oii_ino, id->oii_gen,
4391                                (int)PTR_ERR(child));
4392                         RETURN(PTR_ERR(child));
4393                 }
4394                 child_inode = igrab(child->oo_inode);
4395         }
4396
4397         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4398
4399         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4400                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4401
4402         iput(child_inode);
4403         if (child != NULL)
4404                 osd_object_put(env, child);
4405         LASSERT(osd_invariant(obj));
4406         RETURN(rc);
4407 }
4408
4409 /**
4410  *  Initialize osd Iterator for given osd index object.
4411  *
4412  *  \param  dt      osd index object
4413  */
4414
4415 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4416                                      struct dt_object *dt,
4417                                      __u32 unused)
4418 {
4419         struct osd_it_iam      *it;
4420         struct osd_object      *obj = osd_dt_obj(dt);
4421         struct lu_object       *lo  = &dt->do_lu;
4422         struct iam_path_descr  *ipd;
4423         struct iam_container   *bag = &obj->oo_dir->od_container;
4424
4425         if (!dt_object_exists(dt))
4426                 return ERR_PTR(-ENOENT);
4427
4428         OBD_ALLOC_PTR(it);
4429         if (it == NULL)
4430                 return ERR_PTR(-ENOMEM);
4431
4432         ipd = osd_it_ipd_get(env, bag);
4433         if (likely(ipd != NULL)) {
4434                 it->oi_obj = obj;
4435                 it->oi_ipd = ipd;
4436                 lu_object_get(lo);
4437                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4438                 return (struct dt_it *)it;
4439         } else {
4440                 OBD_FREE_PTR(it);
4441                 return ERR_PTR(-ENOMEM);
4442         }
4443 }
4444
4445 /**
4446  * free given Iterator.
4447  */
4448
4449 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4450 {
4451         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4452         struct osd_object       *obj = it->oi_obj;
4453
4454         iam_it_fini(&it->oi_it);
4455         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4456         lu_object_put(env, &obj->oo_dt.do_lu);
4457         OBD_FREE_PTR(it);
4458 }
4459
4460 /**
4461  *  Move Iterator to record specified by \a key
4462  *
4463  *  \param  di      osd iterator
4464  *  \param  key     key for index
4465  *
4466  *  \retval +ve  di points to record with least key not larger than key
4467  *  \retval  0   di points to exact matched key
4468  *  \retval -ve  failure
4469  */
4470
4471 static int osd_it_iam_get(const struct lu_env *env,
4472                           struct dt_it *di, const struct dt_key *key)
4473 {
4474         struct osd_thread_info  *oti = osd_oti_get(env);
4475         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4476
4477         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4478                 /* swab quota uid/gid */
4479                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4480                 key = (struct dt_key *)&oti->oti_quota_id;
4481         }
4482
4483         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4484 }
4485
4486 /**
4487  *  Release Iterator
4488  *
4489  *  \param  di      osd iterator
4490  */
4491 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4492 {
4493         struct osd_it_iam *it = (struct osd_it_iam *)di;
4494
4495         iam_it_put(&it->oi_it);
4496 }
4497
4498 /**
4499  *  Move iterator by one record
4500  *
4501  *  \param  di      osd iterator
4502  *
4503  *  \retval +1   end of container reached
4504  *  \retval  0   success
4505  *  \retval -ve  failure
4506  */
4507
4508 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4509 {
4510         struct osd_it_iam *it = (struct osd_it_iam *)di;
4511
4512         return iam_it_next(&it->oi_it);
4513 }
4514
4515 /**
4516  * Return pointer to the key under iterator.
4517  */
4518
4519 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4520                                  const struct dt_it *di)
4521 {
4522         struct osd_thread_info *oti = osd_oti_get(env);
4523         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4524         struct osd_object      *obj = it->oi_obj;
4525         struct dt_key          *key;
4526
4527         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4528
4529         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4530                 /* swab quota uid/gid */
4531                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4532                 key = (struct dt_key *)&oti->oti_quota_id;
4533         }
4534
4535         return key;
4536 }
4537
4538 /**
4539  * Return size of key under iterator (in bytes)
4540  */
4541
4542 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4543 {
4544         struct osd_it_iam *it = (struct osd_it_iam *)di;
4545
4546         return iam_it_key_size(&it->oi_it);
4547 }
4548
4549 static inline void
4550 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4551 {
4552         /* check if file type is required */
4553         if (ent->lde_attrs & LUDA_TYPE) {
4554                 struct luda_type *lt;
4555                 int align = sizeof(*lt) - 1;
4556
4557                 len = (len + align) & ~align;
4558                 lt = (struct luda_type *)(ent->lde_name + len);
4559                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4560         }
4561
4562         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4563 }
4564
4565 /**
4566  * build lu direct from backend fs dirent.
4567  */
4568
4569 static inline void
4570 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4571                    char *name, __u16 namelen, __u16 type, __u32 attr)
4572 {
4573         ent->lde_attrs = attr | LUDA_FID;
4574         fid_cpu_to_le(&ent->lde_fid, fid);
4575
4576         ent->lde_hash = cpu_to_le64(offset);
4577         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4578
4579         strncpy(ent->lde_name, name, namelen);
4580         ent->lde_name[namelen] = '\0';
4581         ent->lde_namelen = cpu_to_le16(namelen);
4582
4583         /* append lustre attributes */
4584         osd_it_append_attrs(ent, namelen, type);
4585 }
4586
4587 /**
4588  * Return pointer to the record under iterator.
4589  */
4590 static int osd_it_iam_rec(const struct lu_env *env,
4591                           const struct dt_it *di,
4592                           struct dt_rec *dtrec, __u32 attr)
4593 {
4594         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4595         struct osd_thread_info *info = osd_oti_get(env);
4596         ENTRY;
4597
4598         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4599                 const struct osd_fid_pack *rec;
4600                 struct lu_fid             *fid = &info->oti_fid;
4601                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4602                 char                      *name;
4603                 int                        namelen;
4604                 __u64                      hash;
4605                 int                        rc;
4606
4607                 name = (char *)iam_it_key_get(&it->oi_it);
4608                 if (IS_ERR(name))
4609                         RETURN(PTR_ERR(name));
4610
4611                 namelen = iam_it_key_size(&it->oi_it);
4612
4613                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4614                 if (IS_ERR(rec))
4615                         RETURN(PTR_ERR(rec));
4616
4617                 rc = osd_fid_unpack(fid, rec);
4618                 if (rc)
4619                         RETURN(rc);
4620
4621                 hash = iam_it_store(&it->oi_it);
4622
4623                 /* IAM does not store object type in IAM index (dir) */
4624                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4625                                    0, LUDA_FID);
4626         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4627                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4628                            (struct iam_rec *)dtrec);
4629                 osd_quota_unpack(it->oi_obj, dtrec);
4630         } else {
4631                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4632                            (struct iam_rec *)dtrec);
4633         }
4634
4635         RETURN(0);
4636 }
4637
4638 /**
4639  * Returns cookie for current Iterator position.
4640  */
4641 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4642 {
4643         struct osd_it_iam *it = (struct osd_it_iam *)di;
4644
4645         return iam_it_store(&it->oi_it);
4646 }
4647
4648 /**
4649  * Restore iterator from cookie.
4650  *
4651  * \param  di      osd iterator
4652  * \param  hash    Iterator location cookie
4653  *
4654  * \retval +ve  di points to record with least key not larger than key.
4655  * \retval  0   di points to exact matched key
4656  * \retval -ve  failure
4657  */
4658
4659 static int osd_it_iam_load(const struct lu_env *env,
4660                            const struct dt_it *di, __u64 hash)
4661 {
4662         struct osd_it_iam *it = (struct osd_it_iam *)di;
4663
4664         return iam_it_load(&it->oi_it, hash);
4665 }
4666
4667 static const struct dt_index_operations osd_index_iam_ops = {
4668         .dio_lookup         = osd_index_iam_lookup,
4669         .dio_declare_insert = osd_index_declare_iam_insert,
4670         .dio_insert         = osd_index_iam_insert,
4671         .dio_declare_delete = osd_index_declare_iam_delete,
4672         .dio_delete         = osd_index_iam_delete,
4673         .dio_it     = {
4674                 .init     = osd_it_iam_init,
4675                 .fini     = osd_it_iam_fini,
4676                 .get      = osd_it_iam_get,
4677                 .put      = osd_it_iam_put,
4678                 .next     = osd_it_iam_next,
4679                 .key      = osd_it_iam_key,
4680                 .key_size = osd_it_iam_key_size,
4681                 .rec      = osd_it_iam_rec,
4682                 .store    = osd_it_iam_store,
4683                 .load     = osd_it_iam_load
4684         }
4685 };
4686
4687
4688 /**
4689  * Creates or initializes iterator context.
4690  *
4691  * \retval struct osd_it_ea, iterator structure on success
4692  *
4693  */
4694 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4695                                     struct dt_object *dt,
4696                                     __u32 attr)
4697 {
4698         struct osd_object       *obj  = osd_dt_obj(dt);
4699         struct osd_thread_info  *info = osd_oti_get(env);
4700         struct osd_it_ea        *oie;
4701         struct file             *file;
4702         struct lu_object        *lo   = &dt->do_lu;
4703         struct dentry           *obj_dentry;
4704         ENTRY;
4705
4706         if (!dt_object_exists(dt))
4707                 RETURN(ERR_PTR(-ENOENT));
4708
4709         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4710         if (oie == NULL)
4711                 RETURN(ERR_PTR(-ENOMEM));
4712         obj_dentry = &oie->oie_dentry;
4713
4714         obj_dentry->d_inode = obj->oo_inode;
4715         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4716         obj_dentry->d_name.hash = 0;
4717
4718         oie->oie_rd_dirent       = 0;
4719         oie->oie_it_dirent       = 0;
4720         oie->oie_dirent          = NULL;
4721         if (unlikely(!info->oti_it_ea_buf_used)) {
4722                 oie->oie_buf = info->oti_it_ea_buf;
4723                 info->oti_it_ea_buf_used = 1;
4724         } else {
4725                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4726                 if (oie->oie_buf == NULL)
4727                         RETURN(ERR_PTR(-ENOMEM));
4728         }
4729         oie->oie_obj             = obj;
4730
4731         file = &oie->oie_file;
4732
4733         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4734         if (attr & LUDA_64BITHASH)
4735                 file->f_mode    = FMODE_64BITHASH;
4736         else
4737                 file->f_mode    = FMODE_32BITHASH;
4738         file->f_dentry          = obj_dentry;
4739         file->f_mapping         = obj->oo_inode->i_mapping;
4740         file->f_op              = obj->oo_inode->i_fop;
4741         set_file_inode(file, obj->oo_inode);
4742
4743         lu_object_get(lo);
4744         RETURN((struct dt_it *) oie);
4745 }
4746
4747 /**
4748  * Destroy or finishes iterator context.
4749  *
4750  * \param di iterator structure to be destroyed
4751  */
4752 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4753 {
4754         struct osd_thread_info  *info = osd_oti_get(env);
4755         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
4756         struct osd_object       *obj    = oie->oie_obj;
4757         struct inode            *inode  = obj->oo_inode;
4758
4759         ENTRY;
4760         oie->oie_file.f_op->release(inode, &oie->oie_file);
4761         lu_object_put(env, &obj->oo_dt.do_lu);
4762         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4763                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4764         else
4765                 info->oti_it_ea_buf_used = 0;
4766         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4767         EXIT;
4768 }
4769
4770 /**
4771  * It position the iterator at given key, so that next lookup continues from
4772  * that key Or it is similar to dio_it->load() but based on a key,
4773  * rather than file position.
4774  *
4775  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4776  * to the beginning.
4777  *
4778  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4779  */
4780 static int osd_it_ea_get(const struct lu_env *env,
4781                          struct dt_it *di, const struct dt_key *key)
4782 {
4783         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4784
4785         ENTRY;
4786         LASSERT(((const char *)key)[0] == '\0');
4787         it->oie_file.f_pos      = 0;
4788         it->oie_rd_dirent       = 0;
4789         it->oie_it_dirent       = 0;
4790         it->oie_dirent          = NULL;
4791
4792         RETURN(+1);
4793 }
4794
4795 /**
4796  * Does nothing
4797  */
4798 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4799 {
4800 }
4801
4802 struct osd_filldir_cbs {
4803 #ifdef HAVE_DIR_CONTEXT
4804         struct dir_context ctx;
4805 #endif
4806         struct osd_it_ea  *it;
4807 };
4808 /**
4809  * It is called internally by ->readdir(). It fills the
4810  * iterator's in-memory data structure with required
4811  * information i.e. name, namelen, rec_size etc.
4812  *
4813  * \param buf in which information to be filled in.
4814  * \param name name of the file in given dir
4815  *
4816  * \retval 0 on success
4817  * \retval 1 on buffer full
4818  */
4819 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4820                                loff_t offset, __u64 ino,
4821                                unsigned d_type)
4822 {
4823         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4824         struct osd_object       *obj  = it->oie_obj;
4825         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4826         struct lu_fid           *fid  = &ent->oied_fid;
4827         struct osd_fid_pack     *rec;
4828         ENTRY;
4829
4830         /* this should never happen */
4831         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4832                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4833                 RETURN(-EIO);
4834         }
4835
4836         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4837             OSD_IT_EA_BUFSIZE)
4838                 RETURN(1);
4839
4840         /* "." is just the object itself. */
4841         if (namelen == 1 && name[0] == '.') {
4842                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4843         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4844                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4845                 if (osd_fid_unpack(fid, rec) != 0)
4846                         fid_zero(fid);
4847         } else {
4848                 fid_zero(fid);
4849         }
4850         d_type &= ~LDISKFS_DIRENT_LUFID;
4851
4852         /* NOT export local root. */
4853         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4854                 ino = obj->oo_inode->i_ino;
4855                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4856         }
4857
4858         ent->oied_ino     = ino;
4859         ent->oied_off     = offset;
4860         ent->oied_namelen = namelen;
4861         ent->oied_type    = d_type;
4862
4863         memcpy(ent->oied_name, name, namelen);
4864
4865         it->oie_rd_dirent++;
4866         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4867         RETURN(0);
4868 }
4869
4870 /**
4871  * Calls ->readdir() to load a directory entry at a time
4872  * and stored it in iterator's in-memory data structure.
4873  *
4874  * \param di iterator's in memory structure
4875  *
4876  * \retval   0 on success
4877  * \retval -ve on error
4878  * \retval +1 reach the end of entry
4879  */
4880 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4881                                const struct dt_it *di)
4882 {
4883         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4884         struct osd_object  *obj   = it->oie_obj;
4885         struct inode       *inode = obj->oo_inode;
4886         struct htree_lock  *hlock = NULL;
4887         struct file        *filp  = &it->oie_file;
4888         int                 rc = 0;
4889         struct osd_filldir_cbs buf = {
4890 #ifdef HAVE_DIR_CONTEXT
4891                 .ctx.actor = osd_ldiskfs_filldir,
4892 #endif
4893                 .it = it
4894         };
4895
4896         ENTRY;
4897         it->oie_dirent = it->oie_buf;
4898         it->oie_rd_dirent = 0;
4899
4900         if (obj->oo_hl_head != NULL) {
4901                 hlock = osd_oti_get(env)->oti_hlock;
4902                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4903                                    inode, LDISKFS_HLOCK_READDIR);
4904         } else {
4905                 down_read(&obj->oo_ext_idx_sem);
4906         }
4907
4908 #ifdef HAVE_DIR_CONTEXT
4909         buf.ctx.pos = filp->f_pos;
4910         rc = inode->i_fop->iterate(filp, &buf.ctx);
4911         filp->f_pos = buf.ctx.pos;
4912 #else
4913         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4914 #endif
4915
4916         if (hlock != NULL)
4917                 ldiskfs_htree_unlock(hlock);
4918         else
4919                 up_read(&obj->oo_ext_idx_sem);
4920
4921         if (it->oie_rd_dirent == 0) {
4922                 /*If it does not get any dirent, it means it has been reached
4923                  *to the end of the dir */
4924                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4925                 if (rc == 0)
4926                         rc = 1;
4927         } else {
4928                 it->oie_dirent = it->oie_buf;
4929                 it->oie_it_dirent = 1;
4930         }
4931
4932         RETURN(rc);
4933 }
4934
4935 /**
4936  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4937  * to load a directory entry at a time and stored it in
4938  * iterator's in-memory data structure.
4939  *
4940  * \param di iterator's in memory structure
4941  *
4942  * \retval +ve iterator reached to end
4943  * \retval   0 iterator not reached to end
4944  * \retval -ve on error
4945  */
4946 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4947 {
4948         struct osd_it_ea *it = (struct osd_it_ea *)di;
4949         int rc;
4950
4951         ENTRY;
4952
4953         if (it->oie_it_dirent < it->oie_rd_dirent) {
4954                 it->oie_dirent =
4955                         (void *) it->oie_dirent +
4956                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4957                                        it->oie_dirent->oied_namelen);
4958                 it->oie_it_dirent++;
4959                 RETURN(0);
4960         } else {
4961                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4962                         rc = +1;
4963                 else
4964                         rc = osd_ldiskfs_it_fill(env, di);
4965         }
4966
4967         RETURN(rc);
4968 }
4969
4970 /**
4971  * Returns the key at current position from iterator's in memory structure.
4972  *
4973  * \param di iterator's in memory structure
4974  *
4975  * \retval key i.e. struct dt_key on success
4976  */
4977 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4978                                     const struct dt_it *di)
4979 {
4980         struct osd_it_ea *it = (struct osd_it_ea *)di;
4981
4982         return (struct dt_key *)it->oie_dirent->oied_name;
4983 }
4984
4985 /**
4986  * Returns the key's size at current position from iterator's in memory structure.
4987  *
4988  * \param di iterator's in memory structure
4989  *
4990  * \retval key_size i.e. struct dt_key on success
4991  */
4992 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4993 {
4994         struct osd_it_ea *it = (struct osd_it_ea *)di;
4995
4996         return it->oie_dirent->oied_namelen;
4997 }
4998
4999 static int
5000 osd_dirent_update(handle_t *jh, struct super_block *sb,
5001                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5002                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5003 {
5004         struct osd_fid_pack *rec;
5005         int                  rc;
5006         ENTRY;
5007
5008         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5009         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5010
5011         rc = ldiskfs_journal_get_write_access(jh, bh);
5012         if (rc != 0)
5013                 RETURN(rc);
5014
5015         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5016         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5017         rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5018
5019         RETURN(rc);
5020 }
5021
5022 static inline int
5023 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5024 {
5025         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5026             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5027                 return 1;
5028         else
5029                 return 0;
5030 }
5031
5032 static inline int
5033 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5034 {
5035         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5036                  "dot_dotdot = %d\n", dot_dotdot);
5037
5038         if (LDISKFS_DIR_REC_LEN(de) >=
5039             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5040                 return 1;
5041         else
5042                 return 0;
5043 }
5044
5045 static int
5046 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5047                     struct inode *dir, struct inode *inode,
5048                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5049                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5050                     struct htree_lock *hlock)
5051 {
5052         struct dentry               *dentry;
5053         struct osd_fid_pack         *rec;
5054         struct ldiskfs_dentry_param *ldp;
5055         int                          rc;
5056         ENTRY;
5057
5058         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5059                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5060                 RETURN(0);
5061
5062         /* There is enough space to hold the FID-in-dirent. */
5063         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5064                                  dir->i_sb->s_blocksize)) {
5065                 rc = ldiskfs_journal_get_write_access(jh, bh);
5066                 if (rc != 0)
5067                         RETURN(rc);
5068
5069                 de->name[de->name_len] = 0;
5070                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5071                 rec->fp_len = sizeof(struct lu_fid) + 1;
5072                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5073                 de->file_type |= LDISKFS_DIRENT_LUFID;
5074
5075                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5076
5077                 RETURN(rc);
5078         }
5079
5080         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5081         if (rc != 0)
5082                 RETURN(rc);
5083
5084         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5085                                            ent->oied_namelen);
5086         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5087         osd_get_ldiskfs_dirent_param(ldp, fid);
5088         dentry->d_fsdata = (void *)ldp;
5089         ll_vfs_dq_init(dir);
5090         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5091         /* It is too bad, we cannot reinsert the name entry back.
5092          * That means we lose it! */
5093         if (rc != 0)
5094                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5095                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5096                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5097                        dir->i_ino, dir->i_generation,
5098                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5099
5100         RETURN(rc);
5101 }
5102
5103 static int
5104 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5105                         struct osd_it_ea *it, struct lu_fid *fid,
5106                         struct osd_inode_id *id, __u32 *attr)
5107 {
5108         struct osd_thread_info     *info        = osd_oti_get(env);
5109         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5110         struct osd_device          *dev         = osd_obj2dev(obj);
5111         struct super_block         *sb          = osd_sb(dev);
5112         const char                 *devname     =
5113                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5114         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5115         struct inode               *dir         = obj->oo_inode;
5116         struct htree_lock          *hlock       = NULL;
5117         struct buffer_head         *bh          = NULL;
5118         handle_t                   *jh          = NULL;
5119         struct ldiskfs_dir_entry_2 *de;
5120         struct dentry              *dentry;
5121         struct inode               *inode;
5122         int                         credits;
5123         int                         rc;
5124         int                         dot_dotdot  = 0;
5125         bool                        dirty       = false;
5126         ENTRY;
5127
5128         if (ent->oied_name[0] == '.') {
5129                 if (ent->oied_namelen == 1)
5130                         dot_dotdot = 1;
5131                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5132                         dot_dotdot = 2;
5133         }
5134
5135         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5136                                       ent->oied_namelen);
5137
5138         /* We need to ensure that the name entry is still valid.
5139          * Because it may be removed or renamed by other already.
5140          *
5141          * The unlink or rename operation will start journal before PDO lock,
5142          * so to avoid deadlock, here we need to start journal handle before
5143          * related PDO lock also. But because we do not know whether there
5144          * will be something to be repaired before PDO lock, we just start
5145          * journal without conditions.
5146          *
5147          * We may need to remove the name entry firstly, then insert back.
5148          * One credit is for user quota file update.
5149          * One credit is for group quota file update.
5150          * Two credits are for dirty inode. */
5151         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5152                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5153
5154 again:
5155         if (dev->od_dirent_journal != 0) {
5156                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5157                 if (IS_ERR(jh)) {
5158                         rc = PTR_ERR(jh);
5159                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5160                                "check_repair, dir = %lu/%u, credits = %d, "
5161                                "name = %.*s: rc = %d\n",
5162                                devname, dir->i_ino, dir->i_generation, credits,
5163                                ent->oied_namelen, ent->oied_name, rc);
5164                         RETURN(rc);
5165                 }
5166
5167                 if (obj->oo_hl_head != NULL) {
5168                         hlock = osd_oti_get(env)->oti_hlock;
5169                         /* "0" means exclusive lock for the whole directory.
5170                          * We need to prevent others access such name entry
5171                          * during the delete + insert. Neither HLOCK_ADD nor
5172                          * HLOCK_DEL cannot guarantee the atomicity. */
5173                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5174                 } else {
5175                         down_write(&obj->oo_ext_idx_sem);
5176                 }
5177         } else {
5178                 if (obj->oo_hl_head != NULL) {
5179                         hlock = osd_oti_get(env)->oti_hlock;
5180                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5181                                            LDISKFS_HLOCK_LOOKUP);
5182                 } else {
5183                         down_read(&obj->oo_ext_idx_sem);
5184                 }
5185         }
5186
5187         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5188         /* For dot/dotdot entry, if there is not enough space to hold the
5189          * FID-in-dirent, just keep them there. It only happens when the
5190          * device upgraded from 1.8 or restored from MDT file-level backup.
5191          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5192          * and needs to get FID from LMA when readdir, it will not affect the
5193          * performance much. */
5194         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5195             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5196                 *attr |= LUDA_IGNORE;
5197                 GOTO(out_journal, rc = 0);
5198         }
5199
5200         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5201         inode = osd_iget(info, dev, id);
5202         if (IS_ERR(inode)) {
5203                 rc = PTR_ERR(inode);
5204                 if (rc == -ENOENT || rc == -ESTALE)
5205                         rc = 1;
5206                 else
5207                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5208                                "check_repair, dir = %lu/%u, name = %.*s: "
5209                                "rc = %d\n",
5210                                devname, dir->i_ino, dir->i_generation,
5211                                ent->oied_namelen, ent->oied_name, rc);
5212
5213                 GOTO(out_journal, rc);
5214         }
5215
5216         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5217         if (rc == 0) {
5218                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5219
5220                 if (fid_is_sane(fid)) {
5221                         /* FID-in-dirent is valid. */
5222                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5223                                 GOTO(out_inode, rc = 0);
5224
5225                         /* Do not repair under dryrun mode. */
5226                         if (*attr & LUDA_VERIFY_DRYRUN) {
5227                                 *attr |= LUDA_REPAIR;
5228                                 GOTO(out_inode, rc = 0);
5229                         }
5230
5231                         if (dev->od_dirent_journal == 0) {
5232                                 iput(inode);
5233                                 brelse(bh);
5234                                 if (hlock != NULL)
5235                                         ldiskfs_htree_unlock(hlock);
5236                                 else
5237                                         up_read(&obj->oo_ext_idx_sem);
5238                                 dev->od_dirent_journal = 1;
5239                                 goto again;
5240                         }
5241
5242                         *fid = lma->lma_self_fid;
5243                         dirty = true;
5244                         /* Update the FID-in-dirent. */
5245                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5246                         if (rc == 0)
5247                                 *attr |= LUDA_REPAIR;
5248                         else
5249                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5250                                        "in the dirent, dir = %lu/%u, "
5251                                        "name = %.*s, "DFID": rc = %d\n",
5252                                        devname, dir->i_ino, dir->i_generation,
5253                                        ent->oied_namelen, ent->oied_name,
5254                                        PFID(fid), rc);
5255                 } else {
5256                         /* Do not repair under dryrun mode. */
5257                         if (*attr & LUDA_VERIFY_DRYRUN) {
5258                                 *fid = lma->lma_self_fid;
5259                                 *attr |= LUDA_REPAIR;
5260                                 GOTO(out_inode, rc = 0);
5261                         }
5262
5263                         if (dev->od_dirent_journal == 0) {
5264                                 iput(inode);
5265                                 brelse(bh);
5266                                 if (hlock != NULL)
5267                                         ldiskfs_htree_unlock(hlock);
5268                                 else
5269                                         up_read(&obj->oo_ext_idx_sem);
5270                                 dev->od_dirent_journal = 1;
5271                                 goto again;
5272                         }
5273
5274                         *fid = lma->lma_self_fid;
5275                         dirty = true;
5276                         /* Append the FID-in-dirent. */
5277                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5278                                                  fid, bh, de, hlock);
5279                         if (rc == 0)
5280                                 *attr |= LUDA_REPAIR;
5281                         else
5282                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5283                                        "after the dirent, dir = %lu/%u, "
5284                                        "name = %.*s, "DFID": rc = %d\n",
5285                                        devname, dir->i_ino, dir->i_generation,
5286                                        ent->oied_namelen, ent->oied_name,
5287                                        PFID(fid), rc);
5288                 }
5289         } else if (rc == -ENODATA) {
5290                 /* Do not repair under dryrun mode. */
5291                 if (*attr & LUDA_VERIFY_DRYRUN) {
5292                         if (fid_is_sane(fid)) {
5293                                 *attr |= LUDA_REPAIR;
5294                         } else {
5295                                 lu_igif_build(fid, inode->i_ino,
5296                                               inode->i_generation);
5297                                 *attr |= LUDA_UPGRADE;
5298                         }
5299                         GOTO(out_inode, rc = 0);
5300                 }
5301
5302                 if (dev->od_dirent_journal == 0) {
5303                         iput(inode);
5304                         brelse(bh);
5305                         if (hlock != NULL)
5306                                 ldiskfs_htree_unlock(hlock);
5307                         else
5308                                 up_read(&obj->oo_ext_idx_sem);
5309                         dev->od_dirent_journal = 1;
5310                         goto again;
5311                 }
5312
5313                 dirty = true;
5314                 if (unlikely(fid_is_sane(fid))) {
5315                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5316                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5317                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5318                         if (rc == 0)
5319                                 *attr |= LUDA_REPAIR;
5320                         else
5321                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5322                                        "update dirent, dir = %lu/%u, "
5323                                        "name = %.*s, "DFID": rc = %d\n",
5324                                        devname, dir->i_ino, dir->i_generation,
5325                                        ent->oied_namelen, ent->oied_name,
5326                                        PFID(fid), rc);
5327                 } else {
5328                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5329                         /* It is probably IGIF object. Only aappend the
5330                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5331                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5332                                                  fid, bh, de, hlock);
5333                         if (rc == 0)
5334                                 *attr |= LUDA_UPGRADE;
5335                         else
5336                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5337                                        "after the dirent, dir = %lu/%u, "
5338                                        "name = %.*s, "DFID": rc = %d\n",
5339                                        devname, dir->i_ino, dir->i_generation,
5340                                        ent->oied_namelen, ent->oied_name,
5341                                        PFID(fid), rc);
5342                 }
5343         }
5344
5345         GOTO(out_inode, rc);
5346
5347 out_inode:
5348         iput(inode);
5349
5350 out_journal:
5351         brelse(bh);
5352         if (hlock != NULL) {
5353                 ldiskfs_htree_unlock(hlock);
5354         } else {
5355                 if (dev->od_dirent_journal != 0)
5356                         up_write(&obj->oo_ext_idx_sem);
5357                 else
5358                         up_read(&obj->oo_ext_idx_sem);
5359         }
5360         if (jh != NULL)
5361                 ldiskfs_journal_stop(jh);
5362         if (rc >= 0 && !dirty)
5363                 dev->od_dirent_journal = 0;
5364         return rc;
5365 }
5366
5367 /**
5368  * Returns the value at current position from iterator's in memory structure.
5369  *
5370  * \param di struct osd_it_ea, iterator's in memory structure
5371  * \param attr attr requested for dirent.
5372  * \param lde lustre dirent
5373  *
5374  * \retval   0 no error and \param lde has correct lustre dirent.
5375  * \retval -ve on error
5376  */
5377 static inline int osd_it_ea_rec(const struct lu_env *env,
5378                                 const struct dt_it *di,
5379                                 struct dt_rec *dtrec, __u32 attr)
5380 {
5381         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5382         struct osd_object      *obj   = it->oie_obj;
5383         struct osd_device      *dev   = osd_obj2dev(obj);
5384         struct osd_thread_info *oti   = osd_oti_get(env);
5385         struct osd_inode_id    *id    = &oti->oti_id;
5386         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5387         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5388         __u32                   ino   = it->oie_dirent->oied_ino;
5389         int                     rc    = 0;
5390         ENTRY;
5391
5392         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5393
5394         if (attr & LUDA_VERIFY) {
5395                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5396                         attr |= LUDA_IGNORE;
5397                         /* If the parent is on remote MDT, and there
5398                          * is no FID-in-dirent, then we have to get
5399                          * the parent FID from the linkEA.  */
5400                         if (!fid_is_sane(fid) &&
5401                             it->oie_dirent->oied_namelen == 2 &&
5402                             it->oie_dirent->oied_name[0] == '.' &&
5403                             it->oie_dirent->oied_name[1] == '.')
5404                                 osd_get_pfid_from_linkea(env, obj, fid);
5405                 } else {
5406                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5407                                                      &attr);
5408                 }
5409         } else {
5410                 attr &= ~LU_DIRENT_ATTRS_MASK;
5411                 if (!fid_is_sane(fid)) {
5412                         if (it->oie_dirent->oied_namelen == 2 &&
5413                             it->oie_dirent->oied_name[0] == '.' &&
5414                             it->oie_dirent->oied_name[1] == '.') {
5415                                 /* If the parent is on remote MDT, and there
5416                                  * is no FID-in-dirent, then we have to get
5417                                  * the parent FID from the linkEA.  */
5418                                 if (ino == osd_remote_parent_ino(dev))
5419                                         rc = osd_get_pfid_from_linkea(env, obj,
5420                                                                       fid);
5421                         } else {
5422                                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5423                                         RETURN(-ENOENT);
5424
5425                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5426                         }
5427                 } else {
5428                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5429                 }
5430         }
5431
5432         /* Pack the entry anyway, at least the offset is right. */
5433         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5434                            it->oie_dirent->oied_name,
5435                            it->oie_dirent->oied_namelen,
5436                            it->oie_dirent->oied_type, attr);
5437
5438         if (rc < 0)
5439                 RETURN(rc);
5440
5441         if (osd_remote_fid(env, dev, fid))
5442                 RETURN(0);
5443
5444         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5445                 osd_add_oi_cache(oti, dev, id, fid);
5446
5447         RETURN(rc > 0 ? 0 : rc);
5448 }
5449
5450 /**
5451  * Returns the record size size at current position.
5452  *
5453  * This function will return record(lu_dirent) size in bytes.
5454  *
5455  * \param[in] env       execution environment
5456  * \param[in] di        iterator's in memory structure
5457  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5458  *                      calculate the lu_dirent size.
5459  *
5460  * \retval      record size(in bytes & in memory) of the current lu_dirent
5461  *              entry.
5462  */
5463 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5464                               __u32 attr)
5465 {
5466         struct osd_it_ea *it = (struct osd_it_ea *)di;
5467
5468         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5469 }
5470
5471 /**
5472  * Returns a cookie for current position of the iterator head, so that
5473  * user can use this cookie to load/start the iterator next time.
5474  *
5475  * \param di iterator's in memory structure
5476  *
5477  * \retval cookie for current position, on success
5478  */
5479 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5480 {
5481         struct osd_it_ea *it = (struct osd_it_ea *)di;
5482
5483         return it->oie_dirent->oied_off;
5484 }
5485
5486 /**
5487  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5488  * to load a directory entry at a time and stored it i inn,
5489  * in iterator's in-memory data structure.
5490  *
5491  * \param di struct osd_it_ea, iterator's in memory structure
5492  *
5493  * \retval +ve on success
5494  * \retval -ve on error
5495  */
5496 static int osd_it_ea_load(const struct lu_env *env,
5497                           const struct dt_it *di, __u64 hash)
5498 {
5499         struct osd_it_ea *it = (struct osd_it_ea *)di;
5500         int rc;
5501
5502         ENTRY;
5503         it->oie_file.f_pos = hash;
5504
5505         rc =  osd_ldiskfs_it_fill(env, di);
5506         if (rc > 0)
5507                 rc = -ENODATA;
5508
5509         if (rc == 0)
5510                 rc = +1;
5511
5512         RETURN(rc);
5513 }
5514
5515 /**
5516  * Index lookup function for interoperability mode (b11826).
5517  *
5518  * \param key,  key i.e. file name to be searched
5519  *
5520  * \retval +ve, on success
5521  * \retval -ve, on error
5522  */
5523 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5524                                struct dt_rec *rec, const struct dt_key *key)
5525 {
5526         struct osd_object *obj = osd_dt_obj(dt);
5527         int rc = 0;
5528
5529         ENTRY;
5530
5531         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5532         LINVRNT(osd_invariant(obj));
5533
5534         rc = osd_ea_lookup_rec(env, obj, rec, key);
5535         if (rc == 0)
5536                 rc = +1;
5537         RETURN(rc);
5538 }
5539
5540 /**
5541  * Index and Iterator operations for interoperability
5542  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5543  */
5544 static const struct dt_index_operations osd_index_ea_ops = {
5545         .dio_lookup         = osd_index_ea_lookup,
5546         .dio_declare_insert = osd_index_declare_ea_insert,
5547         .dio_insert         = osd_index_ea_insert,
5548         .dio_declare_delete = osd_index_declare_ea_delete,
5549         .dio_delete         = osd_index_ea_delete,
5550         .dio_it     = {
5551                 .init     = osd_it_ea_init,
5552                 .fini     = osd_it_ea_fini,
5553                 .get      = osd_it_ea_get,
5554                 .put      = osd_it_ea_put,
5555                 .next     = osd_it_ea_next,
5556                 .key      = osd_it_ea_key,
5557                 .key_size = osd_it_ea_key_size,
5558                 .rec      = osd_it_ea_rec,
5559                 .rec_size = osd_it_ea_rec_size,
5560                 .store    = osd_it_ea_store,
5561                 .load     = osd_it_ea_load
5562         }
5563 };
5564
5565 static void *osd_key_init(const struct lu_context *ctx,
5566                           struct lu_context_key *key)
5567 {
5568         struct osd_thread_info *info;
5569
5570         OBD_ALLOC_PTR(info);
5571         if (info == NULL)
5572                 return ERR_PTR(-ENOMEM);
5573
5574         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5575         if (info->oti_it_ea_buf == NULL)
5576                 goto out_free_info;
5577
5578         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5579
5580         info->oti_hlock = ldiskfs_htree_lock_alloc();
5581         if (info->oti_hlock == NULL)
5582                 goto out_free_ea;
5583
5584         return info;
5585
5586  out_free_ea:
5587         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5588  out_free_info:
5589         OBD_FREE_PTR(info);
5590         return ERR_PTR(-ENOMEM);
5591 }
5592
5593 static void osd_key_fini(const struct lu_context *ctx,
5594                          struct lu_context_key *key, void* data)
5595 {
5596         struct osd_thread_info *info = data;
5597
5598         if (info->oti_inode != NULL)
5599                 OBD_FREE_PTR(info->oti_inode);
5600         if (info->oti_hlock != NULL)
5601                 ldiskfs_htree_lock_free(info->oti_hlock);
5602         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5603         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5604         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5605         lu_buf_free(&info->oti_big_buf);
5606         OBD_FREE_PTR(info);
5607 }
5608
5609 static void osd_key_exit(const struct lu_context *ctx,
5610                          struct lu_context_key *key, void *data)
5611 {
5612         struct osd_thread_info *info = data;
5613
5614         LASSERT(info->oti_r_locks == 0);
5615         LASSERT(info->oti_w_locks == 0);
5616         LASSERT(info->oti_txns    == 0);
5617 }
5618
5619 /* type constructor/destructor: osd_type_init, osd_type_fini */
5620 LU_TYPE_INIT_FINI(osd, &osd_key);
5621
5622 struct lu_context_key osd_key = {
5623         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5624         .lct_init = osd_key_init,
5625         .lct_fini = osd_key_fini,
5626         .lct_exit = osd_key_exit
5627 };
5628
5629
5630 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5631                            const char *name, struct lu_device *next)
5632 {
5633         struct osd_device *osd = osd_dev(d);
5634
5635         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5636             >= sizeof(osd->od_svname))
5637                 return -E2BIG;
5638         return osd_procfs_init(osd, name);
5639 }
5640
5641 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5642 {
5643         struct seq_server_site  *ss = osd_seq_site(osd);
5644         int                     rc;
5645         ENTRY;
5646
5647         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5648                 RETURN(0);
5649
5650         if (unlikely(ss == NULL))
5651                 RETURN(-ENODEV);
5652
5653         OBD_ALLOC_PTR(osd->od_cl_seq);
5654         if (osd->od_cl_seq == NULL)
5655                 RETURN(-ENOMEM);
5656
5657         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5658                              osd->od_svname, ss->ss_server_seq);
5659
5660         if (rc != 0) {
5661                 OBD_FREE_PTR(osd->od_cl_seq);
5662                 osd->od_cl_seq = NULL;
5663         }
5664
5665         RETURN(rc);
5666 }
5667
5668 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5669 {
5670         if (osd->od_cl_seq == NULL)
5671                 return;
5672
5673         seq_client_fini(osd->od_cl_seq);
5674         OBD_FREE_PTR(osd->od_cl_seq);
5675         osd->od_cl_seq = NULL;
5676 }
5677
5678 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5679 {
5680         ENTRY;
5681
5682         /* shutdown quota slave instance associated with the device */
5683         if (o->od_quota_slave != NULL) {
5684                 qsd_fini(env, o->od_quota_slave);
5685                 o->od_quota_slave = NULL;
5686         }
5687
5688         osd_fid_fini(env, o);
5689
5690         RETURN(0);
5691 }
5692
5693 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5694 {
5695         ENTRY;
5696
5697         if (o->od_mnt != NULL) {
5698                 shrink_dcache_sb(osd_sb(o));
5699                 osd_sync(env, &o->od_dt_dev);
5700
5701                 mntput(o->od_mnt);
5702                 o->od_mnt = NULL;
5703         }
5704
5705         EXIT;
5706 }
5707
5708 static int osd_mount(const struct lu_env *env,
5709                      struct osd_device *o, struct lustre_cfg *cfg)
5710 {
5711         const char              *name  = lustre_cfg_string(cfg, 0);
5712         const char              *dev  = lustre_cfg_string(cfg, 1);
5713         const char              *opts;
5714         unsigned long            page, s_flags, lmd_flags = 0;
5715         struct page             *__page;
5716         struct file_system_type *type;
5717         char                    *options = NULL;
5718         char                    *str;
5719         struct osd_thread_info  *info = osd_oti_get(env);
5720         struct lu_fid           *fid = &info->oti_fid;
5721         struct inode            *inode;
5722         int                      rc = 0, force_over_128tb = 0;
5723         ENTRY;
5724
5725         if (o->od_mnt != NULL)
5726                 RETURN(0);
5727
5728         if (strlen(dev) >= sizeof(o->od_mntdev))
5729                 RETURN(-E2BIG);
5730         strcpy(o->od_mntdev, dev);
5731
5732         str = lustre_cfg_string(cfg, 2);
5733         s_flags = simple_strtoul(str, NULL, 0);
5734         str = strstr(str, ":");
5735         if (str)
5736                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5737         opts = lustre_cfg_string(cfg, 3);
5738 #ifdef __BIG_ENDIAN
5739         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5740                 CERROR("%s: device %s extents feature is not guaranteed to "
5741                        "work on big-endian systems. Use \"bigendian_extents\" "
5742                        "mount option to override.\n", name, dev);
5743                 RETURN(-EINVAL);
5744         }
5745 #endif
5746         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5747                 force_over_128tb = 1;
5748
5749         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5750         if (__page == NULL)
5751                 GOTO(out, rc = -ENOMEM);
5752         page = (unsigned long)page_address(__page);
5753         options = (char *)page;
5754         *options = '\0';
5755         if (opts != NULL) {
5756                 /* strip out the options for back compatiblity */
5757                 static char *sout[] = {
5758                         "mballoc",
5759                         "iopen",
5760                         "noiopen",
5761                         "iopen_nopriv",
5762                         "extents",
5763                         "noextents",
5764                         /* strip out option we processed in osd */
5765                         "bigendian_extents",
5766                         "force_over_128tb",
5767                         NULL
5768                 };
5769                 strcat(options, opts);
5770                 for (rc = 0, str = options; sout[rc]; ) {
5771                         char *op = strstr(str, sout[rc]);
5772                         if (op == NULL) {
5773                                 rc++;
5774                                 str = options;
5775                                 continue;
5776                         }
5777                         if (op == options || *(op - 1) == ',') {
5778                                 str = op + strlen(sout[rc]);
5779                                 if (*str == ',' || *str == '\0') {
5780                                         *str == ',' ? str++ : str;
5781                                         memmove(op, str, strlen(str) + 1);
5782                                 }
5783                         }
5784                         for (str = op; *str != ',' && *str != '\0'; str++)
5785                                 ;
5786                 }
5787         } else {
5788                 strncat(options, "user_xattr,acl", 14);
5789         }
5790
5791         /* Glom up mount options */
5792         if (*options != '\0')
5793                 strcat(options, ",");
5794         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5795
5796         type = get_fs_type("ldiskfs");
5797         if (!type) {
5798                 CERROR("%s: cannot find ldiskfs module\n", name);
5799                 GOTO(out, rc = -ENODEV);
5800         }
5801
5802         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5803         module_put(type->owner);
5804
5805         if (IS_ERR(o->od_mnt)) {
5806                 rc = PTR_ERR(o->od_mnt);
5807                 o->od_mnt = NULL;
5808                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5809                 GOTO(out, rc);
5810         }
5811
5812         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
5813             force_over_128tb == 0) {
5814                 CERROR("%s: device %s LDISKFS does not support filesystems "
5815                        "greater than 128TB and can cause data corruption. "
5816                        "Use \"force_over_128tb\" mount option to override.\n",
5817                        name, dev);
5818                 GOTO(out, rc = -EINVAL);
5819         }
5820
5821 #ifdef HAVE_DEV_SET_RDONLY
5822         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5823                 CERROR("%s: underlying device %s is marked as read-only. "
5824                        "Setup failed\n", name, dev);
5825                 GOTO(out_mnt, rc = -EROFS);
5826         }
5827 #endif
5828
5829         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5830                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5831                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5832                 GOTO(out_mnt, rc = -EINVAL);
5833         }
5834
5835 #ifdef LDISKFS_MOUNT_DIRDATA
5836         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5837                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5838                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5839         else if (!o->od_is_ost)
5840                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5841                       "enabling the dirdata feature. If you do not want to "
5842                       "downgrade to Lustre-1.x again, you can enable it via "
5843                       "'tune2fs -O dirdata device'\n", name, dev);
5844 #endif
5845         inode = osd_sb(o)->s_root->d_inode;
5846         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5847         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5848         if (rc != 0) {
5849                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5850                 GOTO(out_mnt, rc);
5851         }
5852
5853         if (lmd_flags & LMD_FLG_NOSCRUB)
5854                 o->od_noscrub = 1;
5855
5856         GOTO(out, rc = 0);
5857
5858 out_mnt:
5859         mntput(o->od_mnt);
5860         o->od_mnt = NULL;
5861
5862 out:
5863         if (__page)
5864                 OBD_PAGE_FREE(__page);
5865
5866         return rc;
5867 }
5868
5869 static struct lu_device *osd_device_fini(const struct lu_env *env,
5870                                          struct lu_device *d)
5871 {
5872         struct osd_device *o = osd_dev(d);
5873         ENTRY;
5874
5875         osd_shutdown(env, o);
5876         osd_procfs_fini(o);
5877         osd_scrub_cleanup(env, o);
5878         osd_obj_map_fini(o);
5879         osd_umount(env, o);
5880
5881         RETURN(NULL);
5882 }
5883
5884 static int osd_device_init0(const struct lu_env *env,
5885                             struct osd_device *o,
5886                             struct lustre_cfg *cfg)
5887 {
5888         struct lu_device        *l = osd2lu_dev(o);
5889         struct osd_thread_info *info;
5890         int                     rc;
5891         int                     cplen = 0;
5892
5893         /* if the module was re-loaded, env can loose its keys */
5894         rc = lu_env_refill((struct lu_env *) env);
5895         if (rc)
5896                 GOTO(out, rc);
5897         info = osd_oti_get(env);
5898         LASSERT(info);
5899
5900         l->ld_ops = &osd_lu_ops;
5901         o->od_dt_dev.dd_ops = &osd_dt_ops;
5902
5903         spin_lock_init(&o->od_osfs_lock);
5904         mutex_init(&o->od_otable_mutex);
5905
5906         o->od_read_cache = 1;
5907         o->od_writethrough_cache = 1;
5908         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5909
5910         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5911                         sizeof(o->od_svname));
5912         if (cplen >= sizeof(o->od_svname)) {
5913                 rc = -E2BIG;
5914                 GOTO(out, rc);
5915         }
5916
5917         if (server_name_is_ost(o->od_svname))
5918                 o->od_is_ost = 1;
5919
5920         o->od_full_scrub_ratio = OFSR_DEFAULT;
5921         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
5922         rc = osd_mount(env, o, cfg);
5923         if (rc != 0)
5924                 GOTO(out, rc);
5925
5926         rc = osd_obj_map_init(env, o);
5927         if (rc != 0)
5928                 GOTO(out_mnt, rc);
5929
5930         rc = lu_site_init(&o->od_site, l);
5931         if (rc != 0)
5932                 GOTO(out_compat, rc);
5933         o->od_site.ls_bottom_dev = l;
5934
5935         rc = lu_site_init_finish(&o->od_site);
5936         if (rc != 0)
5937                 GOTO(out_site, rc);
5938
5939         INIT_LIST_HEAD(&o->od_ios_list);
5940         /* setup scrub, including OI files initialization */
5941         rc = osd_scrub_setup(env, o);
5942         if (rc < 0)
5943                 GOTO(out_site, rc);
5944
5945         rc = osd_procfs_init(o, o->od_svname);
5946         if (rc != 0) {
5947                 CERROR("%s: can't initialize procfs: rc = %d\n",
5948                        o->od_svname, rc);
5949                 GOTO(out_scrub, rc);
5950         }
5951
5952         LASSERT(l->ld_site->ls_linkage.next != NULL);
5953         LASSERT(l->ld_site->ls_linkage.prev != NULL);
5954
5955         /* initialize quota slave instance */
5956         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5957                                      o->od_proc_entry);
5958         if (IS_ERR(o->od_quota_slave)) {
5959                 rc = PTR_ERR(o->od_quota_slave);
5960                 o->od_quota_slave = NULL;
5961                 GOTO(out_procfs, rc);
5962         }
5963
5964         RETURN(0);
5965
5966 out_procfs:
5967         osd_procfs_fini(o);
5968 out_scrub:
5969         osd_scrub_cleanup(env, o);
5970 out_site:
5971         lu_site_fini(&o->od_site);
5972 out_compat:
5973         osd_obj_map_fini(o);
5974 out_mnt:
5975         osd_umount(env, o);
5976 out:
5977         return rc;
5978 }
5979
5980 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5981                                           struct lu_device_type *t,
5982                                           struct lustre_cfg *cfg)
5983 {
5984         struct osd_device *o;
5985         int                rc;
5986
5987         OBD_ALLOC_PTR(o);
5988         if (o == NULL)
5989                 return ERR_PTR(-ENOMEM);
5990
5991         rc = dt_device_init(&o->od_dt_dev, t);
5992         if (rc == 0) {
5993                 /* Because the ctx might be revived in dt_device_init,
5994                  * refill the env here */
5995                 lu_env_refill((struct lu_env *)env);
5996                 rc = osd_device_init0(env, o, cfg);
5997                 if (rc)
5998                         dt_device_fini(&o->od_dt_dev);
5999         }
6000
6001         if (unlikely(rc != 0))
6002                 OBD_FREE_PTR(o);
6003
6004         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6005 }
6006
6007 static struct lu_device *osd_device_free(const struct lu_env *env,
6008                                          struct lu_device *d)
6009 {
6010         struct osd_device *o = osd_dev(d);
6011         ENTRY;
6012
6013         /* XXX: make osd top device in order to release reference */
6014         d->ld_site->ls_top_dev = d;
6015         lu_site_purge(env, d->ld_site, -1);
6016         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6017                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6018                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6019         }
6020         lu_site_fini(&o->od_site);
6021         dt_device_fini(&o->od_dt_dev);
6022         OBD_FREE_PTR(o);
6023         RETURN(NULL);
6024 }
6025
6026 static int osd_process_config(const struct lu_env *env,
6027                               struct lu_device *d, struct lustre_cfg *cfg)
6028 {
6029         struct osd_device               *o = osd_dev(d);
6030         int                             rc;
6031         ENTRY;
6032
6033         switch (cfg->lcfg_command) {
6034         case LCFG_SETUP:
6035                 rc = osd_mount(env, o, cfg);
6036                 break;
6037         case LCFG_CLEANUP:
6038                 lu_dev_del_linkage(d->ld_site, d);
6039                 rc = osd_shutdown(env, o);
6040                 break;
6041         case LCFG_PARAM:
6042                 LASSERT(&o->od_dt_dev);
6043                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6044                                               cfg, &o->od_dt_dev);
6045                 if (rc > 0 || rc == -ENOSYS)
6046                         rc = class_process_proc_param(PARAM_OST,
6047                                                       lprocfs_osd_obd_vars,
6048                                                       cfg, &o->od_dt_dev);
6049                 break;
6050         default:
6051                 rc = -ENOSYS;
6052         }
6053
6054         RETURN(rc);
6055 }
6056
6057 static int osd_recovery_complete(const struct lu_env *env,
6058                                  struct lu_device *d)
6059 {
6060         struct osd_device       *osd = osd_dev(d);
6061         int                      rc = 0;
6062         ENTRY;
6063
6064         if (osd->od_quota_slave == NULL)
6065                 RETURN(0);
6066
6067         /* start qsd instance on recovery completion, this notifies the quota
6068          * slave code that we are about to process new requests now */
6069         rc = qsd_start(env, osd->od_quota_slave);
6070         RETURN(rc);
6071 }
6072
6073 /*
6074  * we use exports to track all osd users
6075  */
6076 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6077                            struct obd_device *obd, struct obd_uuid *cluuid,
6078                            struct obd_connect_data *data, void *localdata)
6079 {
6080         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6081         struct lustre_handle  conn;
6082         int                   rc;
6083         ENTRY;
6084
6085         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6086
6087         rc = class_connect(&conn, obd, cluuid);
6088         if (rc)
6089                 RETURN(rc);
6090
6091         *exp = class_conn2export(&conn);
6092
6093         spin_lock(&osd->od_osfs_lock);
6094         osd->od_connects++;
6095         spin_unlock(&osd->od_osfs_lock);
6096
6097         RETURN(0);
6098 }
6099
6100 /*
6101  * once last export (we don't count self-export) disappeared
6102  * osd can be released
6103  */
6104 static int osd_obd_disconnect(struct obd_export *exp)
6105 {
6106         struct obd_device *obd = exp->exp_obd;
6107         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6108         int                rc, release = 0;
6109         ENTRY;
6110
6111         /* Only disconnect the underlying layers on the final disconnect. */
6112         spin_lock(&osd->od_osfs_lock);
6113         osd->od_connects--;
6114         if (osd->od_connects == 0)
6115                 release = 1;
6116         spin_unlock(&osd->od_osfs_lock);
6117
6118         rc = class_disconnect(exp); /* bz 9811 */
6119
6120         if (rc == 0 && release)
6121                 class_manual_cleanup(obd);
6122         RETURN(rc);
6123 }
6124
6125 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6126                        struct lu_device *dev)
6127 {
6128         struct osd_device       *osd    = osd_dev(dev);
6129         struct lr_server_data   *lsd    =
6130                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6131         int                      result = 0;
6132         ENTRY;
6133
6134         if (osd->od_quota_slave != NULL) {
6135                 /* set up quota slave objects */
6136                 result = qsd_prepare(env, osd->od_quota_slave);
6137                 if (result != 0)
6138                         RETURN(result);
6139         }
6140
6141         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6142 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6143                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6144                         osd->od_index_in_idif = 1;
6145                 } else {
6146                         osd->od_index_in_idif = 0;
6147                         result = osd_register_proc_index_in_idif(osd);
6148                         if (result != 0)
6149                                 RETURN(result);
6150                 }
6151 #else
6152                 osd->od_index_in_idif = 1;
6153 #endif
6154         }
6155
6156         result = osd_fid_init(env, osd);
6157
6158         RETURN(result);
6159 }
6160
6161 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6162                          struct lu_fid *fid, struct md_op_data *op_data)
6163 {
6164         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6165
6166         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6167 }
6168
6169 static const struct lu_object_operations osd_lu_obj_ops = {
6170         .loo_object_init      = osd_object_init,
6171         .loo_object_delete    = osd_object_delete,
6172         .loo_object_release   = osd_object_release,
6173         .loo_object_free      = osd_object_free,
6174         .loo_object_print     = osd_object_print,
6175         .loo_object_invariant = osd_object_invariant
6176 };
6177
6178 const struct lu_device_operations osd_lu_ops = {
6179         .ldo_object_alloc      = osd_object_alloc,
6180         .ldo_process_config    = osd_process_config,
6181         .ldo_recovery_complete = osd_recovery_complete,
6182         .ldo_prepare           = osd_prepare,
6183 };
6184
6185 static const struct lu_device_type_operations osd_device_type_ops = {
6186         .ldto_init = osd_type_init,
6187         .ldto_fini = osd_type_fini,
6188
6189         .ldto_start = osd_type_start,
6190         .ldto_stop  = osd_type_stop,
6191
6192         .ldto_device_alloc = osd_device_alloc,
6193         .ldto_device_free  = osd_device_free,
6194
6195         .ldto_device_init    = osd_device_init,
6196         .ldto_device_fini    = osd_device_fini
6197 };
6198
6199 static struct lu_device_type osd_device_type = {
6200         .ldt_tags     = LU_DEVICE_DT,
6201         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6202         .ldt_ops      = &osd_device_type_ops,
6203         .ldt_ctx_tags = LCT_LOCAL,
6204 };
6205
6206 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6207 {
6208         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6209         struct super_block *sb = osd_sb(osd);
6210
6211         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6212 }
6213
6214 /*
6215  * lprocfs legacy support.
6216  */
6217 static struct obd_ops osd_obd_device_ops = {
6218         .o_owner = THIS_MODULE,
6219         .o_connect      = osd_obd_connect,
6220         .o_disconnect   = osd_obd_disconnect,
6221         .o_fid_alloc    = osd_fid_alloc,
6222         .o_health_check = osd_health_check,
6223 };
6224
6225 static int __init osd_mod_init(void)
6226 {
6227         int rc;
6228
6229 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6230         /* please, try to keep osd_thread_info smaller than a page */
6231         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6232 #endif
6233
6234         osd_oi_mod_init();
6235
6236         rc = lu_kmem_init(ldiskfs_caches);
6237         if (rc)
6238                 return rc;
6239
6240         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6241                                  lprocfs_osd_module_vars,
6242                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6243         if (rc)
6244                 lu_kmem_fini(ldiskfs_caches);
6245         return rc;
6246 }
6247
6248 static void __exit osd_mod_exit(void)
6249 {
6250         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6251         lu_kmem_fini(ldiskfs_caches);
6252 }
6253
6254 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6255 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6256 MODULE_VERSION(LUSTRE_VERSION_STRING);
6257 MODULE_LICENSE("GPL");
6258
6259 module_init(osd_mod_init);
6260 module_exit(osd_mod_exit);