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 /**
 305  * \retval +v: new filter_fid, does not contain self-fid
 306  * \retval 0:  filter_fid_old, contains self-fid
 307  * \retval -v: other failure cases
 308  */
 309 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 310                  struct dentry *dentry, struct lu_fid *fid)
 311 {
 312         struct filter_fid_old   *ff     = &info->oti_ff;
 313         struct ost_id           *ostid  = &info->oti_ostid;
 314         int                      rc;
 315
 316         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 317         if (rc == sizeof(*ff)) {
 318                 rc = 0;
 319                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 320                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 321                 /* XXX: should use real OST index in the future. LU-3569 */
 322                 ostid_to_fid(fid, ostid, 0);
 323         } else if (rc == sizeof(struct filter_fid)) {
 324                 rc = 1;
 325         } else if (rc >= 0) {
 326                 rc = -EINVAL;
 327         }
 328
 329         return rc;
 330 }
 331
 332 static int osd_lma_self_repair(struct osd_thread_info *info,
 333                                struct osd_device *osd, struct inode *inode,
 334                                const struct lu_fid *fid, __u32 compat)
 335 {
 336         handle_t *jh;
 337         int       rc;
 338
 339         LASSERT(current->journal_info == NULL);
 340
 341         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 342                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 343         if (IS_ERR(jh)) {
 344                 rc = PTR_ERR(jh);
 345                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 346                       osd_name(osd), rc);
 347                 return rc;
 348         }
 349
 350         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 351         if (rc != 0)
 352                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 353                       osd_name(osd), rc);
 354         ldiskfs_journal_stop(jh);
 355         return rc;
 356 }
 357
 358 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 359 {
 360         struct osd_thread_info  *info   = osd_oti_get(env);
 361         struct osd_device       *osd    = osd_obj2dev(obj);
 362         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 363         struct inode            *inode  = obj->oo_inode;
 364         struct dentry           *dentry = &info->oti_obj_dentry;
 365         struct lu_fid           *fid    = NULL;
 366         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 367         int                      rc;
 368         ENTRY;
 369
 370         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 371         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 372                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 373         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 374                 fid = &lma->lma_self_fid;
 375                 rc = osd_get_idif(info, inode, dentry, fid);
 376                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 377                         /* For the given OST-object, if it has neither LMA nor
 378                          * FID in XATTR_NAME_FID, then the given FID (which is
 379                          * contained in the @obj, from client RPC for locating
 380                          * the OST-object) is trusted. We use it to generate
 381                          * the LMA. */
 382                         osd_lma_self_repair(info, osd, inode, rfid,
 383                                             LMAC_FID_ON_OST);
 384                         RETURN(0);
 385                 }
 386         }
 387
 388         if (unlikely(rc == -ENODATA))
 389                 RETURN(0);
 390
 391         if (rc < 0)
 392                 RETURN(rc);
 393
 394         if (rc > 0) {
 395                 rc = 0;
 396                 lustre_lma_swab(lma);
 397                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 398                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 399                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 400                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 401                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 402                               PFID(rfid), inode->i_ino);
 403                         rc = -EOPNOTSUPP;
 404                 } else {
 405                         fid = &lma->lma_self_fid;
 406                 }
 407         }
 408
 409         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 410                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 411                         struct ost_id   *oi   = &info->oti_ostid;
 412                         struct lu_fid   *fid1 = &info->oti_fid3;
 413                         __u32            idx  = fid_idif_ost_idx(rfid);
 414
 415                         /* For old IDIF, the OST index is not part of the IDIF,
 416                          * Means that different OSTs may have the same IDIFs.
 417                          * Under such case, we need to make some compatible
 418                          * check to make sure to trigger OI scrub properly. */
 419                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 420                                 /* Given @rfid is new, LMA is old. */
 421                                 fid_to_ostid(fid, oi);
 422                                 ostid_to_fid(fid1, oi, idx);
 423                                 if (lu_fid_eq(fid1, rfid)) {
 424                                         if (osd->od_index_in_idif)
 425                                                 osd_lma_self_repair(info, osd,
 426                                                         inode, rfid,
 427                                                         LMAC_FID_ON_OST);
 428                                         RETURN(0);
 429                                 }
 430                         }
 431                 }
 432
 433                 rc = -EREMCHG;
 434         }
 435
 436         RETURN(rc);
 437 }
 438
 439 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 440                           const struct lu_fid *fid,
 441                           const struct lu_object_conf *conf)
 442 {
 443         struct osd_thread_info *info;
 444         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 445         struct osd_device      *dev;
 446         struct osd_idmap_cache *oic;
 447         struct osd_inode_id    *id;
 448         struct inode           *inode;
 449         struct osd_scrub       *scrub;
 450         struct scrub_file      *sf;
 451         int                     result;
 452         int                     saved  = 0;
 453         bool                    in_oi  = false;
 454         bool                    in_cache = false;
 455         bool                    triggered = false;
 456         ENTRY;
 457
 458         LINVRNT(osd_invariant(obj));
 459         LASSERT(obj->oo_inode == NULL);
 460         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 461
 462         dev = osd_dev(ldev);
 463         scrub = &dev->od_scrub;
 464         sf = &scrub->os_file;
 465         info = osd_oti_get(env);
 466         LASSERT(info);
 467         oic = &info->oti_cache;
 468
 469         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 470                 RETURN(-ENOENT);
 471
 472         /* For the object is created as locking anchor, or for the object to
 473          * be created on disk. No need to osd_oi_lookup() at here because FID
 474          * shouldn't never be re-used, if it's really a duplicate FID from
 475          * unexpected reason, we should be able to detect it later by calling
 476          * do_create->osd_oi_insert(). */
 477         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 478                 GOTO(out, result = 0);
 479
 480         /* Search order: 1. per-thread cache. */
 481         if (lu_fid_eq(fid, &oic->oic_fid) &&
 482             likely(oic->oic_dev == dev)) {
 483                 id = &oic->oic_lid;
 484                 in_cache = true;
 485                 goto iget;
 486         }
 487
 488         id = &info->oti_id;
 489         if (!list_empty(&scrub->os_inconsistent_items)) {
 490                 /* Search order: 2. OI scrub pending list. */
 491                 result = osd_oii_lookup(dev, fid, id);
 492                 if (result == 0)
 493                         goto iget;
 494         }
 495
 496         /* Search order: 3. OI files. */
 497         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 498         if (result == -ENOENT) {
 499                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 500                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 501                                       sf->sf_oi_bitmap))
 502                         GOTO(out, result = 0);
 503
 504                 goto trigger;
 505         }
 506
 507         if (result != 0)
 508                 GOTO(out, result);
 509
 510         in_oi = true;
 511
 512 iget:
 513         inode = osd_iget(info, dev, id);
 514         if (IS_ERR(inode)) {
 515                 result = PTR_ERR(inode);
 516                 if (result != -ENOENT && result != -ESTALE)
 517                         GOTO(out, result);
 518
 519                 if (in_cache)
 520                         fid_zero(&oic->oic_fid);
 521
 522                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 523                 if (result != 0)
 524                         GOTO(out, result = (result == -ENOENT ? 0 : result));
 525
 526                 /* The OI mapping is there, but the inode is NOT there.
 527                  * Two possible cases for that:
 528                  *
 529                  * 1) Backup/restore caused the OI invalid.
 530                  * 2) Someone unlinked the object but NOT removed
 531                  *    the OI mapping, such as mount target device
 532                  *    as ldiskfs, and modify something directly.
 533                  *
 534                  * Generally, when the device is mounted, it will
 535                  * auto check whether the system is restored from
 536                  * file-level backup or not. We trust such detect
 537                  * to distinguish the 1st case from the 2nd case. */
 538                 if (!(scrub->os_file.sf_flags & SF_INCONSISTENT))
 539                         GOTO(out, result = 0);
 540
 541 trigger:
 542                 if (unlikely(triggered))
 543                         GOTO(out, result = saved);
 544
 545                 triggered = true;
 546                 if (thread_is_running(&scrub->os_thread)) {
 547                         result = -EINPROGRESS;
 548                 } else if (!dev->od_noscrub) {
 549                         /* Since we do not know the right OI mapping, we have
 550                          * to trigger OI scrub to scan the whole device. */
 551                         result = osd_scrub_start(dev, SS_AUTO_FULL |
 552                                 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 553                         CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
 554                                "scrub by RPC for "DFID", rc = %d [1]\n",
 555                                osd_name(dev), PFID(fid), result);
 556                         if (result == 0 || result == -EALREADY)
 557                                 result = -EINPROGRESS;
 558                         else
 559                                 result = -EREMCHG;
 560                 }
 561
 562                 /* We still have chance to get the valid inode: for the
 563                  * object which is referenced by remote name entry, the
 564                  * object on the local MDT will be linked under the dir
 565                  * of "/REMOTE_PARENT_DIR" with its FID string as name.
 566                  *
 567                  * We do not know whether the object for the given FID
 568                  * is referenced by some remote name entry or not, and
 569                  * especially for DNE II, a multiple-linked object may
 570                  * have many name entries reside on many MDTs.
 571                  *
 572                  * To simplify the operation, OSD will not distinguish
 573                  * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 574                  * only happened for the RPC from other MDT during the
 575                  * OI scrub, or for the client side RPC with FID only,
 576                  * such as FID to path, or from old connected client. */
 577                 saved = result;
 578                 result = osd_lookup_in_remote_parent(info, dev, fid, id);
 579                 if (result == 0) {
 580                         in_oi = false;
 581                         goto iget;
 582                 }
 583
 584                 GOTO(out, result = saved);
 585         }
 586
 587         obj->oo_inode = inode;
 588         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 589
 590         result = osd_check_lma(env, obj);
 591         if (result != 0) {
 592                 iput(inode);
 593                 obj->oo_inode = NULL;
 594
 595                 if (result != -EREMCHG)
 596                         GOTO(out, result);
 597
 598                 if (in_cache)
 599                         fid_zero(&oic->oic_fid);
 600
 601                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 602                 if (result == 0)
 603                         goto trigger;
 604
 605                 if (result != -ENOENT)
 606                         GOTO(out, result);
 607
 608                 if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) &&
 609                     ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
 610                                      sf->sf_oi_bitmap))
 611                         goto trigger;
 612
 613                 GOTO(out, result = 0);
 614         }
 615
 616         obj->oo_compat_dot_created = 1;
 617         obj->oo_compat_dotdot_created = 1;
 618
 619         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 620                 GOTO(out, result = 0);
 621
 622         LASSERT(obj->oo_hl_head == NULL);
 623         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 624         if (obj->oo_hl_head == NULL) {
 625                 obj->oo_inode = NULL;
 626                 iput(inode);
 627                 GOTO(out, result = -ENOMEM);
 628         }
 629         GOTO(out, result = 0);
 630
 631 out:
 632         LINVRNT(osd_invariant(obj));
 633         return result;
 634 }
 635
 636 /*
 637  * Concurrency: shouldn't matter.
 638  */
 639 static void osd_object_init0(struct osd_object *obj)
 640 {
 641         LASSERT(obj->oo_inode != NULL);
 642         obj->oo_dt.do_body_ops = &osd_body_ops;
 643         obj->oo_dt.do_lu.lo_header->loh_attr |=
 644                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 645 }
 646
 647 /*
 648  * Concurrency: no concurrent access is possible that early in object
 649  * life-cycle.
 650  */
 651 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 652                            const struct lu_object_conf *conf)
 653 {
 654         struct osd_object *obj = osd_obj(l);
 655         int result;
 656
 657         LINVRNT(osd_invariant(obj));
 658
 659         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 660                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 661                 l->lo_header->loh_attr |= LOHA_EXISTS;
 662                 return 0;
 663         }
 664
 665         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 666         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 667         if (result == 0 && obj->oo_inode != NULL)
 668                 osd_object_init0(obj);
 669
 670         LINVRNT(osd_invariant(obj));
 671         return result;
 672 }
 673
 674 /*
 675  * Concurrency: no concurrent access is possible that late in object
 676  * life-cycle.
 677  */
 678 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 679 {
 680         struct osd_object *obj = osd_obj(l);
 681
 682         LINVRNT(osd_invariant(obj));
 683
 684         dt_object_fini(&obj->oo_dt);
 685         if (obj->oo_hl_head != NULL)
 686                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 687         OBD_FREE_PTR(obj);
 688 }
 689
 690 /*
 691  * Concurrency: no concurrent access is possible that late in object
 692  * life-cycle.
 693  */
 694 static void osd_index_fini(struct osd_object *o)
 695 {
 696         struct iam_container *bag;
 697
 698         if (o->oo_dir != NULL) {
 699                 bag = &o->oo_dir->od_container;
 700                 if (o->oo_inode != NULL) {
 701                         if (bag->ic_object == o->oo_inode)
 702                                 iam_container_fini(bag);
 703                 }
 704                 OBD_FREE_PTR(o->oo_dir);
 705                 o->oo_dir = NULL;
 706         }
 707 }
 708
 709 /*
 710  * Concurrency: no concurrent access is possible that late in object
 711  * life-cycle (for all existing callers, that is. New callers have to provide
 712  * their own locking.)
 713  */
 714 static int osd_inode_unlinked(const struct inode *inode)
 715 {
 716         return inode->i_nlink == 0;
 717 }
 718
 719 enum {
 720         OSD_TXN_OI_DELETE_CREDITS    = 20,
 721         OSD_TXN_INODE_DELETE_CREDITS = 20
 722 };
 723
 724 /*
 725  * Journal
 726  */
 727
 728 #if OSD_THANDLE_STATS
 729 /**
 730  * Set time when the handle is allocated
 731  */
 732 static void osd_th_alloced(struct osd_thandle *oth)
 733 {
 734         oth->oth_alloced = cfs_time_current();
 735 }
 736
 737 /**
 738  * Set time when the handle started
 739  */
 740 static void osd_th_started(struct osd_thandle *oth)
 741 {
 742         oth->oth_started = cfs_time_current();
 743 }
 744
 745 /**
 746  * Helper function to convert time interval to microseconds packed in
 747  * long int.
 748  */
 749 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 750 {
 751         struct timeval val;
 752
 753         cfs_duration_usec(cfs_time_sub(end, start), &val);
 754         return val.tv_sec * 1000000 + val.tv_usec;
 755 }
 756
 757 /**
 758  * Check whether the we deal with this handle for too long.
 759  */
 760 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 761                                 cfs_time_t alloced, cfs_time_t started,
 762                                 cfs_time_t closed)
 763 {
 764         cfs_time_t now = cfs_time_current();
 765
 766         LASSERT(dev != NULL);
 767
 768         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 769                             interval_to_usec(alloced, started));
 770         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 771                             interval_to_usec(started, closed));
 772         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 773                             interval_to_usec(closed, now));
 774
 775         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 776                 CWARN("transaction handle %p was open for too long: "
 777                       "now "CFS_TIME_T" ,"
 778                       "alloced "CFS_TIME_T" ,"
 779                       "started "CFS_TIME_T" ,"
 780                       "closed "CFS_TIME_T"\n",
 781                       oth, now, alloced, started, closed);
 782                 libcfs_debug_dumpstack(NULL);
 783         }
 784 }
 785
 786 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 787 {                                                                       \
 788         cfs_time_t __closed = cfs_time_current();                       \
 789         cfs_time_t __alloced = oth->oth_alloced;                        \
 790         cfs_time_t __started = oth->oth_started;                        \
 791                                                                         \
 792         expr;                                                           \
 793         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 794 }
 795
 796 #else /* OSD_THANDLE_STATS */
 797
 798 #define osd_th_alloced(h)                  do {} while(0)
 799 #define osd_th_started(h)                  do {} while(0)
 800 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 801
 802 #endif /* OSD_THANDLE_STATS */
 803
 804 /*
 805  * Concurrency: doesn't access mutable data.
 806  */
 807 static int osd_param_is_not_sane(const struct osd_device *dev,
 808                                  const struct thandle *th)
 809 {
 810         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 811
 812         return oh->ot_credits > osd_transaction_size(dev);
 813 }
 814
 815 /*
 816  * Concurrency: shouldn't matter.
 817  */
 818 static void osd_trans_commit_cb(struct super_block *sb,
 819                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 820 {
 821         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 822         struct thandle     *th  = &oh->ot_super;
 823         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 824         struct dt_txn_commit_cb *dcb, *tmp;
 825
 826         LASSERT(oh->ot_handle == NULL);
 827
 828         if (error)
 829                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 830
 831         dt_txn_hook_commit(th);
 832
 833         /* call per-transaction callbacks if any */
 834         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
 835                                  dcb_linkage) {
 836                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 837                          "commit callback entry: magic=%x name='%s'\n",
 838                          dcb->dcb_magic, dcb->dcb_name);
 839                 list_del_init(&dcb->dcb_linkage);
 840                 dcb->dcb_func(NULL, th, dcb, error);
 841         }
 842
 843         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 844         lu_device_put(lud);
 845         th->th_dev = NULL;
 846
 847         lu_context_exit(&th->th_ctx);
 848         lu_context_fini(&th->th_ctx);
 849         OBD_FREE_PTR(oh);
 850 }
 851
 852 static struct thandle *osd_trans_create(const struct lu_env *env,
 853                                         struct dt_device *d)
 854 {
 855         struct osd_thread_info  *oti = osd_oti_get(env);
 856         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 857         struct osd_thandle      *oh;
 858         struct thandle          *th;
 859         ENTRY;
 860
 861         /* on pending IO in this thread should left from prev. request */
 862         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 863
 864         th = ERR_PTR(-ENOMEM);
 865         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 866         if (oh != NULL) {
 867                 oh->ot_quota_trans = &oti->oti_quota_trans;
 868                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 869                 th = &oh->ot_super;
 870                 th->th_dev = d;
 871                 th->th_result = 0;
 872                 th->th_tags = LCT_TX_HANDLE;
 873                 oh->ot_credits = 0;
 874                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
 875                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
 876                 osd_th_alloced(oh);
 877
 878                 memset(oti->oti_declare_ops, 0,
 879                        sizeof(oti->oti_declare_ops));
 880                 memset(oti->oti_declare_ops_cred, 0,
 881                        sizeof(oti->oti_declare_ops_cred));
 882                 memset(oti->oti_declare_ops_used, 0,
 883                        sizeof(oti->oti_declare_ops_used));
 884         }
 885         RETURN(th);
 886 }
 887
 888 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
 889 {
 890         struct osd_thread_info  *oti = osd_oti_get(env);
 891         struct osd_thandle      *oh;
 892
 893         oh = container_of0(th, struct osd_thandle, ot_super);
 894         LASSERT(oh != NULL);
 895
 896         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
 897               oti->oti_declare_ops[OSD_OT_CREATE],
 898               oti->oti_declare_ops_cred[OSD_OT_CREATE],
 899               oti->oti_declare_ops_used[OSD_OT_CREATE],
 900               oti->oti_declare_ops[OSD_OT_DESTROY],
 901               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
 902               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
 903         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
 904               oti->oti_declare_ops[OSD_OT_ATTR_SET],
 905               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
 906               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
 907               oti->oti_declare_ops[OSD_OT_XATTR_SET],
 908               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
 909               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
 910         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
 911               oti->oti_declare_ops[OSD_OT_WRITE],
 912               oti->oti_declare_ops_cred[OSD_OT_WRITE],
 913               oti->oti_declare_ops_used[OSD_OT_WRITE],
 914               oti->oti_declare_ops[OSD_OT_PUNCH],
 915               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
 916               oti->oti_declare_ops_used[OSD_OT_PUNCH],
 917               oti->oti_declare_ops[OSD_OT_QUOTA],
 918               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
 919               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
 920         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
 921               oti->oti_declare_ops[OSD_OT_INSERT],
 922               oti->oti_declare_ops_cred[OSD_OT_INSERT],
 923               oti->oti_declare_ops_used[OSD_OT_INSERT],
 924               oti->oti_declare_ops[OSD_OT_DELETE],
 925               oti->oti_declare_ops_cred[OSD_OT_DELETE],
 926               oti->oti_declare_ops_used[OSD_OT_DELETE]);
 927         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
 928               oti->oti_declare_ops[OSD_OT_REF_ADD],
 929               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
 930               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
 931               oti->oti_declare_ops[OSD_OT_REF_DEL],
 932               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
 933               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
 934 }
 935
 936 /*
 937  * Concurrency: shouldn't matter.
 938  */
 939 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 940                            struct thandle *th)
 941 {
 942         struct osd_thread_info *oti = osd_oti_get(env);
 943         struct osd_device  *dev = osd_dt_dev(d);
 944         handle_t           *jh;
 945         struct osd_thandle *oh;
 946         int rc;
 947
 948         ENTRY;
 949
 950         LASSERT(current->journal_info == NULL);
 951
 952         oh = container_of0(th, struct osd_thandle, ot_super);
 953         LASSERT(oh != NULL);
 954         LASSERT(oh->ot_handle == NULL);
 955
 956         rc = dt_txn_hook_start(env, d, th);
 957         if (rc != 0)
 958                 GOTO(out, rc);
 959
 960         if (unlikely(osd_param_is_not_sane(dev, th))) {
 961                 static unsigned long last_printed;
 962                 static int last_credits;
 963
 964                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 965                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 966                       oh->ot_credits,
 967                       osd_journal(dev)->j_max_transaction_buffers);
 968
 969                 osd_trans_dump_creds(env, th);
 970
 971                 if (last_credits != oh->ot_credits &&
 972                     time_after(jiffies, last_printed +
 973                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
 974                         libcfs_debug_dumpstack(NULL);
 975                         last_credits = oh->ot_credits;
 976                         last_printed = jiffies;
 977                 }
 978                 /* XXX Limit the credits to 'max_transaction_buffers', and
 979                  *     let the underlying filesystem to catch the error if
 980                  *     we really need so many credits.
 981                  *
 982                  *     This should be removed when we can calculate the
 983                  *     credits precisely. */
 984                 oh->ot_credits = osd_transaction_size(dev);
 985         }
 986
 987         /*
 988          * XXX temporary stuff. Some abstraction layer should
 989          * be used.
 990          */
 991         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
 992         osd_th_started(oh);
 993         if (!IS_ERR(jh)) {
 994                 oh->ot_handle = jh;
 995                 LASSERT(oti->oti_txns == 0);
 996                 lu_context_init(&th->th_ctx, th->th_tags);
 997                 lu_context_enter(&th->th_ctx);
 998
 999                 lu_device_get(&d->dd_lu_dev);
1000                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1001                               "osd-tx", th);
1002                 oti->oti_txns++;
1003                 rc = 0;
1004         } else {
1005                 rc = PTR_ERR(jh);
1006         }
1007 out:
1008         RETURN(rc);
1009 }
1010
1011 static int osd_seq_exists(const struct lu_env *env,
1012                           struct osd_device *osd, u64 seq)
1013 {
1014         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1015         struct seq_server_site  *ss = osd_seq_site(osd);
1016         int                     rc;
1017         ENTRY;
1018
1019         LASSERT(ss != NULL);
1020         LASSERT(ss->ss_server_fld != NULL);
1021
1022         rc = osd_fld_lookup(env, osd, seq, range);
1023         if (rc != 0) {
1024                 if (rc != -ENOENT)
1025                         CERROR("%s: can't lookup FLD sequence "LPX64
1026                                ": rc = %d\n", osd_name(osd), seq, rc);
1027                 RETURN(0);
1028         }
1029
1030         RETURN(ss->ss_node_id == range->lsr_index);
1031 }
1032
1033 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1034 {
1035         struct dt_txn_commit_cb *dcb;
1036         struct dt_txn_commit_cb *tmp;
1037
1038         /* call per-transaction stop callbacks if any */
1039         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1040                                  dcb_linkage) {
1041                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1042                          "commit callback entry: magic=%x name='%s'\n",
1043                          dcb->dcb_magic, dcb->dcb_name);
1044                 list_del_init(&dcb->dcb_linkage);
1045                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1046         }
1047 }
1048
1049 /*
1050  * Concurrency: shouldn't matter.
1051  */
1052 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1053                           struct thandle *th)
1054 {
1055         int                     rc = 0;
1056         struct osd_thandle     *oh;
1057         struct osd_thread_info *oti = osd_oti_get(env);
1058         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1059         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1060         struct qsd_instance    *qsd = osd->od_quota_slave;
1061         struct lquota_trans    *qtrans;
1062         ENTRY;
1063
1064         oh = container_of0(th, struct osd_thandle, ot_super);
1065
1066         qtrans = oh->ot_quota_trans;
1067         oh->ot_quota_trans = NULL;
1068
1069         if (oh->ot_handle != NULL) {
1070                 handle_t *hdl = oh->ot_handle;
1071
1072                 /*
1073                  * add commit callback
1074                  * notice we don't do this in osd_trans_start()
1075                  * as underlying transaction can change during truncate
1076                  */
1077                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1078                                          &oh->ot_jcb);
1079
1080                 LASSERT(oti->oti_txns == 1);
1081                 oti->oti_txns--;
1082
1083                 rc = dt_txn_hook_stop(env, th);
1084                 if (rc != 0)
1085                         CERROR("%s: failed in transaction hook: rc = %d\n",
1086                                osd_name(osd), rc);
1087
1088                 osd_trans_stop_cb(oh, rc);
1089                 /* hook functions might modify th_sync */
1090                 hdl->h_sync = th->th_sync;
1091
1092                 oh->ot_handle = NULL;
1093                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1094                 if (rc != 0)
1095                         CERROR("%s: failed to stop transaction: rc = %d\n",
1096                                osd_name(osd), rc);
1097         } else {
1098                 osd_trans_stop_cb(oh, th->th_result);
1099                 OBD_FREE_PTR(oh);
1100         }
1101
1102         /* inform the quota slave device that the transaction is stopping */
1103         qsd_op_end(env, qsd, qtrans);
1104
1105         /* as we want IO to journal and data IO be concurrent, we don't block
1106          * awaiting data IO completion in osd_do_bio(), instead we wait here
1107          * once transaction is submitted to the journal. all reqular requests
1108          * don't do direct IO (except read/write), thus this wait_event becomes
1109          * no-op for them.
1110          *
1111          * IMPORTANT: we have to wait till any IO submited by the thread is
1112          * completed otherwise iobuf may be corrupted by different request
1113          */
1114         wait_event(iobuf->dr_wait,
1115                        atomic_read(&iobuf->dr_numreqs) == 0);
1116         osd_fini_iobuf(osd, iobuf);
1117         if (!rc)
1118                 rc = iobuf->dr_error;
1119
1120         RETURN(rc);
1121 }
1122
1123 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1124 {
1125         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1126                                                ot_super);
1127
1128         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1129         LASSERT(&dcb->dcb_func != NULL);
1130         if (dcb->dcb_flags & DCB_TRANS_STOP)
1131                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1132         else
1133                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1134
1135         return 0;
1136 }
1137
1138 /*
1139  * Called just before object is freed. Releases all resources except for
1140  * object itself (that is released by osd_object_free()).
1141  *
1142  * Concurrency: no concurrent access is possible that late in object
1143  * life-cycle.
1144  */
1145 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1146 {
1147         struct osd_object *obj   = osd_obj(l);
1148         struct inode      *inode = obj->oo_inode;
1149
1150         LINVRNT(osd_invariant(obj));
1151
1152         /*
1153          * If object is unlinked remove fid->ino mapping from object index.
1154          */
1155
1156         osd_index_fini(obj);
1157         if (inode != NULL) {
1158                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1159                 qid_t                    uid = i_uid_read(inode);
1160                 qid_t                    gid = i_gid_read(inode);
1161
1162                 iput(inode);
1163                 obj->oo_inode = NULL;
1164
1165                 if (qsd != NULL) {
1166                         struct osd_thread_info  *info = osd_oti_get(env);
1167                         struct lquota_id_info   *qi = &info->oti_qi;
1168
1169                         /* Release granted quota to master if necessary */
1170                         qi->lqi_id.qid_uid = uid;
1171                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1172
1173                         qi->lqi_id.qid_uid = gid;
1174                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1175                 }
1176         }
1177 }
1178
1179 /*
1180  * Concurrency: ->loo_object_release() is called under site spin-lock.
1181  */
1182 static void osd_object_release(const struct lu_env *env,
1183                                struct lu_object *l)
1184 {
1185 }
1186
1187 /*
1188  * Concurrency: shouldn't matter.
1189  */
1190 static int osd_object_print(const struct lu_env *env, void *cookie,
1191                             lu_printer_t p, const struct lu_object *l)
1192 {
1193         struct osd_object *o = osd_obj(l);
1194         struct iam_descr  *d;
1195
1196         if (o->oo_dir != NULL)
1197                 d = o->oo_dir->od_container.ic_descr;
1198         else
1199                 d = NULL;
1200         return (*p)(env, cookie,
1201                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1202                     o, o->oo_inode,
1203                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1204                     o->oo_inode ? o->oo_inode->i_generation : 0,
1205                     d ? d->id_ops->id_name : "plain");
1206 }
1207
1208 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1209
1210 /*
1211  * Concurrency: shouldn't matter.
1212  */
1213 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1214                struct obd_statfs *sfs)
1215 {
1216         struct osd_device  *osd = osd_dt_dev(d);
1217         struct super_block *sb = osd_sb(osd);
1218         struct kstatfs     *ksfs;
1219         int result = 0;
1220
1221         if (unlikely(osd->od_mnt == NULL))
1222                 return -EINPROGRESS;
1223
1224         /* osd_lproc.c call this without env, allocate ksfs for that case */
1225         if (unlikely(env == NULL)) {
1226                 OBD_ALLOC_PTR(ksfs);
1227                 if (ksfs == NULL)
1228                         return -ENOMEM;
1229         } else {
1230                 ksfs = &osd_oti_get(env)->oti_ksfs;
1231         }
1232
1233         spin_lock(&osd->od_osfs_lock);
1234         result = sb->s_op->statfs(sb->s_root, ksfs);
1235         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1236                 statfs_pack(sfs, ksfs);
1237                 if (unlikely(sb->s_flags & MS_RDONLY))
1238                         sfs->os_state = OS_STATE_READONLY;
1239                 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1240                                               LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1241                         sfs->os_maxbytes = sb->s_maxbytes;
1242                 else
1243                         sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1244         }
1245         spin_unlock(&osd->od_osfs_lock);
1246
1247         if (unlikely(env == NULL))
1248                 OBD_FREE_PTR(ksfs);
1249
1250         /* Reserve a small amount of space for local objects like last_rcvd,
1251          * llog, quota files, ... */
1252         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1253                 sfs->os_bavail = 0;
1254         } else {
1255                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1256                 /** Take out metadata overhead for indirect blocks */
1257                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1258         }
1259
1260         return result;
1261 }
1262
1263 /**
1264  * Estimate space needed for file creations. We assume the largest filename
1265  * which is 2^64 - 1, hence a filename of 20 chars.
1266  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1267  */
1268 #ifdef __LDISKFS_DIR_REC_LEN
1269 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1270 #else
1271 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1272 #endif
1273
1274 /*
1275  * Concurrency: doesn't access mutable data.
1276  */
1277 static void osd_conf_get(const struct lu_env *env,
1278                          const struct dt_device *dev,
1279                          struct dt_device_param *param)
1280 {
1281         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1282         int                ea_overhead;
1283
1284         /*
1285          * XXX should be taken from not-yet-existing fs abstraction layer.
1286          */
1287         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1288         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1289         param->ddp_block_shift  = sb->s_blocksize_bits;
1290         param->ddp_mount_type     = LDD_MT_LDISKFS;
1291         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1292                 param->ddp_maxbytes = sb->s_maxbytes;
1293         else
1294                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1295         /* Overhead estimate should be fairly accurate, so we really take a tiny
1296          * error margin which also avoids fragmenting the filesystem too much */
1297         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1298         /* inode are statically allocated, so per-inode space consumption
1299          * is the space consumed by the directory entry */
1300         param->ddp_inodespace     = PER_OBJ_USAGE;
1301         /* per-fragment overhead to be used by the client code */
1302         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1303         param->ddp_mntopts      = 0;
1304         if (test_opt(sb, XATTR_USER))
1305                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1306         if (test_opt(sb, POSIX_ACL))
1307                 param->ddp_mntopts |= MNTOPT_ACL;
1308
1309         /* LOD might calculate the max stripe count based on max_ea_size,
1310          * so we need take account in the overhead as well,
1311          * xattr_header + magic + xattr_entry_head */
1312         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1313                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1314
1315 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1316         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1317                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1318                                                                 ea_overhead;
1319         else
1320 #endif
1321                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1322 }
1323
1324 /*
1325  * Concurrency: shouldn't matter.
1326  */
1327 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1328 {
1329         int rc;
1330
1331         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1332
1333         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1334
1335         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1336                LUSTRE_OSD_LDISKFS_NAME, rc);
1337
1338         return rc;
1339 }
1340
1341 /**
1342  * Start commit for OSD device.
1343  *
1344  * An implementation of dt_commit_async method for OSD device.
1345  * Asychronously starts underlayng fs sync and thereby a transaction
1346  * commit.
1347  *
1348  * \param env environment
1349  * \param d dt device
1350  *
1351  * \see dt_device_operations
1352  */
1353 static int osd_commit_async(const struct lu_env *env,
1354                             struct dt_device *d)
1355 {
1356         struct super_block *s = osd_sb(osd_dt_dev(d));
1357         ENTRY;
1358
1359         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1360         RETURN(s->s_op->sync_fs(s, 0));
1361 }
1362
1363 /*
1364  * Concurrency: shouldn't matter.
1365  */
1366
1367 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1368 {
1369         struct super_block *sb = osd_sb(osd_dt_dev(d));
1370         struct block_device *dev = sb->s_bdev;
1371 #ifdef HAVE_DEV_SET_RDONLY
1372         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1373         int rc = 0;
1374 #else
1375         int rc = -EOPNOTSUPP;
1376 #endif
1377         ENTRY;
1378
1379 #ifdef HAVE_DEV_SET_RDONLY
1380         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1381
1382         if (jdev && (jdev != dev)) {
1383                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1384                        (long)jdev);
1385                 dev_set_rdonly(jdev);
1386         }
1387         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1388         dev_set_rdonly(dev);
1389 #else
1390         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1391                osd_dt_dev(d)->od_svname, (long)dev, rc);
1392 #endif
1393         RETURN(rc);
1394 }
1395
1396 /**
1397  * Note: we do not count into QUOTA here.
1398  * If we mount with --data_journal we may need more.
1399  */
1400 const int osd_dto_credits_noquota[DTO_NR] = {
1401         /**
1402          * Insert.
1403          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1404          * SINGLEDATA_TRANS_BLOCKS(8)
1405          * XXX Note: maybe iam need more, since iam have more level than
1406          *           EXT3 htree.
1407          */
1408         [DTO_INDEX_INSERT]  = 16,
1409         /**
1410          * Delete
1411          * just modify a single entry, probably merge few within a block
1412          */
1413         [DTO_INDEX_DELETE]  = 1,
1414         /**
1415          * Used for OI scrub
1416          */
1417         [DTO_INDEX_UPDATE]  = 16,
1418         /**
1419          * 4(inode, inode bits, groups, GDT)
1420          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1421          */
1422         [DTO_OBJECT_CREATE] = 4,
1423         /**
1424          * 4(inode, inode bits, groups, GDT)
1425          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1426          */
1427         [DTO_OBJECT_DELETE] = 4,
1428         /**
1429          * Attr set credits (inode)
1430          */
1431         [DTO_ATTR_SET_BASE] = 1,
1432         /**
1433          * Xattr set. The same as xattr of EXT3.
1434          * DATA_TRANS_BLOCKS(14)
1435          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1436          * are also counted in. Do not know why?
1437          */
1438         [DTO_XATTR_SET]     = 14,
1439         /**
1440          * credits for inode change during write.
1441          */
1442         [DTO_WRITE_BASE]    = 3,
1443         /**
1444          * credits for single block write.
1445          */
1446         [DTO_WRITE_BLOCK]   = 14,
1447         /**
1448          * Attr set credits for chown.
1449          * This is extra credits for setattr, and it is null without quota
1450          */
1451         [DTO_ATTR_SET_CHOWN] = 0
1452 };
1453
1454 static const struct dt_device_operations osd_dt_ops = {
1455         .dt_root_get       = osd_root_get,
1456         .dt_statfs         = osd_statfs,
1457         .dt_trans_create   = osd_trans_create,
1458         .dt_trans_start    = osd_trans_start,
1459         .dt_trans_stop     = osd_trans_stop,
1460         .dt_trans_cb_add   = osd_trans_cb_add,
1461         .dt_conf_get       = osd_conf_get,
1462         .dt_sync           = osd_sync,
1463         .dt_ro             = osd_ro,
1464         .dt_commit_async   = osd_commit_async,
1465 };
1466
1467 static void osd_object_read_lock(const struct lu_env *env,
1468                                  struct dt_object *dt, unsigned role)
1469 {
1470         struct osd_object *obj = osd_dt_obj(dt);
1471         struct osd_thread_info *oti = osd_oti_get(env);
1472
1473         LINVRNT(osd_invariant(obj));
1474
1475         LASSERT(obj->oo_owner != env);
1476         down_read_nested(&obj->oo_sem, role);
1477
1478         LASSERT(obj->oo_owner == NULL);
1479         oti->oti_r_locks++;
1480 }
1481
1482 static void osd_object_write_lock(const struct lu_env *env,
1483                                   struct dt_object *dt, unsigned role)
1484 {
1485         struct osd_object *obj = osd_dt_obj(dt);
1486         struct osd_thread_info *oti = osd_oti_get(env);
1487
1488         LINVRNT(osd_invariant(obj));
1489
1490         LASSERT(obj->oo_owner != env);
1491         down_write_nested(&obj->oo_sem, role);
1492
1493         LASSERT(obj->oo_owner == NULL);
1494         obj->oo_owner = env;
1495         oti->oti_w_locks++;
1496 }
1497
1498 static void osd_object_read_unlock(const struct lu_env *env,
1499                                    struct dt_object *dt)
1500 {
1501         struct osd_object *obj = osd_dt_obj(dt);
1502         struct osd_thread_info *oti = osd_oti_get(env);
1503
1504         LINVRNT(osd_invariant(obj));
1505
1506         LASSERT(oti->oti_r_locks > 0);
1507         oti->oti_r_locks--;
1508         up_read(&obj->oo_sem);
1509 }
1510
1511 static void osd_object_write_unlock(const struct lu_env *env,
1512                                     struct dt_object *dt)
1513 {
1514         struct osd_object *obj = osd_dt_obj(dt);
1515         struct osd_thread_info *oti = osd_oti_get(env);
1516
1517         LINVRNT(osd_invariant(obj));
1518
1519         LASSERT(obj->oo_owner == env);
1520         LASSERT(oti->oti_w_locks > 0);
1521         oti->oti_w_locks--;
1522         obj->oo_owner = NULL;
1523         up_write(&obj->oo_sem);
1524 }
1525
1526 static int osd_object_write_locked(const struct lu_env *env,
1527                                    struct dt_object *dt)
1528 {
1529         struct osd_object *obj = osd_dt_obj(dt);
1530
1531         LINVRNT(osd_invariant(obj));
1532
1533         return obj->oo_owner == env;
1534 }
1535
1536 static struct timespec *osd_inode_time(const struct lu_env *env,
1537                                        struct inode *inode, __u64 seconds)
1538 {
1539         struct osd_thread_info  *oti = osd_oti_get(env);
1540         struct timespec         *t   = &oti->oti_time;
1541
1542         t->tv_sec = seconds;
1543         t->tv_nsec = 0;
1544         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1545         return t;
1546 }
1547
1548
1549 static void osd_inode_getattr(const struct lu_env *env,
1550                               struct inode *inode, struct lu_attr *attr)
1551 {
1552         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1553                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1554                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1555                            LA_TYPE;
1556
1557         attr->la_atime   = LTIME_S(inode->i_atime);
1558         attr->la_mtime   = LTIME_S(inode->i_mtime);
1559         attr->la_ctime   = LTIME_S(inode->i_ctime);
1560         attr->la_mode    = inode->i_mode;
1561         attr->la_size    = i_size_read(inode);
1562         attr->la_blocks  = inode->i_blocks;
1563         attr->la_uid     = i_uid_read(inode);
1564         attr->la_gid     = i_gid_read(inode);
1565         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1566         attr->la_nlink   = inode->i_nlink;
1567         attr->la_rdev    = inode->i_rdev;
1568         attr->la_blksize = 1 << inode->i_blkbits;
1569         attr->la_blkbits = inode->i_blkbits;
1570 }
1571
1572 static int osd_attr_get(const struct lu_env *env,
1573                         struct dt_object *dt,
1574                         struct lu_attr *attr)
1575 {
1576         struct osd_object *obj = osd_dt_obj(dt);
1577
1578         if (!dt_object_exists(dt))
1579                 return -ENOENT;
1580
1581         LASSERT(!dt_object_remote(dt));
1582         LINVRNT(osd_invariant(obj));
1583
1584         spin_lock(&obj->oo_guard);
1585         osd_inode_getattr(env, obj->oo_inode, attr);
1586         spin_unlock(&obj->oo_guard);
1587         return 0;
1588 }
1589
1590 static int osd_declare_attr_set(const struct lu_env *env,
1591                                 struct dt_object *dt,
1592                                 const struct lu_attr *attr,
1593                                 struct thandle *handle)
1594 {
1595         struct osd_thandle     *oh;
1596         struct osd_object      *obj;
1597         struct osd_thread_info *info = osd_oti_get(env);
1598         struct lquota_id_info  *qi = &info->oti_qi;
1599         qid_t                   uid;
1600         qid_t                   gid;
1601         long long               bspace;
1602         int                     rc = 0;
1603         bool                    enforce;
1604         ENTRY;
1605
1606         LASSERT(dt != NULL);
1607         LASSERT(handle != NULL);
1608
1609         obj = osd_dt_obj(dt);
1610         LASSERT(osd_invariant(obj));
1611
1612         oh = container_of0(handle, struct osd_thandle, ot_super);
1613         LASSERT(oh->ot_handle == NULL);
1614
1615         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1616                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1617
1618         if (attr == NULL || obj->oo_inode == NULL)
1619                 RETURN(rc);
1620
1621         bspace   = obj->oo_inode->i_blocks;
1622         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1623         bspace   = toqb(bspace);
1624
1625         /* Changing ownership is always preformed by super user, it should not
1626          * fail with EDQUOT.
1627          *
1628          * We still need to call the osd_declare_qid() to calculate the journal
1629          * credits for updating quota accounting files and to trigger quota
1630          * space adjustment once the operation is completed.*/
1631         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1632                 /* USERQUOTA */
1633                 uid = i_uid_read(obj->oo_inode);
1634                 qi->lqi_type = USRQUOTA;
1635                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1636                 /* inode accounting */
1637                 qi->lqi_is_blk = false;
1638
1639                 /* one more inode for the new uid ... */
1640                 qi->lqi_id.qid_uid = attr->la_uid;
1641                 qi->lqi_space      = 1;
1642                 /* Reserve credits for the new uid */
1643                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1644                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1645                         rc = 0;
1646                 if (rc)
1647                         RETURN(rc);
1648
1649                 /* and one less inode for the current uid */
1650                 qi->lqi_id.qid_uid = uid;
1651                 qi->lqi_space      = -1;
1652                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1653                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1654                         rc = 0;
1655                 if (rc)
1656                         RETURN(rc);
1657
1658                 /* block accounting */
1659                 qi->lqi_is_blk = true;
1660
1661                 /* more blocks for the new uid ... */
1662                 qi->lqi_id.qid_uid = attr->la_uid;
1663                 qi->lqi_space      = bspace;
1664                 /*
1665                  * Credits for the new uid has been reserved, re-use "obj"
1666                  * to save credit reservation.
1667                  */
1668                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1669                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1670                         rc = 0;
1671                 if (rc)
1672                         RETURN(rc);
1673
1674                 /* and finally less blocks for the current uid */
1675                 qi->lqi_id.qid_uid = uid;
1676                 qi->lqi_space      = -bspace;
1677                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1678                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1679                         rc = 0;
1680                 if (rc)
1681                         RETURN(rc);
1682
1683                 /* GROUP QUOTA */
1684                 gid = i_gid_read(obj->oo_inode);
1685                 qi->lqi_type = GRPQUOTA;
1686                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1687
1688                 /* inode accounting */
1689                 qi->lqi_is_blk = false;
1690
1691                 /* one more inode for the new gid ... */
1692                 qi->lqi_id.qid_gid = attr->la_gid;
1693                 qi->lqi_space      = 1;
1694                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1695                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1696                         rc = 0;
1697                 if (rc)
1698                         RETURN(rc);
1699
1700                 /* and one less inode for the current gid */
1701                 qi->lqi_id.qid_gid = gid;
1702                 qi->lqi_space      = -1;
1703                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1704                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1705                         rc = 0;
1706                 if (rc)
1707                         RETURN(rc);
1708
1709                 /* block accounting */
1710                 qi->lqi_is_blk = true;
1711
1712                 /* more blocks for the new gid ... */
1713                 qi->lqi_id.qid_gid = attr->la_gid;
1714                 qi->lqi_space      = bspace;
1715                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1716                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1717                         rc = 0;
1718                 if (rc)
1719                         RETURN(rc);
1720
1721                 /* and finally less blocks for the current gid */
1722                 qi->lqi_id.qid_gid = gid;
1723                 qi->lqi_space      = -bspace;
1724                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1725                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1726                         rc = 0;
1727                 if (rc)
1728                         RETURN(rc);
1729         }
1730
1731         RETURN(rc);
1732 }
1733
1734 static int osd_inode_setattr(const struct lu_env *env,
1735                              struct inode *inode, const struct lu_attr *attr)
1736 {
1737         __u64 bits = attr->la_valid;
1738
1739         /* Only allow set size for regular file */
1740         if (!S_ISREG(inode->i_mode))
1741                 bits &= ~(LA_SIZE | LA_BLOCKS);
1742
1743         if (bits == 0)
1744                 return 0;
1745
1746         if (bits & LA_ATIME)
1747                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1748         if (bits & LA_CTIME)
1749                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1750         if (bits & LA_MTIME)
1751                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1752         if (bits & LA_SIZE) {
1753                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1754                 i_size_write(inode, attr->la_size);
1755         }
1756
1757 #if 0
1758         /* OSD should not change "i_blocks" which is used by quota.
1759          * "i_blocks" should be changed by ldiskfs only. */
1760         if (bits & LA_BLOCKS)
1761                 inode->i_blocks = attr->la_blocks;
1762 #endif
1763         if (bits & LA_MODE)
1764                 inode->i_mode = (inode->i_mode & S_IFMT) |
1765                                 (attr->la_mode & ~S_IFMT);
1766         if (bits & LA_UID)
1767                 i_uid_write(inode, attr->la_uid);
1768         if (bits & LA_GID)
1769                 i_gid_write(inode, attr->la_gid);
1770         if (bits & LA_NLINK)
1771                 set_nlink(inode, attr->la_nlink);
1772         if (bits & LA_RDEV)
1773                 inode->i_rdev = attr->la_rdev;
1774
1775         if (bits & LA_FLAGS) {
1776                 /* always keep S_NOCMTIME */
1777                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1778                                  S_NOCMTIME;
1779         }
1780         return 0;
1781 }
1782
1783 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1784 {
1785         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1786             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1787                 struct iattr    iattr;
1788                 int             rc;
1789
1790                 ll_vfs_dq_init(inode);
1791                 iattr.ia_valid = 0;
1792                 if (attr->la_valid & LA_UID)
1793                         iattr.ia_valid |= ATTR_UID;
1794                 if (attr->la_valid & LA_GID)
1795                         iattr.ia_valid |= ATTR_GID;
1796                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1797                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1798
1799                 rc = ll_vfs_dq_transfer(inode, &iattr);
1800                 if (rc) {
1801                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1802                                "enforcement enabled on the ldiskfs "
1803                                "filesystem?\n", inode->i_sb->s_id, rc);
1804                         return rc;
1805                 }
1806         }
1807         return 0;
1808 }
1809
1810 static int osd_attr_set(const struct lu_env *env,
1811                         struct dt_object *dt,
1812                         const struct lu_attr *attr,
1813                         struct thandle *handle)
1814 {
1815         struct osd_object *obj = osd_dt_obj(dt);
1816         struct inode      *inode;
1817         int rc;
1818
1819         if (!dt_object_exists(dt))
1820                 return -ENOENT;
1821
1822         LASSERT(handle != NULL);
1823         LASSERT(!dt_object_remote(dt));
1824         LASSERT(osd_invariant(obj));
1825
1826         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1827
1828         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1829                 struct osd_thread_info  *oti  = osd_oti_get(env);
1830                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1831                 struct lu_fid           *fid1 = &oti->oti_fid;
1832                 struct osd_inode_id     *id   = &oti->oti_id;
1833                 struct iam_path_descr   *ipd;
1834                 struct iam_container    *bag;
1835                 struct osd_thandle      *oh;
1836                 int                      rc;
1837
1838                 fid_cpu_to_be(fid1, fid0);
1839                 memset(id, 1, sizeof(*id));
1840                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1841                                   fid0)->oi_dir.od_container;
1842                 ipd = osd_idx_ipd_get(env, bag);
1843                 if (unlikely(ipd == NULL))
1844                         RETURN(-ENOMEM);
1845
1846                 oh = container_of0(handle, struct osd_thandle, ot_super);
1847                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1848                                 (const struct iam_rec *)id, ipd);
1849                 osd_ipd_put(env, bag, ipd);
1850                 return(rc > 0 ? 0 : rc);
1851         }
1852
1853         inode = obj->oo_inode;
1854
1855         rc = osd_quota_transfer(inode, attr);
1856         if (rc)
1857                 return rc;
1858
1859         spin_lock(&obj->oo_guard);
1860         rc = osd_inode_setattr(env, inode, attr);
1861         spin_unlock(&obj->oo_guard);
1862
1863         if (!rc)
1864                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1865
1866         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
1867
1868         return rc;
1869 }
1870
1871 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1872                                            struct osd_object *obj,
1873                                            const char *name, const int namelen)
1874 {
1875         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1876 }
1877
1878 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1879                       umode_t mode, struct dt_allocation_hint *hint,
1880                       struct thandle *th)
1881 {
1882         int result;
1883         struct osd_device  *osd = osd_obj2dev(obj);
1884         struct osd_thandle *oth;
1885         struct dt_object   *parent = NULL;
1886         struct inode       *inode;
1887
1888         LINVRNT(osd_invariant(obj));
1889         LASSERT(obj->oo_inode == NULL);
1890         LASSERT(obj->oo_hl_head == NULL);
1891
1892         if (S_ISDIR(mode) && ldiskfs_pdo) {
1893                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1894                 if (obj->oo_hl_head == NULL)
1895                         return -ENOMEM;
1896         }
1897
1898         oth = container_of(th, struct osd_thandle, ot_super);
1899         LASSERT(oth->ot_handle->h_transaction != NULL);
1900
1901         if (hint != NULL && hint->dah_parent != NULL &&
1902             !dt_object_remote(hint->dah_parent))
1903                 parent = hint->dah_parent;
1904
1905         inode = ldiskfs_create_inode(oth->ot_handle,
1906                                      parent ? osd_dt_obj(parent)->oo_inode :
1907                                               osd_sb(osd)->s_root->d_inode,
1908                                      mode);
1909         if (!IS_ERR(inode)) {
1910                 /* Do not update file c/mtime in ldiskfs. */
1911                 inode->i_flags |= S_NOCMTIME;
1912
1913                 /* For new created object, it must be consistent,
1914                  * and it is unnecessary to scrub against it. */
1915                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1916
1917                 obj->oo_inode = inode;
1918                 result = 0;
1919         } else {
1920                 if (obj->oo_hl_head != NULL) {
1921                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1922                         obj->oo_hl_head = NULL;
1923                 }
1924                 result = PTR_ERR(inode);
1925         }
1926         LINVRNT(osd_invariant(obj));
1927         return result;
1928 }
1929
1930 enum {
1931         OSD_NAME_LEN = 255
1932 };
1933
1934 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1935                      struct lu_attr *attr,
1936                      struct dt_allocation_hint *hint,
1937                      struct dt_object_format *dof,
1938                      struct thandle *th)
1939 {
1940         int result;
1941         struct osd_thandle *oth;
1942         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1943
1944         LASSERT(S_ISDIR(attr->la_mode));
1945
1946         oth = container_of(th, struct osd_thandle, ot_super);
1947         LASSERT(oth->ot_handle->h_transaction != NULL);
1948         result = osd_mkfile(info, obj, mode, hint, th);
1949
1950         return result;
1951 }
1952
1953 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1954                         struct lu_attr *attr,
1955                         struct dt_allocation_hint *hint,
1956                         struct dt_object_format *dof,
1957                         struct thandle *th)
1958 {
1959         int result;
1960         struct osd_thandle *oth;
1961         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1962
1963         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1964
1965         LASSERT(S_ISREG(attr->la_mode));
1966
1967         oth = container_of(th, struct osd_thandle, ot_super);
1968         LASSERT(oth->ot_handle->h_transaction != NULL);
1969
1970         result = osd_mkfile(info, obj, mode, hint, th);
1971         if (result == 0) {
1972                 LASSERT(obj->oo_inode != NULL);
1973                 if (feat->dif_flags & DT_IND_VARKEY)
1974                         result = iam_lvar_create(obj->oo_inode,
1975                                                  feat->dif_keysize_max,
1976                                                  feat->dif_ptrsize,
1977                                                  feat->dif_recsize_max,
1978                                                  oth->ot_handle);
1979                 else
1980                         result = iam_lfix_create(obj->oo_inode,
1981                                                  feat->dif_keysize_max,
1982                                                  feat->dif_ptrsize,
1983                                                  feat->dif_recsize_max,
1984                                                  oth->ot_handle);
1985
1986         }
1987         return result;
1988 }
1989
1990 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1991                      struct lu_attr *attr,
1992                      struct dt_allocation_hint *hint,
1993                      struct dt_object_format *dof,
1994                      struct thandle *th)
1995 {
1996         LASSERT(S_ISREG(attr->la_mode));
1997         return osd_mkfile(info, obj, (attr->la_mode &
1998                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1999 }
2000
2001 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2002                      struct lu_attr *attr,
2003                      struct dt_allocation_hint *hint,
2004                      struct dt_object_format *dof,
2005                      struct thandle *th)
2006 {
2007         LASSERT(S_ISLNK(attr->la_mode));
2008         return osd_mkfile(info, obj, (attr->la_mode &
2009                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2010 }
2011
2012 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2013                      struct lu_attr *attr,
2014                      struct dt_allocation_hint *hint,
2015                      struct dt_object_format *dof,
2016                      struct thandle *th)
2017 {
2018         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2019         int result;
2020
2021         LINVRNT(osd_invariant(obj));
2022         LASSERT(obj->oo_inode == NULL);
2023         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2024                 S_ISFIFO(mode) || S_ISSOCK(mode));
2025
2026         result = osd_mkfile(info, obj, mode, hint, th);
2027         if (result == 0) {
2028                 LASSERT(obj->oo_inode != NULL);
2029                 /*
2030                  * This inode should be marked dirty for i_rdev.  Currently
2031                  * that is done in the osd_attr_init().
2032                  */
2033                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2034                                    attr->la_rdev);
2035         }
2036         LINVRNT(osd_invariant(obj));
2037         return result;
2038 }
2039
2040 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2041                               struct lu_attr *,
2042                               struct dt_allocation_hint *hint,
2043                               struct dt_object_format *dof,
2044                               struct thandle *);
2045
2046 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2047 {
2048         osd_obj_type_f result;
2049
2050         switch (type) {
2051         case DFT_DIR:
2052                 result = osd_mkdir;
2053                 break;
2054         case DFT_REGULAR:
2055                 result = osd_mkreg;
2056                 break;
2057         case DFT_SYM:
2058                 result = osd_mksym;
2059                 break;
2060         case DFT_NODE:
2061                 result = osd_mknod;
2062                 break;
2063         case DFT_INDEX:
2064                 result = osd_mk_index;
2065                 break;
2066
2067         default:
2068                 LBUG();
2069                 break;
2070         }
2071         return result;
2072 }
2073
2074
2075 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2076                         struct dt_object *parent, struct dt_object *child,
2077                         umode_t child_mode)
2078 {
2079         LASSERT(ah);
2080
2081         ah->dah_parent = parent;
2082         ah->dah_mode = child_mode;
2083 }
2084
2085 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2086                           struct lu_attr *attr, struct dt_object_format *dof)
2087 {
2088         struct inode   *inode = obj->oo_inode;
2089         __u64           valid = attr->la_valid;
2090         int             result;
2091
2092         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2093
2094         if (dof->dof_type != DFT_NODE)
2095                 attr->la_valid &= ~LA_RDEV;
2096         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2097                 attr->la_valid &= ~LA_ATIME;
2098         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2099                 attr->la_valid &= ~LA_CTIME;
2100         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2101                 attr->la_valid &= ~LA_MTIME;
2102
2103         result = osd_quota_transfer(inode, attr);
2104         if (result)
2105                 return;
2106
2107         if (attr->la_valid != 0) {
2108                 result = osd_inode_setattr(info->oti_env, inode, attr);
2109                 /*
2110                  * The osd_inode_setattr() should always succeed here.  The
2111                  * only error that could be returned is EDQUOT when we are
2112                  * trying to change the UID or GID of the inode. However, this
2113                  * should not happen since quota enforcement is no longer
2114                  * enabled on ldiskfs (lquota takes care of it).
2115                  */
2116                 LASSERTF(result == 0, "%d\n", result);
2117                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2118         }
2119
2120         attr->la_valid = valid;
2121 }
2122
2123 /**
2124  * Helper function for osd_object_create()
2125  *
2126  * \retval 0, on success
2127  */
2128 static int __osd_object_create(struct osd_thread_info *info,
2129                                struct osd_object *obj, struct lu_attr *attr,
2130                                struct dt_allocation_hint *hint,
2131                                struct dt_object_format *dof,
2132                                struct thandle *th)
2133 {
2134         int     result;
2135         __u32   umask;
2136
2137         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2138
2139         /* we drop umask so that permissions we pass are not affected */
2140         umask = current->fs->umask;
2141         current->fs->umask = 0;
2142
2143         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2144                                                   th);
2145         if (likely(obj->oo_inode != NULL)) {
2146                 LASSERT(obj->oo_inode->i_state & I_NEW);
2147
2148                 /* Unlock the inode before attr initialization to avoid
2149                  * unnecessary dqget operations. LU-6378 */
2150                 unlock_new_inode(obj->oo_inode);
2151         }
2152
2153         if (likely(result == 0)) {
2154                 osd_attr_init(info, obj, attr, dof);
2155                 osd_object_init0(obj);
2156         }
2157
2158         /* restore previous umask value */
2159         current->fs->umask = umask;
2160
2161         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2162
2163         return result;
2164 }
2165
2166 /**
2167  * Helper function for osd_object_create()
2168  *
2169  * \retval 0, on success
2170  */
2171 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2172                            const struct lu_fid *fid, struct thandle *th)
2173 {
2174         struct osd_thread_info *info = osd_oti_get(env);
2175         struct osd_inode_id    *id   = &info->oti_id;
2176         struct osd_device      *osd  = osd_obj2dev(obj);
2177         struct osd_thandle     *oh;
2178         int                     rc;
2179
2180         LASSERT(obj->oo_inode != NULL);
2181
2182         oh = container_of0(th, struct osd_thandle, ot_super);
2183         LASSERT(oh->ot_handle);
2184         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2185
2186         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2187         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2188         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2189
2190         return rc;
2191 }
2192
2193 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2194                    u64 seq, struct lu_seq_range *range)
2195 {
2196         struct seq_server_site  *ss = osd_seq_site(osd);
2197
2198         if (fid_seq_is_idif(seq)) {
2199                 fld_range_set_ost(range);
2200                 range->lsr_index = idif_ost_idx(seq);
2201                 return 0;
2202         }
2203
2204         if (!fid_seq_in_fldb(seq)) {
2205                 fld_range_set_mdt(range);
2206                 if (ss != NULL)
2207                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2208                          * at all */
2209                         range->lsr_index = ss->ss_node_id;
2210                 return 0;
2211         }
2212
2213         LASSERT(ss != NULL);
2214         fld_range_set_any(range);
2215         /* OSD will only do local fld lookup */
2216         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2217 }
2218
2219 /*
2220  * Concurrency: no external locking is necessary.
2221  */
2222 static int osd_declare_object_create(const struct lu_env *env,
2223                                      struct dt_object *dt,
2224                                      struct lu_attr *attr,
2225                                      struct dt_allocation_hint *hint,
2226                                      struct dt_object_format *dof,
2227                                      struct thandle *handle)
2228 {
2229         struct osd_thandle      *oh;
2230         int                      rc;
2231         ENTRY;
2232
2233         LASSERT(handle != NULL);
2234
2235         oh = container_of0(handle, struct osd_thandle, ot_super);
2236         LASSERT(oh->ot_handle == NULL);
2237
2238         /* EA object consumes more credits than regular object: osd_mk_index
2239          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2240          * leaf_node, could involves the block, block bitmap, groups, GDT
2241          * change for each block, so add 4 * 2 credits in that case. */
2242         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2243                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2244                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2245         /* Reuse idle OI block may cause additional one OI block
2246          * to be changed. */
2247         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2248                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2249
2250         if (!attr)
2251                 RETURN(0);
2252
2253         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2254                                    osd_dt_obj(dt), false, NULL, false);
2255         if (rc != 0)
2256                 RETURN(rc);
2257
2258         RETURN(rc);
2259 }
2260
2261 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2262                              struct lu_attr *attr,
2263                              struct dt_allocation_hint *hint,
2264                              struct dt_object_format *dof, struct thandle *th)
2265 {
2266         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2267         struct osd_object       *obj    = osd_dt_obj(dt);
2268         struct osd_thread_info  *info   = osd_oti_get(env);
2269         int result;
2270         ENTRY;
2271
2272         if (dt_object_exists(dt))
2273                 return -EEXIST;
2274
2275         LINVRNT(osd_invariant(obj));
2276         LASSERT(!dt_object_remote(dt));
2277         LASSERT(osd_write_locked(env, obj));
2278         LASSERT(th != NULL);
2279
2280         if (unlikely(fid_is_acct(fid)))
2281                 /* Quota files can't be created from the kernel any more,
2282                  * 'tune2fs -O quota' will take care of creating them */
2283                 RETURN(-EPERM);
2284
2285         result = __osd_object_create(info, obj, attr, hint, dof, th);
2286         if (result == 0) {
2287                 result = __osd_oi_insert(env, obj, fid, th);
2288                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2289                         obj->oo_dt.do_body_ops = &osd_body_ops;
2290         }
2291         LASSERT(ergo(result == 0,
2292                 dt_object_exists(dt) && !dt_object_remote(dt)));
2293
2294         LASSERT(osd_invariant(obj));
2295         RETURN(result);
2296 }
2297
2298 /**
2299  * Called to destroy on-disk representation of the object
2300  *
2301  * Concurrency: must be locked
2302  */
2303 static int osd_declare_object_destroy(const struct lu_env *env,
2304                                       struct dt_object *dt,
2305                                       struct thandle *th)
2306 {
2307         struct osd_object  *obj = osd_dt_obj(dt);
2308         struct inode       *inode = obj->oo_inode;
2309         struct osd_thandle *oh;
2310         int                 rc;
2311         ENTRY;
2312
2313         oh = container_of0(th, struct osd_thandle, ot_super);
2314         LASSERT(oh->ot_handle == NULL);
2315         LASSERT(inode);
2316
2317         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2318                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2319         /* Recycle idle OI leaf may cause additional three OI blocks
2320          * to be changed. */
2321         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2322                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2323         /* one less inode */
2324         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2325                                    -1, oh, obj, false, NULL, false);
2326         if (rc)
2327                 RETURN(rc);
2328         /* data to be truncated */
2329         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2330                                    0, oh, obj, true, NULL, false);
2331         RETURN(rc);
2332 }
2333
2334 static int osd_object_destroy(const struct lu_env *env,
2335                               struct dt_object *dt,
2336                               struct thandle *th)
2337 {
2338         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2339         struct osd_object      *obj = osd_dt_obj(dt);
2340         struct inode           *inode = obj->oo_inode;
2341         struct osd_device      *osd = osd_obj2dev(obj);
2342         struct osd_thandle     *oh;
2343         int                     result;
2344         ENTRY;
2345
2346         oh = container_of0(th, struct osd_thandle, ot_super);
2347         LASSERT(oh->ot_handle);
2348         LASSERT(inode);
2349         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2350
2351         if (unlikely(fid_is_acct(fid)))
2352                 RETURN(-EPERM);
2353
2354         if (S_ISDIR(inode->i_mode)) {
2355                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2356                         inode->i_nlink == 2);
2357                 /* it will check/delete the inode from remote parent,
2358                  * how to optimize it? unlink performance impaction XXX */
2359                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2360                 if (result != 0 && result != -ENOENT) {
2361                         CERROR("%s: delete inode "DFID": rc = %d\n",
2362                                osd_name(osd), PFID(fid), result);
2363                 }
2364                 spin_lock(&obj->oo_guard);
2365                 clear_nlink(inode);
2366                 spin_unlock(&obj->oo_guard);
2367                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2368         }
2369
2370         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2371
2372         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2373                                OI_CHECK_FLD);
2374
2375         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2376         /* XXX: add to ext3 orphan list */
2377         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2378
2379         /* not needed in the cache anymore */
2380         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2381         obj->oo_destroyed = 1;
2382
2383         RETURN(0);
2384 }
2385
2386 /**
2387  * Put the fid into lustre_mdt_attrs, and then place the structure
2388  * inode's ea. This fid should not be altered during the life time
2389  * of the inode.
2390  *
2391  * \retval +ve, on success
2392  * \retval -ve, on error
2393  *
2394  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2395  */
2396 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2397                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2398 {
2399         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2400         int                      rc;
2401         ENTRY;
2402
2403         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2404                 RETURN(0);
2405
2406         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2407                 rc = -ENOMEM;
2408
2409         lustre_lma_init(lma, fid, compat, incompat);
2410         lustre_lma_swab(lma);
2411
2412         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2413                              XATTR_CREATE);
2414         /* LMA may already exist, but we need to check that all the
2415          * desired compat/incompat flags have been added. */
2416         if (unlikely(rc == -EEXIST)) {
2417                 if (compat == 0 && incompat == 0)
2418                         RETURN(0);
2419
2420                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2421                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2422                                      LMA_OLD_SIZE);
2423                 if (rc <= 0)
2424                         RETURN(-EINVAL);
2425
2426                 lustre_lma_swab(lma);
2427                 if (!(~lma->lma_compat & compat) &&
2428                     !(~lma->lma_incompat & incompat))
2429                         RETURN(0);
2430
2431                 lma->lma_compat |= compat;
2432                 lma->lma_incompat |= incompat;
2433                 lustre_lma_swab(lma);
2434                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2435                                      sizeof(*lma), XATTR_REPLACE);
2436         }
2437
2438         RETURN(rc);
2439 }
2440
2441 /**
2442  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2443  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2444  * To have compatilibility with 1.8 ldiskfs driver we need to have
2445  * magic number at start of fid data.
2446  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2447  * its inmemory API.
2448  */
2449 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2450                                          const struct lu_fid *fid)
2451 {
2452         if (!fid_is_namespace_visible(fid) ||
2453             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2454                 param->edp_magic = 0;
2455                 return;
2456         }
2457
2458         param->edp_magic = LDISKFS_LUFID_MAGIC;
2459         param->edp_len =  sizeof(struct lu_fid) + 1;
2460         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2461 }
2462
2463 /**
2464  * Try to read the fid from inode ea into dt_rec.
2465  *
2466  * \param fid object fid.
2467  *
2468  * \retval 0 on success
2469  */
2470 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2471                           __u32 ino, struct lu_fid *fid,
2472                           struct osd_inode_id *id)
2473 {
2474         struct osd_thread_info *info  = osd_oti_get(env);
2475         struct inode           *inode;
2476         ENTRY;
2477
2478         osd_id_gen(id, ino, OSD_OII_NOGEN);
2479         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2480         if (IS_ERR(inode))
2481                 RETURN(PTR_ERR(inode));
2482
2483         iput(inode);
2484         RETURN(0);
2485 }
2486
2487 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2488                                         struct inode *dir,
2489                                         struct inode *parent_dir,
2490                                         const struct lu_fid *dot_fid,
2491                                         const struct lu_fid *dot_dot_fid,
2492                                         struct osd_thandle *oth)
2493 {
2494         struct ldiskfs_dentry_param *dot_ldp;
2495         struct ldiskfs_dentry_param *dot_dot_ldp;
2496
2497         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2498         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2499
2500         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2501         dot_ldp->edp_magic = 0;
2502
2503         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2504                                       dir, dot_ldp, dot_dot_ldp);
2505 }
2506
2507 /**
2508  * Create an local agent inode for remote entry
2509  */
2510 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2511                                                   struct osd_device *osd,
2512                                                   struct osd_object *pobj,
2513                                                   const struct lu_fid *fid,
2514                                                   __u32 type,
2515                                                   struct thandle *th)
2516 {
2517         struct osd_thread_info  *info = osd_oti_get(env);
2518         struct inode            *local;
2519         struct osd_thandle      *oh;
2520         int                     rc;
2521         ENTRY;
2522
2523         LASSERT(th);
2524         oh = container_of(th, struct osd_thandle, ot_super);
2525         LASSERT(oh->ot_handle->h_transaction != NULL);
2526
2527         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2528         if (IS_ERR(local)) {
2529                 CERROR("%s: create local error %d\n", osd_name(osd),
2530                        (int)PTR_ERR(local));
2531                 RETURN(local);
2532         }
2533
2534         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2535         unlock_new_inode(local);
2536
2537         /* Set special LMA flag for local agent inode */
2538         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2539         if (rc != 0) {
2540                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2541                        osd_name(osd), PFID(fid), rc);
2542                 RETURN(ERR_PTR(rc));
2543         }
2544
2545         if (!S_ISDIR(type))
2546                 RETURN(local);
2547
2548         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2549                                          lu_object_fid(&pobj->oo_dt.do_lu),
2550                                          fid, oh);
2551         if (rc != 0) {
2552                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2553                         osd_name(osd), PFID(fid), rc);
2554                 RETURN(ERR_PTR(rc));
2555         }
2556
2557         RETURN(local);
2558 }
2559
2560 /**
2561  * Delete local agent inode for remote entry
2562  */
2563 static int osd_delete_local_agent_inode(const struct lu_env *env,
2564                                         struct osd_device *osd,
2565                                         const struct lu_fid *fid,
2566                                         __u32 ino, struct osd_thandle *oh)
2567 {
2568         struct osd_thread_info  *oti = osd_oti_get(env);
2569         struct osd_inode_id     *id = &oti->oti_id;
2570         struct inode            *inode;
2571         ENTRY;
2572
2573         id->oii_ino = le32_to_cpu(ino);
2574         id->oii_gen = OSD_OII_NOGEN;
2575         inode = osd_iget(oti, osd, id);
2576         if (IS_ERR(inode)) {
2577                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2578                        PFID(fid), id->oii_ino, id->oii_gen);
2579                 RETURN(PTR_ERR(inode));
2580         }
2581
2582         clear_nlink(inode);
2583         mark_inode_dirty(inode);
2584         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2585                 osd_name(osd), PFID(fid), inode->i_ino);
2586         iput(inode);
2587         RETURN(0);
2588 }
2589
2590 /**
2591  * OSD layer object create function for interoperability mode (b11826).
2592  * This is mostly similar to osd_object_create(). Only difference being, fid is
2593  * inserted into inode ea here.
2594  *
2595  * \retval   0, on success
2596  * \retval -ve, on error
2597  */
2598 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2599                                 struct lu_attr *attr,
2600                                 struct dt_allocation_hint *hint,
2601                                 struct dt_object_format *dof,
2602                                 struct thandle *th)
2603 {
2604         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2605         struct osd_object       *obj    = osd_dt_obj(dt);
2606         struct osd_thread_info  *info   = osd_oti_get(env);
2607         int                      result;
2608
2609         ENTRY;
2610
2611         if (dt_object_exists(dt))
2612                 RETURN(-EEXIST);
2613
2614         LASSERT(osd_invariant(obj));
2615         LASSERT(!dt_object_remote(dt));
2616         LASSERT(osd_write_locked(env, obj));
2617         LASSERT(th != NULL);
2618
2619         if (unlikely(fid_is_acct(fid)))
2620                 /* Quota files can't be created from the kernel any more,
2621                  * 'tune2fs -O quota' will take care of creating them */
2622                 RETURN(-EPERM);
2623
2624         result = __osd_object_create(info, obj, attr, hint, dof, th);
2625         if (result == 0) {
2626                 if (fid_is_idif(fid) &&
2627                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2628                         struct lu_fid *tfid = &info->oti_fid;
2629                         struct ost_id *oi   = &info->oti_ostid;
2630
2631                         fid_to_ostid(fid, oi);
2632                         ostid_to_fid(tfid, oi, 0);
2633                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2634                                                 LMAC_FID_ON_OST, 0);
2635                 } else {
2636                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2637                                 fid_is_on_ost(info, osd_obj2dev(obj),
2638                                               fid, OI_CHECK_FLD) ?
2639                                 LMAC_FID_ON_OST : 0, 0);
2640                 }
2641                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2642                         obj->oo_dt.do_body_ops = &osd_body_ops;
2643         }
2644
2645         if (result == 0)
2646                 result = __osd_oi_insert(env, obj, fid, th);
2647
2648         LASSERT(ergo(result == 0,
2649                      dt_object_exists(dt) && !dt_object_remote(dt)));
2650         LINVRNT(osd_invariant(obj));
2651         RETURN(result);
2652 }
2653
2654 static int osd_declare_object_ref_add(const struct lu_env *env,
2655                                       struct dt_object *dt,
2656                                       struct thandle *handle)
2657 {
2658         struct osd_thandle       *oh;
2659
2660         /* it's possible that object doesn't exist yet */
2661         LASSERT(handle != NULL);
2662
2663         oh = container_of0(handle, struct osd_thandle, ot_super);
2664         LASSERT(oh->ot_handle == NULL);
2665
2666         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2667                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2668
2669         return 0;
2670 }
2671
2672 /*
2673  * Concurrency: @dt is write locked.
2674  */
2675 static int osd_object_ref_add(const struct lu_env *env,
2676                               struct dt_object *dt, struct thandle *th)
2677 {
2678         struct osd_object  *obj = osd_dt_obj(dt);
2679         struct inode       *inode = obj->oo_inode;
2680         struct osd_thandle *oh;
2681         int                 rc = 0;
2682
2683         if (!dt_object_exists(dt))
2684                 return -ENOENT;
2685
2686         LINVRNT(osd_invariant(obj));
2687         LASSERT(!dt_object_remote(dt));
2688         LASSERT(osd_write_locked(env, obj));
2689         LASSERT(th != NULL);
2690
2691         oh = container_of0(th, struct osd_thandle, ot_super);
2692         LASSERT(oh->ot_handle != NULL);
2693
2694         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2695
2696         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2697                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2698         /*
2699          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2700          * (65000) subdirectories by storing "1" in i_nlink if the link count
2701          * would otherwise overflow. Directory tranversal tools understand
2702          * that (st_nlink == 1) indicates that the filesystem dose not track
2703          * hard links count on the directory, and will not abort subdirectory
2704          * scanning early once (st_nlink - 2) subdirs have been found.
2705          *
2706          * This also has to properly handle the case of inodes with nlink == 0
2707          * in case they are being linked into the PENDING directory
2708          */
2709         spin_lock(&obj->oo_guard);
2710         if (unlikely(inode->i_nlink == 0))
2711                 /* inc_nlink from 0 may cause WARN_ON */
2712                 set_nlink(inode, 1);
2713         else {
2714                 ldiskfs_inc_count(oh->ot_handle, inode);
2715                 if (!S_ISDIR(inode->i_mode))
2716                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2717         }
2718         spin_unlock(&obj->oo_guard);
2719
2720         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2721         LINVRNT(osd_invariant(obj));
2722
2723         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2724
2725         return rc;
2726 }
2727
2728 static int osd_declare_object_ref_del(const struct lu_env *env,
2729                                       struct dt_object *dt,
2730                                       struct thandle *handle)
2731 {
2732         struct osd_thandle *oh;
2733
2734         LASSERT(!dt_object_remote(dt));
2735         LASSERT(handle != NULL);
2736
2737         oh = container_of0(handle, struct osd_thandle, ot_super);
2738         LASSERT(oh->ot_handle == NULL);
2739
2740         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2741                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2742
2743         return 0;
2744 }
2745
2746 /*
2747  * Concurrency: @dt is write locked.
2748  */
2749 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2750                               struct thandle *th)
2751 {
2752         struct osd_object       *obj = osd_dt_obj(dt);
2753         struct inode            *inode = obj->oo_inode;
2754         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2755         struct osd_thandle      *oh;
2756
2757         if (!dt_object_exists(dt))
2758                 return -ENOENT;
2759
2760         LINVRNT(osd_invariant(obj));
2761         LASSERT(!dt_object_remote(dt));
2762         LASSERT(osd_write_locked(env, obj));
2763         LASSERT(th != NULL);
2764
2765         oh = container_of0(th, struct osd_thandle, ot_super);
2766         LASSERT(oh->ot_handle != NULL);
2767
2768         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2769
2770         spin_lock(&obj->oo_guard);
2771         /* That can be result of upgrade from old Lustre version and
2772          * applied only to local files.  Just skip this ref_del call.
2773          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2774         if (inode->i_nlink == 0) {
2775                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2776                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2777                              ", maybe an upgraded file? (LU-3915)\n",
2778                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2779                 spin_unlock(&obj->oo_guard);
2780                 return 0;
2781         }
2782
2783         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2784                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2785
2786         ldiskfs_dec_count(oh->ot_handle, inode);
2787         spin_unlock(&obj->oo_guard);
2788
2789         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2790         LINVRNT(osd_invariant(obj));
2791
2792         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
2793
2794         return 0;
2795 }
2796
2797 /*
2798  * Get the 64-bit version for an inode.
2799  */
2800 static int osd_object_version_get(const struct lu_env *env,
2801                                   struct dt_object *dt, dt_obj_version_t *ver)
2802 {
2803         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2804
2805         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2806                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2807         *ver = LDISKFS_I(inode)->i_fs_version;
2808         return 0;
2809 }
2810
2811 /*
2812  * Concurrency: @dt is read locked.
2813  */
2814 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2815                          struct lu_buf *buf, const char *name)
2816 {
2817         struct osd_object      *obj    = osd_dt_obj(dt);
2818         struct inode           *inode  = obj->oo_inode;
2819         struct osd_thread_info *info   = osd_oti_get(env);
2820         struct dentry          *dentry = &info->oti_obj_dentry;
2821
2822         /* version get is not real XATTR but uses xattr API */
2823         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2824                 /* for version we are just using xattr API but change inode
2825                  * field instead */
2826                 if (buf->lb_len == 0)
2827                         return sizeof(dt_obj_version_t);
2828
2829                 if (buf->lb_len < sizeof(dt_obj_version_t))
2830                         return -ERANGE;
2831
2832                 osd_object_version_get(env, dt, buf->lb_buf);
2833
2834                 return sizeof(dt_obj_version_t);
2835         }
2836
2837         if (!dt_object_exists(dt))
2838                 return -ENOENT;
2839
2840         LASSERT(!dt_object_remote(dt));
2841         LASSERT(inode->i_op != NULL);
2842         LASSERT(inode->i_op->getxattr != NULL);
2843
2844         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2845 }
2846
2847
2848 static int osd_declare_xattr_set(const struct lu_env *env,
2849                                  struct dt_object *dt,
2850                                  const struct lu_buf *buf, const char *name,
2851                                  int fl, struct thandle *handle)
2852 {
2853         struct osd_thandle *oh;
2854         int credits;
2855         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2856
2857         LASSERT(handle != NULL);
2858
2859         oh = container_of0(handle, struct osd_thandle, ot_super);
2860         LASSERT(oh->ot_handle == NULL);
2861
2862         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2863                 /* For non-upgrading case, the LMA is set first and
2864                  * usually fit inode. But for upgrade case, the LMA
2865                  * may be in another separated EA block. */
2866                 if (!dt_object_exists(dt))
2867                         credits = 0;
2868                 else if (fl == LU_XATTR_REPLACE)
2869                         credits = 1;
2870                 else
2871                         goto upgrade;
2872         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2873                 credits = 1;
2874         } else {
2875 upgrade:
2876                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2877                 if (buf && buf->lb_len > sb->s_blocksize) {
2878                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2879                                         sb->s_blocksize_bits;
2880                 }
2881                 /*
2882                  * xattr set may involve inode quota change, reserve credits for
2883                  * dquot_initialize()
2884                  */
2885                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2886         }
2887
2888         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2889
2890         return 0;
2891 }
2892
2893 /*
2894  * Set the 64-bit version for object
2895  */
2896 static void osd_object_version_set(const struct lu_env *env,
2897                                    struct dt_object *dt,
2898                                    dt_obj_version_t *new_version)
2899 {
2900         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2901
2902         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2903                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2904
2905         LDISKFS_I(inode)->i_fs_version = *new_version;
2906         /** Version is set after all inode operations are finished,
2907          *  so we should mark it dirty here */
2908         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2909 }
2910
2911 /*
2912  * Concurrency: @dt is write locked.
2913  */
2914 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2915                          const struct lu_buf *buf, const char *name, int fl,
2916                          struct thandle *handle)
2917 {
2918         struct osd_object      *obj      = osd_dt_obj(dt);
2919         struct inode           *inode    = obj->oo_inode;
2920         struct osd_thread_info *info     = osd_oti_get(env);
2921         int                     fs_flags = 0;
2922         int                     rc;
2923         ENTRY;
2924
2925         LASSERT(handle != NULL);
2926
2927         /* version set is not real XATTR */
2928         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2929                 /* for version we are just using xattr API but change inode
2930                  * field instead */
2931                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2932                 osd_object_version_set(env, dt, buf->lb_buf);
2933                 return sizeof(dt_obj_version_t);
2934         }
2935
2936         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2937                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2938
2939         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2940         if (fl & LU_XATTR_REPLACE)
2941                 fs_flags |= XATTR_REPLACE;
2942
2943         if (fl & LU_XATTR_CREATE)
2944                 fs_flags |= XATTR_CREATE;
2945
2946         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2947                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2948                 int                      rc;
2949
2950                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2951                 if (rc != 0)
2952                         RETURN(rc);
2953
2954                 lma->lma_incompat |= LMAI_STRIPED;
2955                 lustre_lma_swab(lma);
2956                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2957                                      sizeof(*lma), XATTR_REPLACE);
2958                 if (rc != 0)
2959                         RETURN(rc);
2960         }
2961
2962         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
2963             strcmp(name, XATTR_NAME_LINK) == 0)
2964                 return -ENOSPC;
2965
2966         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2967                                fs_flags);
2968         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2969
2970         return rc;
2971 }
2972
2973 /*
2974  * Concurrency: @dt is read locked.
2975  */
2976 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2977                           const struct lu_buf *buf)
2978 {
2979         struct osd_object      *obj    = osd_dt_obj(dt);
2980         struct inode           *inode  = obj->oo_inode;
2981         struct osd_thread_info *info   = osd_oti_get(env);
2982         struct dentry          *dentry = &info->oti_obj_dentry;
2983
2984         if (!dt_object_exists(dt))
2985                 return -ENOENT;
2986
2987         LASSERT(!dt_object_remote(dt));
2988         LASSERT(inode->i_op != NULL);
2989         LASSERT(inode->i_op->listxattr != NULL);
2990
2991         dentry->d_inode = inode;
2992         dentry->d_sb = inode->i_sb;
2993         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2994 }
2995
2996 static int osd_declare_xattr_del(const struct lu_env *env,
2997                                  struct dt_object *dt, const char *name,
2998                                  struct thandle *handle)
2999 {
3000         struct osd_thandle *oh;
3001         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3002
3003         LASSERT(!dt_object_remote(dt));
3004         LASSERT(handle != NULL);
3005
3006         oh = container_of0(handle, struct osd_thandle, ot_super);
3007         LASSERT(oh->ot_handle == NULL);
3008
3009         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3010                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3011         /*
3012          * xattr del may involve inode quota change, reserve credits for
3013          * dquot_initialize()
3014          */
3015         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3016
3017         return 0;
3018 }
3019
3020 /*
3021  * Concurrency: @dt is write locked.
3022  */
3023 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3024                          const char *name, struct thandle *handle)
3025 {
3026         struct osd_object      *obj    = osd_dt_obj(dt);
3027         struct inode           *inode  = obj->oo_inode;
3028         struct osd_thread_info *info   = osd_oti_get(env);
3029         struct dentry          *dentry = &info->oti_obj_dentry;
3030         int                     rc;
3031
3032         if (!dt_object_exists(dt))
3033                 return -ENOENT;
3034
3035         LASSERT(!dt_object_remote(dt));
3036         LASSERT(inode->i_op != NULL);
3037         LASSERT(inode->i_op->removexattr != NULL);
3038         LASSERT(handle != NULL);
3039
3040         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3041
3042         ll_vfs_dq_init(inode);
3043         dentry->d_inode = inode;
3044         dentry->d_sb = inode->i_sb;
3045         rc = inode->i_op->removexattr(dentry, name);
3046         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3047         return rc;
3048 }
3049
3050 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3051                            __u64 start, __u64 end)
3052 {
3053         struct osd_object       *obj    = osd_dt_obj(dt);
3054         struct inode            *inode  = obj->oo_inode;
3055         struct osd_thread_info  *info   = osd_oti_get(env);
3056         struct dentry           *dentry = &info->oti_obj_dentry;
3057         struct file             *file   = &info->oti_file;
3058         int                     rc;
3059
3060         ENTRY;
3061
3062         dentry->d_inode = inode;
3063         dentry->d_sb = inode->i_sb;
3064         file->f_path.dentry = dentry;
3065         file->f_mapping = inode->i_mapping;
3066         file->f_op = inode->i_fop;
3067         set_file_inode(file, inode);
3068
3069         rc = ll_vfs_fsync_range(file, start, end, 0);
3070
3071         RETURN(rc);
3072 }
3073
3074 /*
3075  * Index operations.
3076  */
3077
3078 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3079                            const struct dt_index_features *feat)
3080 {
3081         struct iam_descr *descr;
3082
3083         if (osd_object_is_root(o))
3084                 return feat == &dt_directory_features;
3085
3086         LASSERT(o->oo_dir != NULL);
3087
3088         descr = o->oo_dir->od_container.ic_descr;
3089         if (feat == &dt_directory_features) {
3090                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3091                         return 1;
3092                 else
3093                         return 0;
3094         } else {
3095                 return
3096                         feat->dif_keysize_min <= descr->id_key_size &&
3097                         descr->id_key_size <= feat->dif_keysize_max &&
3098                         feat->dif_recsize_min <= descr->id_rec_size &&
3099                         descr->id_rec_size <= feat->dif_recsize_max &&
3100                         !(feat->dif_flags & (DT_IND_VARKEY |
3101                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3102                         ergo(feat->dif_flags & DT_IND_UPDATE,
3103                              1 /* XXX check that object (and file system) is
3104                                 * writable */);
3105         }
3106 }
3107
3108 static int osd_iam_container_init(const struct lu_env *env,
3109                                   struct osd_object *obj,
3110                                   struct osd_directory *dir)
3111 {
3112         struct iam_container *bag = &dir->od_container;
3113         int result;
3114
3115         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3116         if (result != 0)
3117                 return result;
3118
3119         result = iam_container_setup(bag);
3120         if (result == 0)
3121                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3122         else
3123                 iam_container_fini(bag);
3124
3125         return result;
3126 }
3127
3128
3129 /*
3130  * Concurrency: no external locking is necessary.
3131  */
3132 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3133                          const struct dt_index_features *feat)
3134 {
3135         int                      result;
3136         int                      skip_iam = 0;
3137         struct osd_object       *obj = osd_dt_obj(dt);
3138
3139         LINVRNT(osd_invariant(obj));
3140
3141         if (osd_object_is_root(obj)) {
3142                 dt->do_index_ops = &osd_index_ea_ops;
3143                 result = 0;
3144         } else if (feat == &dt_directory_features) {
3145                 dt->do_index_ops = &osd_index_ea_ops;
3146                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3147                         result = 0;
3148                 else
3149                         result = -ENOTDIR;
3150                 skip_iam = 1;
3151         } else if (unlikely(feat == &dt_otable_features)) {
3152                 dt->do_index_ops = &osd_otable_ops;
3153                 return 0;
3154         } else if (unlikely(feat == &dt_acct_features)) {
3155                 dt->do_index_ops = &osd_acct_index_ops;
3156                 result = 0;
3157                 skip_iam = 1;
3158         } else if (!osd_has_index(obj)) {
3159                 struct osd_directory *dir;
3160
3161                 OBD_ALLOC_PTR(dir);
3162                 if (dir != NULL) {
3163
3164                         spin_lock(&obj->oo_guard);
3165                         if (obj->oo_dir == NULL)
3166                                 obj->oo_dir = dir;
3167                         else
3168                                 /*
3169                                  * Concurrent thread allocated container data.
3170                                  */
3171                                 OBD_FREE_PTR(dir);
3172                         spin_unlock(&obj->oo_guard);
3173                         /*
3174                          * Now, that we have container data, serialize its
3175                          * initialization.
3176                          */
3177                         down_write(&obj->oo_ext_idx_sem);
3178                         /*
3179                          * recheck under lock.
3180                          */
3181                         if (!osd_has_index(obj))
3182                                 result = osd_iam_container_init(env, obj,
3183                                                                 obj->oo_dir);
3184                         else
3185                                 result = 0;
3186                         up_write(&obj->oo_ext_idx_sem);
3187                 } else {
3188                         result = -ENOMEM;
3189                 }
3190         } else {
3191                 result = 0;
3192         }
3193
3194         if (result == 0 && skip_iam == 0) {
3195                 if (!osd_iam_index_probe(env, obj, feat))
3196                         result = -ENOTDIR;
3197         }
3198         LINVRNT(osd_invariant(obj));
3199
3200         if (result == 0 && feat == &dt_quota_glb_features &&
3201             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3202                 result = osd_quota_migration(env, dt);
3203
3204         return result;
3205 }
3206
3207 static int osd_otable_it_attr_get(const struct lu_env *env,
3208                                  struct dt_object *dt,
3209                                  struct lu_attr *attr)
3210 {
3211         attr->la_valid = 0;
3212         return 0;
3213 }
3214
3215 static const struct dt_object_operations osd_obj_ops = {
3216         .do_read_lock         = osd_object_read_lock,
3217         .do_write_lock        = osd_object_write_lock,
3218         .do_read_unlock       = osd_object_read_unlock,
3219         .do_write_unlock      = osd_object_write_unlock,
3220         .do_write_locked      = osd_object_write_locked,
3221         .do_attr_get          = osd_attr_get,
3222         .do_declare_attr_set  = osd_declare_attr_set,
3223         .do_attr_set          = osd_attr_set,
3224         .do_ah_init           = osd_ah_init,
3225         .do_declare_create    = osd_declare_object_create,
3226         .do_create            = osd_object_create,
3227         .do_declare_destroy   = osd_declare_object_destroy,
3228         .do_destroy           = osd_object_destroy,
3229         .do_index_try         = osd_index_try,
3230         .do_declare_ref_add   = osd_declare_object_ref_add,
3231         .do_ref_add           = osd_object_ref_add,
3232         .do_declare_ref_del   = osd_declare_object_ref_del,
3233         .do_ref_del           = osd_object_ref_del,
3234         .do_xattr_get         = osd_xattr_get,
3235         .do_declare_xattr_set = osd_declare_xattr_set,
3236         .do_xattr_set         = osd_xattr_set,
3237         .do_declare_xattr_del = osd_declare_xattr_del,
3238         .do_xattr_del         = osd_xattr_del,
3239         .do_xattr_list        = osd_xattr_list,
3240         .do_object_sync       = osd_object_sync,
3241 };
3242
3243 /**
3244  * dt_object_operations for interoperability mode
3245  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3246  */
3247 static const struct dt_object_operations osd_obj_ea_ops = {
3248         .do_read_lock         = osd_object_read_lock,
3249         .do_write_lock        = osd_object_write_lock,
3250         .do_read_unlock       = osd_object_read_unlock,
3251         .do_write_unlock      = osd_object_write_unlock,
3252         .do_write_locked      = osd_object_write_locked,
3253         .do_attr_get          = osd_attr_get,
3254         .do_declare_attr_set  = osd_declare_attr_set,
3255         .do_attr_set          = osd_attr_set,
3256         .do_ah_init           = osd_ah_init,
3257         .do_declare_create    = osd_declare_object_create,
3258         .do_create            = osd_object_ea_create,
3259         .do_declare_destroy   = osd_declare_object_destroy,
3260         .do_destroy           = osd_object_destroy,
3261         .do_index_try         = osd_index_try,
3262         .do_declare_ref_add   = osd_declare_object_ref_add,
3263         .do_ref_add           = osd_object_ref_add,
3264         .do_declare_ref_del   = osd_declare_object_ref_del,
3265         .do_ref_del           = osd_object_ref_del,
3266         .do_xattr_get         = osd_xattr_get,
3267         .do_declare_xattr_set = osd_declare_xattr_set,
3268         .do_xattr_set         = osd_xattr_set,
3269         .do_declare_xattr_del = osd_declare_xattr_del,
3270         .do_xattr_del         = osd_xattr_del,
3271         .do_xattr_list        = osd_xattr_list,
3272         .do_object_sync       = osd_object_sync,
3273 };
3274
3275 static const struct dt_object_operations osd_obj_otable_it_ops = {
3276         .do_attr_get    = osd_otable_it_attr_get,
3277         .do_index_try   = osd_index_try,
3278 };
3279
3280 static int osd_index_declare_iam_delete(const struct lu_env *env,
3281                                         struct dt_object *dt,
3282                                         const struct dt_key *key,
3283                                         struct thandle *handle)
3284 {
3285         struct osd_thandle    *oh;
3286
3287         oh = container_of0(handle, struct osd_thandle, ot_super);
3288         LASSERT(oh->ot_handle == NULL);
3289
3290         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3291                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3292
3293         return 0;
3294 }
3295
3296 /**
3297  *      delete a (key, value) pair from index \a dt specified by \a key
3298  *
3299  *      \param  dt      osd index object
3300  *      \param  key     key for index
3301  *      \param  rec     record reference
3302  *      \param  handle  transaction handler
3303  *
3304  *      \retval  0  success
3305  *      \retval -ve   failure
3306  */
3307 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3308                                 const struct dt_key *key,
3309                                 struct thandle *handle)
3310 {
3311         struct osd_thread_info *oti = osd_oti_get(env);
3312         struct osd_object      *obj = osd_dt_obj(dt);
3313         struct osd_thandle     *oh;
3314         struct iam_path_descr  *ipd;
3315         struct iam_container   *bag = &obj->oo_dir->od_container;
3316         int                     rc;
3317         ENTRY;
3318
3319         if (!dt_object_exists(dt))
3320                 RETURN(-ENOENT);
3321
3322         LINVRNT(osd_invariant(obj));
3323         LASSERT(!dt_object_remote(dt));
3324         LASSERT(bag->ic_object == obj->oo_inode);
3325         LASSERT(handle != NULL);
3326
3327         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3328
3329         ipd = osd_idx_ipd_get(env, bag);
3330         if (unlikely(ipd == NULL))
3331                 RETURN(-ENOMEM);
3332
3333         oh = container_of0(handle, struct osd_thandle, ot_super);
3334         LASSERT(oh->ot_handle != NULL);
3335         LASSERT(oh->ot_handle->h_transaction != NULL);
3336
3337         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3338                 /* swab quota uid/gid provided by caller */
3339                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3340                 key = (const struct dt_key *)&oti->oti_quota_id;
3341         }
3342
3343         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3344         osd_ipd_put(env, bag, ipd);
3345         LINVRNT(osd_invariant(obj));
3346         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3347         RETURN(rc);
3348 }
3349
3350 static int osd_index_declare_ea_delete(const struct lu_env *env,
3351                                        struct dt_object *dt,
3352                                        const struct dt_key *key,
3353                                        struct thandle *handle)
3354 {
3355         struct osd_thandle *oh;
3356         struct inode       *inode;
3357         int                 rc;
3358         ENTRY;
3359
3360         LASSERT(!dt_object_remote(dt));
3361         LASSERT(handle != NULL);
3362
3363         oh = container_of0(handle, struct osd_thandle, ot_super);
3364         LASSERT(oh->ot_handle == NULL);
3365
3366         /* due to DNE we may need to remove an agent inode */
3367         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3368                              osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3369                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3370
3371         inode = osd_dt_obj(dt)->oo_inode;
3372         LASSERT(inode);
3373
3374         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3375                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3376         RETURN(rc);
3377 }
3378
3379 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3380                                           struct dt_rec *fid)
3381 {
3382         struct osd_fid_pack *rec;
3383         int                  rc = -ENODATA;
3384
3385         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3386                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3387                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3388                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3389                         rc = -EINVAL;
3390         }
3391         return rc;
3392 }
3393
3394 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3395                           const struct lu_fid *fid)
3396 {
3397         struct seq_server_site  *ss = osd_seq_site(osd);
3398         ENTRY;
3399
3400         /* FID seqs not in FLDB, must be local seq */
3401         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3402                 RETURN(0);
3403
3404         /* If FLD is not being initialized yet, it only happens during the
3405          * initialization, likely during mgs initialization, and we assume
3406          * this is local FID. */
3407         if (ss == NULL || ss->ss_server_fld == NULL)
3408                 RETURN(0);
3409
3410         /* Only check the local FLDB here */
3411         if (osd_seq_exists(env, osd, fid_seq(fid)))
3412                 RETURN(0);
3413
3414         RETURN(1);
3415 }
3416
3417 /**
3418  * Index delete function for interoperability mode (b11826).
3419  * It will remove the directory entry added by osd_index_ea_insert().
3420  * This entry is needed to maintain name->fid mapping.
3421  *
3422  * \param key,  key i.e. file entry to be deleted
3423  *
3424  * \retval   0, on success
3425  * \retval -ve, on error
3426  */
3427 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3428                                const struct dt_key *key, struct thandle *handle)
3429 {
3430         struct osd_object          *obj = osd_dt_obj(dt);
3431         struct inode               *dir = obj->oo_inode;
3432         struct dentry              *dentry;
3433         struct osd_thandle         *oh;
3434         struct ldiskfs_dir_entry_2 *de = NULL;
3435         struct buffer_head         *bh;
3436         struct htree_lock          *hlock = NULL;
3437         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3438         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3439         int                        rc;
3440         ENTRY;
3441
3442         if (!dt_object_exists(dt))
3443                 RETURN(-ENOENT);
3444
3445         LINVRNT(osd_invariant(obj));
3446         LASSERT(!dt_object_remote(dt));
3447         LASSERT(handle != NULL);
3448
3449         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3450
3451         oh = container_of(handle, struct osd_thandle, ot_super);
3452         LASSERT(oh->ot_handle != NULL);
3453         LASSERT(oh->ot_handle->h_transaction != NULL);
3454
3455         ll_vfs_dq_init(dir);
3456         dentry = osd_child_dentry_get(env, obj,
3457                                       (char *)key, strlen((char *)key));
3458
3459         if (obj->oo_hl_head != NULL) {
3460                 hlock = osd_oti_get(env)->oti_hlock;
3461                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3462                                    dir, LDISKFS_HLOCK_DEL);
3463         } else {
3464                 down_write(&obj->oo_ext_idx_sem);
3465         }
3466
3467         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3468         if (bh) {
3469                 __u32 ino = 0;
3470
3471                 /* If this is not the ".." entry, it might be a remote DNE
3472                  * entry and  we need to check if the FID is for a remote
3473                  * MDT.  If the FID is  not in the directory entry (e.g.
3474                  * upgraded 1.8 filesystem without dirdata enabled) then
3475                  * we need to get the FID from the LMA. For a remote directory
3476                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3477                  * case.
3478                  *
3479                  * Delete the entry before the agent inode in order to
3480                  * simplify error handling.  At worst an error after deleting
3481                  * the entry first might leak the agent inode afterward. The
3482                  * reverse would need filesystem abort in case of error deleting
3483                  * the entry after the agent had been removed, or leave a
3484                  * dangling entry pointing at a random inode. */
3485                 if (strcmp((char *)key, dotdot) != 0) {
3486                         LASSERT(de != NULL);
3487                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3488                         /* If Fid is not in dentry, try to get it from LMA */
3489                         if (rc == -ENODATA) {
3490                                 struct osd_inode_id *id;
3491                                 struct inode *inode;
3492
3493                                 /* Before trying to get fid from the inode,
3494                                  * check whether the inode is valid.
3495                                  *
3496                                  * If the inode has been deleted, do not go
3497                                  * ahead to do osd_ea_fid_get, which will set
3498                                  * the inode to bad inode, which might cause
3499                                  * the inode to be deleted uncorrectly */
3500                                 inode = ldiskfs_iget(osd_sb(osd),
3501                                                      le32_to_cpu(de->inode));
3502                                 if (IS_ERR(inode)) {
3503                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3504                                                "error.\n", osd_name(osd),
3505                                                PFID(fid));
3506                                         rc = PTR_ERR(inode);
3507                                 } else {
3508                                         if (likely(inode->i_nlink != 0)) {
3509                                                 id = &osd_oti_get(env)->oti_id;
3510                                                 rc = osd_ea_fid_get(env, obj,
3511                                                         le32_to_cpu(de->inode),
3512                                                                     fid, id);
3513                                         } else {
3514                                                 CDEBUG(D_INFO, "%s: %u "DFID
3515                                                        "deleted.\n",
3516                                                        osd_name(osd),
3517                                                        le32_to_cpu(de->inode),
3518                                                        PFID(fid));
3519                                                 rc = -ESTALE;
3520                                         }
3521                                         iput(inode);
3522                                 }
3523                         }
3524                         if (rc == 0 &&
3525                             unlikely(osd_remote_fid(env, osd, fid)))
3526                                 /* Need to delete agent inode */
3527                                 ino = le32_to_cpu(de->inode);
3528                 }
3529                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3530                 brelse(bh);
3531                 if (rc == 0 && unlikely(ino != 0)) {
3532                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3533                                                           oh);
3534                         if (rc != 0)
3535                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3536                                        osd_name(osd), PFID(fid), rc);
3537                 }
3538         } else {
3539                 rc = -ENOENT;
3540         }
3541         if (hlock != NULL)
3542                 ldiskfs_htree_unlock(hlock);
3543         else
3544                 up_write(&obj->oo_ext_idx_sem);
3545
3546         if (rc != 0)
3547                 GOTO(out, rc);
3548
3549         /* For inode on the remote MDT, .. will point to
3550          * /Agent directory, Check whether it needs to delete
3551          * from agent directory */
3552         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3553                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3554                                                    oh);
3555                 if (rc != 0 && rc != -ENOENT) {
3556                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3557                                osd_name(osd), PFID(fid), rc);
3558                 }
3559
3560                 if (rc == -ENOENT)
3561                         rc = 0;
3562
3563                 GOTO(out, rc);
3564         }
3565 out:
3566
3567         LASSERT(osd_invariant(obj));
3568         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3569         RETURN(rc);
3570 }
3571
3572 /**
3573  *      Lookup index for \a key and copy record to \a rec.
3574  *
3575  *      \param  dt      osd index object
3576  *      \param  key     key for index
3577  *      \param  rec     record reference
3578  *
3579  *      \retval  +ve  success : exact mach
3580  *      \retval  0    return record with key not greater than \a key
3581  *      \retval -ve   failure
3582  */
3583 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3584                                 struct dt_rec *rec, const struct dt_key *key)
3585 {
3586         struct osd_object      *obj = osd_dt_obj(dt);
3587         struct iam_path_descr  *ipd;
3588         struct iam_container   *bag = &obj->oo_dir->od_container;
3589         struct osd_thread_info *oti = osd_oti_get(env);
3590         struct iam_iterator    *it = &oti->oti_idx_it;
3591         struct iam_rec         *iam_rec;
3592         int                     rc;
3593         ENTRY;
3594
3595         if (!dt_object_exists(dt))
3596                 RETURN(-ENOENT);
3597
3598         LASSERT(osd_invariant(obj));
3599         LASSERT(!dt_object_remote(dt));
3600         LASSERT(bag->ic_object == obj->oo_inode);
3601
3602         ipd = osd_idx_ipd_get(env, bag);
3603         if (IS_ERR(ipd))
3604                 RETURN(-ENOMEM);
3605
3606         /* got ipd now we can start iterator. */
3607         iam_it_init(it, bag, 0, ipd);
3608
3609         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3610                 /* swab quota uid/gid provided by caller */
3611                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3612                 key = (const struct dt_key *)&oti->oti_quota_id;
3613         }
3614
3615         rc = iam_it_get(it, (struct iam_key *)key);
3616         if (rc >= 0) {
3617                 if (S_ISDIR(obj->oo_inode->i_mode))
3618                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3619                 else
3620                         iam_rec = (struct iam_rec *) rec;
3621
3622                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3623
3624                 if (S_ISDIR(obj->oo_inode->i_mode))
3625                         osd_fid_unpack((struct lu_fid *) rec,
3626                                        (struct osd_fid_pack *)iam_rec);
3627                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3628                         osd_quota_unpack(obj, rec);
3629         }
3630
3631         iam_it_put(it);
3632         iam_it_fini(it);
3633         osd_ipd_put(env, bag, ipd);
3634
3635         LINVRNT(osd_invariant(obj));
3636
3637         RETURN(rc);
3638 }
3639
3640 static int osd_index_declare_iam_insert(const struct lu_env *env,
3641                                         struct dt_object *dt,
3642                                         const struct dt_rec *rec,
3643                                         const struct dt_key *key,
3644                                         struct thandle *handle)
3645 {
3646         struct osd_thandle *oh;
3647
3648         LASSERT(handle != NULL);
3649
3650         oh = container_of0(handle, struct osd_thandle, ot_super);
3651         LASSERT(oh->ot_handle == NULL);
3652
3653         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3654                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3655
3656         return 0;
3657 }
3658
3659 /**
3660  *      Inserts (key, value) pair in \a dt index object.
3661  *
3662  *      \param  dt      osd index object
3663  *      \param  key     key for index
3664  *      \param  rec     record reference
3665  *      \param  th      transaction handler
3666  *
3667  *      \retval  0  success
3668  *      \retval -ve failure
3669  */
3670 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3671                                 const struct dt_rec *rec,
3672                                 const struct dt_key *key, struct thandle *th,
3673                                 int ignore_quota)
3674 {
3675         struct osd_object     *obj = osd_dt_obj(dt);
3676         struct iam_path_descr *ipd;
3677         struct osd_thandle    *oh;
3678         struct iam_container  *bag;
3679         struct osd_thread_info *oti = osd_oti_get(env);
3680         struct iam_rec         *iam_rec;
3681         int                     rc;
3682         ENTRY;
3683
3684         if (!dt_object_exists(dt))
3685                 RETURN(-ENOENT);
3686
3687         LINVRNT(osd_invariant(obj));
3688         LASSERT(!dt_object_remote(dt));
3689
3690         bag = &obj->oo_dir->od_container;
3691         LASSERT(bag->ic_object == obj->oo_inode);
3692         LASSERT(th != NULL);
3693
3694         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3695
3696         ipd = osd_idx_ipd_get(env, bag);
3697         if (unlikely(ipd == NULL))
3698                 RETURN(-ENOMEM);
3699
3700         oh = container_of0(th, struct osd_thandle, ot_super);
3701         LASSERT(oh->ot_handle != NULL);
3702         LASSERT(oh->ot_handle->h_transaction != NULL);
3703         if (S_ISDIR(obj->oo_inode->i_mode)) {
3704                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3705                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3706         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3707                 /* pack quota uid/gid */
3708                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3709                 key = (const struct dt_key *)&oti->oti_quota_id;
3710                 /* pack quota record */
3711                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3712                 iam_rec = (struct iam_rec *)rec;
3713         } else {
3714                 iam_rec = (struct iam_rec *)rec;
3715         }
3716
3717         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3718                         iam_rec, ipd);
3719         osd_ipd_put(env, bag, ipd);
3720         LINVRNT(osd_invariant(obj));
3721         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3722         RETURN(rc);
3723 }
3724
3725 /**
3726  * Calls ldiskfs_add_entry() to add directory entry
3727  * into the directory. This is required for
3728  * interoperability mode (b11826)
3729  *
3730  * \retval   0, on success
3731  * \retval -ve, on error
3732  */
3733 static int __osd_ea_add_rec(struct osd_thread_info *info,
3734                             struct osd_object *pobj, struct inode  *cinode,
3735                             const char *name, const struct lu_fid *fid,
3736                             struct htree_lock *hlock, struct thandle *th)
3737 {
3738         struct ldiskfs_dentry_param *ldp;
3739         struct dentry               *child;
3740         struct osd_thandle          *oth;
3741         int                          rc;
3742
3743         oth = container_of(th, struct osd_thandle, ot_super);
3744         LASSERT(oth->ot_handle != NULL);
3745         LASSERT(oth->ot_handle->h_transaction != NULL);
3746         LASSERT(pobj->oo_inode);
3747
3748         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3749         if (unlikely(pobj->oo_inode ==
3750                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3751                 ldp->edp_magic = 0;
3752         else
3753                 osd_get_ldiskfs_dirent_param(ldp, fid);
3754         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3755         child->d_fsdata = (void *)ldp;
3756         ll_vfs_dq_init(pobj->oo_inode);
3757         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3758         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3759                 struct ldiskfs_dir_entry_2      *de;
3760                 struct buffer_head              *bh;
3761                 int                              rc1;
3762
3763                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3764                                             NULL, hlock);
3765                 if (bh != NULL) {
3766                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3767                                                                bh);
3768                         if (rc1 == 0) {
3769                                 if (S_ISDIR(cinode->i_mode))
3770                                         de->file_type = LDISKFS_DIRENT_LUFID |
3771                                                         LDISKFS_FT_REG_FILE;
3772                                 else
3773                                         de->file_type = LDISKFS_DIRENT_LUFID |
3774                                                         LDISKFS_FT_DIR;
3775                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3776                                                               NULL, bh);
3777                                 brelse(bh);
3778                         }
3779                 }
3780         }
3781
3782         RETURN(rc);
3783 }
3784
3785 /**
3786  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3787  * into the directory.Also sets flags into osd object to
3788  * indicate dot and dotdot are created. This is required for
3789  * interoperability mode (b11826)
3790  *
3791  * \param dir   directory for dot and dotdot fixup.
3792  * \param obj   child object for linking
3793  *
3794  * \retval   0, on success
3795  * \retval -ve, on error
3796  */
3797 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3798                               struct osd_object *dir,
3799                               struct inode *parent_dir, const char *name,
3800                               const struct lu_fid *dot_fid,
3801                               const struct lu_fid *dot_dot_fid,
3802                               struct thandle *th)
3803 {
3804         struct inode                *inode = dir->oo_inode;
3805         struct osd_thandle          *oth;
3806         int result = 0;
3807
3808         oth = container_of(th, struct osd_thandle, ot_super);
3809         LASSERT(oth->ot_handle->h_transaction != NULL);
3810         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3811
3812         if (strcmp(name, dot) == 0) {
3813                 if (dir->oo_compat_dot_created) {
3814                         result = -EEXIST;
3815                 } else {
3816                         LASSERT(inode == parent_dir);
3817                         dir->oo_compat_dot_created = 1;
3818                         result = 0;
3819                 }
3820         } else if (strcmp(name, dotdot) == 0) {
3821                 if (!dir->oo_compat_dot_created)
3822                         return -EINVAL;
3823                 /* in case of rename, dotdot is already created */
3824                 if (dir->oo_compat_dotdot_created) {
3825                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3826                                                 dot_dot_fid, NULL, th);
3827                 }
3828
3829                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
3830                         struct lu_fid tfid = *dot_dot_fid;
3831
3832                         tfid.f_oid--;
3833                         result = osd_add_dot_dotdot_internal(info,
3834                                         dir->oo_inode, parent_dir, dot_fid,
3835                                         &tfid, oth);
3836                 } else {
3837                         result = osd_add_dot_dotdot_internal(info,
3838                                         dir->oo_inode, parent_dir, dot_fid,
3839                                         dot_dot_fid, oth);
3840                 }
3841
3842                 if (result == 0)
3843                         dir->oo_compat_dotdot_created = 1;
3844         }
3845
3846         return result;
3847 }
3848
3849
3850 /**
3851  * It will call the appropriate osd_add* function and return the
3852  * value, return by respective functions.
3853  */
3854 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3855                           struct inode *cinode, const char *name,
3856                           const struct lu_fid *fid, struct thandle *th)
3857 {
3858         struct osd_thread_info *info   = osd_oti_get(env);
3859         struct htree_lock      *hlock;
3860         int                     rc;
3861
3862         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3863
3864         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3865                                                    name[2] =='\0'))) {
3866                 if (hlock != NULL) {
3867                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3868                                            pobj->oo_inode, 0);
3869                 } else {
3870                         down_write(&pobj->oo_ext_idx_sem);
3871                 }
3872
3873                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3874                                         lu_object_fid(&pobj->oo_dt.do_lu),
3875                                         fid, th);
3876         } else {
3877                 if (hlock != NULL) {
3878                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3879                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3880                 } else {
3881                         down_write(&pobj->oo_ext_idx_sem);
3882                 }
3883
3884                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3885                         struct lu_fid *tfid = &info->oti_fid;
3886
3887                         *tfid = *fid;
3888                         tfid->f_ver = ~0;
3889                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3890                                               tfid, hlock, th);
3891                 } else {
3892                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3893                                               hlock, th);
3894                 }
3895         }
3896         if (hlock != NULL)
3897                 ldiskfs_htree_unlock(hlock);
3898         else
3899                 up_write(&pobj->oo_ext_idx_sem);
3900
3901         return rc;
3902 }
3903
3904 static void
3905 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3906                       struct osd_idmap_cache *oic)
3907 {
3908         struct osd_scrub    *scrub = &dev->od_scrub;
3909         struct lu_fid       *fid   = &oic->oic_fid;
3910         struct osd_inode_id *id    = &oti->oti_id;
3911         int                  once  = 0;
3912         int                  rc;
3913         ENTRY;
3914
3915         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3916                 RETURN_EXIT;
3917
3918         if (scrub->os_pos_current > id->oii_ino)
3919                 RETURN_EXIT;
3920
3921 again:
3922         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
3923         if (rc != 0 && rc != -ENOENT)
3924                 RETURN_EXIT;
3925
3926         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3927                 RETURN_EXIT;
3928
3929         if (thread_is_running(&scrub->os_thread)) {
3930                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3931                 /* There is race condition between osd_oi_lookup and OI scrub.
3932                  * The OI scrub finished just after osd_oi_lookup() failure.
3933                  * Under such case, it is unnecessary to trigger OI scrub again,
3934                  * but try to call osd_oi_lookup() again. */
3935                 if (unlikely(rc == -EAGAIN))
3936                         goto again;
3937
3938                 RETURN_EXIT;
3939         }
3940
3941         if (!dev->od_noscrub && ++once == 1) {
3942                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
3943                                      SS_CLEAR_FAILOUT);
3944                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
3945                        "for "DFID", rc = %d [2]\n",
3946                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3947                        PFID(fid), rc);
3948                 if (rc == 0 || rc == -EALREADY)
3949                         goto again;
3950         }
3951
3952         EXIT;
3953 }
3954
3955 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3956                                struct osd_device *dev,
3957                                struct osd_idmap_cache *oic,
3958                                struct lu_fid *fid, __u32 ino)
3959 {
3960         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3961         struct inode            *inode;
3962         int                      rc;
3963
3964         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3965         inode = osd_iget(oti, dev, &oic->oic_lid);
3966         if (IS_ERR(inode)) {
3967                 fid_zero(&oic->oic_fid);
3968                 return PTR_ERR(inode);
3969         }
3970
3971         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3972         iput(inode);
3973         if (rc != 0)
3974                 fid_zero(&oic->oic_fid);
3975         else
3976                 *fid = oic->oic_fid = lma->lma_self_fid;
3977         return rc;
3978 }
3979
3980 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
3981                       struct osd_inode_id *id, const struct lu_fid *fid)
3982 {
3983         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
3984                id->oii_ino, id->oii_gen, info);
3985         info->oti_cache.oic_lid = *id;
3986         info->oti_cache.oic_fid = *fid;
3987         info->oti_cache.oic_dev = osd;
3988 }
3989
3990 /**
3991  * Get parent FID from the linkEA.
3992  *
3993  * For a directory which parent resides on remote MDT, to satisfy the
3994  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
3995  * the other hand, to make the lookup(..) on the directory can return
3996  * the real parent FID, we append the real parent FID after its ".."
3997  * name entry in the /REMOTE_PARENT_DIR.
3998  *
3999  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4000  * backup. So after the restore, we cannot get the right parent FID from
4001  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4002  * we have stored the real parent FID in the directory object's linkEA,
4003  * we can parse the linkEA for the real parent FID.
4004  *
4005  * \param[in] env       pointer to the thread context
4006  * \param[in] obj       pointer to the object to be handled
4007  * \param[out]fid       pointer to the buffer to hold the parent FID
4008  *
4009  * \retval              0 for getting the real parent FID successfully
4010  * \retval              negative error number on failure
4011  */
4012 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4013                                     struct osd_object *obj,
4014                                     struct lu_fid *fid)
4015 {
4016         struct osd_thread_info  *oti    = osd_oti_get(env);
4017         struct lu_buf           *buf    = &oti->oti_big_buf;
4018         struct dentry           *dentry = &oti->oti_obj_dentry;
4019         struct inode            *inode  = obj->oo_inode;
4020         struct linkea_data       ldata  = { NULL };
4021         int                      rc;
4022         ENTRY;
4023
4024         fid_zero(fid);
4025         if (!S_ISDIR(inode->i_mode))
4026                 RETURN(-EIO);
4027
4028 again:
4029         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4030                              buf->lb_buf, buf->lb_len);
4031         if (rc == -ERANGE) {
4032                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4033                                      NULL, 0);
4034                 if (rc > 0) {
4035                         lu_buf_realloc(buf, rc);
4036                         if (buf->lb_buf == NULL)
4037                                 RETURN(-ENOMEM);
4038
4039                         goto again;
4040                 }
4041         }
4042
4043         if (unlikely(rc == 0))
4044                 RETURN(-ENODATA);
4045
4046         if (rc < 0)
4047                 RETURN(rc);
4048
4049         if (unlikely(buf->lb_buf == NULL)) {
4050                 lu_buf_realloc(buf, rc);
4051                 if (buf->lb_buf == NULL)
4052                         RETURN(-ENOMEM);
4053
4054                 goto again;
4055         }
4056
4057         ldata.ld_buf = buf;
4058         rc = linkea_init(&ldata);
4059         if (rc == 0) {
4060                 linkea_first_entry(&ldata);
4061                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4062         }
4063
4064         RETURN(rc);
4065 }
4066
4067 /**
4068  * Calls ->lookup() to find dentry. From dentry get inode and
4069  * read inode's ea to get fid. This is required for  interoperability
4070  * mode (b11826)
4071  *
4072  * \retval   0, on success
4073  * \retval -ve, on error
4074  */
4075 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4076                              struct dt_rec *rec, const struct dt_key *key)
4077 {
4078         struct inode                    *dir    = obj->oo_inode;
4079         struct dentry                   *dentry;
4080         struct ldiskfs_dir_entry_2      *de;
4081         struct buffer_head              *bh;
4082         struct lu_fid                   *fid = (struct lu_fid *) rec;
4083         struct htree_lock               *hlock = NULL;
4084         int                             ino;
4085         int                             rc;
4086         ENTRY;
4087
4088         LASSERT(dir->i_op != NULL);
4089         LASSERT(dir->i_op->lookup != NULL);
4090
4091         dentry = osd_child_dentry_get(env, obj,
4092                                       (char *)key, strlen((char *)key));
4093
4094         if (obj->oo_hl_head != NULL) {
4095                 hlock = osd_oti_get(env)->oti_hlock;
4096                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4097                                    dir, LDISKFS_HLOCK_LOOKUP);
4098         } else {
4099                 down_read(&obj->oo_ext_idx_sem);
4100         }
4101
4102         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4103         if (bh) {
4104                 struct osd_thread_info *oti = osd_oti_get(env);
4105                 struct osd_inode_id *id = &oti->oti_id;
4106                 struct osd_idmap_cache *oic = &oti->oti_cache;
4107                 struct osd_device *dev = osd_obj2dev(obj);
4108
4109                 ino = le32_to_cpu(de->inode);
4110                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4111                         brelse(bh);
4112                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4113                         GOTO(out, rc);
4114                 }
4115
4116                 rc = osd_get_fid_from_dentry(de, rec);
4117
4118                 /* done with de, release bh */
4119                 brelse(bh);
4120                 if (rc != 0) {
4121                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
4122                                 const char *name = (const char *)key;
4123
4124                                 /* If the parent is on remote MDT, and there
4125                                  * is no FID-in-dirent, then we have to get
4126                                  * the parent FID from the linkEA.  */
4127                                 if (likely(strlen(name) == 2 &&
4128                                            name[0] == '.' && name[1] == '.'))
4129                                         rc = osd_get_pfid_from_linkea(env, obj,
4130                                                                       fid);
4131                         } else {
4132                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4133                         }
4134                 } else {
4135                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4136                 }
4137
4138                 if (rc != 0) {
4139                         fid_zero(&oic->oic_fid);
4140                         GOTO(out, rc);
4141                 }
4142
4143                 if (osd_remote_fid(env, dev, fid))
4144                         GOTO(out, rc = 0);
4145
4146                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4147                 osd_consistency_check(oti, dev, oic);
4148         } else {
4149                 rc = -ENOENT;
4150         }
4151
4152         GOTO(out, rc);
4153
4154 out:
4155         if (hlock != NULL)
4156                 ldiskfs_htree_unlock(hlock);
4157         else
4158                 up_read(&obj->oo_ext_idx_sem);
4159         return rc;
4160 }
4161
4162 /**
4163  * Find the osd object for given fid.
4164  *
4165  * \param fid need to find the osd object having this fid
4166  *
4167  * \retval osd_object on success
4168  * \retval        -ve on error
4169  */
4170 static struct osd_object *osd_object_find(const struct lu_env *env,
4171                                           struct dt_object *dt,
4172                                           const struct lu_fid *fid)
4173 {
4174         struct lu_device  *ludev = dt->do_lu.lo_dev;
4175         struct osd_object *child = NULL;
4176         struct lu_object  *luch;
4177         struct lu_object  *lo;
4178
4179         /*
4180          * at this point topdev might not exist yet
4181          * (i.e. MGS is preparing profiles). so we can
4182          * not rely on topdev and instead lookup with
4183          * our device passed as topdev. this can't work
4184          * if the object isn't cached yet (as osd doesn't
4185          * allocate lu_header). IOW, the object must be
4186          * in the cache, otherwise lu_object_alloc() crashes
4187          * -bzzz
4188          */
4189         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4190                                  ludev : ludev->ld_site->ls_top_dev,
4191                                  fid, NULL);
4192         if (!IS_ERR(luch)) {
4193                 if (lu_object_exists(luch)) {
4194                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4195                         if (lo != NULL)
4196                                 child = osd_obj(lo);
4197                         else
4198                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4199                                                 "lu_object can't be located"
4200                                                 DFID"\n", PFID(fid));
4201
4202                         if (child == NULL) {
4203                                 lu_object_put(env, luch);
4204                                 CERROR("Unable to get osd_object\n");
4205                                 child = ERR_PTR(-ENOENT);
4206                         }
4207                 } else {
4208                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4209                                         "lu_object does not exists "DFID"\n",
4210                                         PFID(fid));
4211                         lu_object_put(env, luch);
4212                         child = ERR_PTR(-ENOENT);
4213                 }
4214         } else {
4215                 child = ERR_CAST(luch);
4216         }
4217
4218         return child;
4219 }
4220
4221 /**
4222  * Put the osd object once done with it.
4223  *
4224  * \param obj osd object that needs to be put
4225  */
4226 static inline void osd_object_put(const struct lu_env *env,
4227                                   struct osd_object *obj)
4228 {
4229         lu_object_put(env, &obj->oo_dt.do_lu);
4230 }
4231
4232 static int osd_index_declare_ea_insert(const struct lu_env *env,
4233                                        struct dt_object *dt,
4234                                        const struct dt_rec *rec,
4235                                        const struct dt_key *key,
4236                                        struct thandle *handle)
4237 {
4238         struct osd_thandle      *oh;
4239         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4240         struct lu_fid           *fid = (struct lu_fid *)rec;
4241         int                      credits, rc = 0;
4242         ENTRY;
4243
4244         LASSERT(!dt_object_remote(dt));
4245         LASSERT(handle != NULL);
4246
4247         oh = container_of0(handle, struct osd_thandle, ot_super);
4248         LASSERT(oh->ot_handle == NULL);
4249
4250         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4251         if (fid != NULL) {
4252                 rc = osd_remote_fid(env, osd, fid);
4253                 if (unlikely(rc < 0))
4254                         RETURN(rc);
4255                 if (rc > 0) {
4256                         /* a reference to remote inode is represented by an
4257                          * agent inode which we have to create */
4258                         credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4259                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4260                 }
4261                 rc = 0;
4262         }
4263
4264         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4265
4266         if (osd_dt_obj(dt)->oo_inode != NULL) {
4267                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4268
4269                 /* We ignore block quota on meta pool (MDTs), so needn't
4270                  * calculate how many blocks will be consumed by this index
4271                  * insert */
4272                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4273                                            i_gid_read(inode), 0, oh,
4274                                            osd_dt_obj(dt), true, NULL, false);
4275         }
4276
4277         RETURN(rc);
4278 }
4279
4280 /**
4281  * Index add function for interoperability mode (b11826).
4282  * It will add the directory entry.This entry is needed to
4283  * maintain name->fid mapping.
4284  *
4285  * \param key it is key i.e. file entry to be inserted
4286  * \param rec it is value of given key i.e. fid
4287  *
4288  * \retval   0, on success
4289  * \retval -ve, on error
4290  */
4291 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4292                                const struct dt_rec *rec,
4293                                const struct dt_key *key, struct thandle *th,
4294                                int ignore_quota)
4295 {
4296         struct osd_object       *obj = osd_dt_obj(dt);
4297         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4298         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4299         const struct lu_fid     *fid    = rec1->rec_fid;
4300         const char              *name = (const char *)key;
4301         struct osd_thread_info  *oti   = osd_oti_get(env);
4302         struct osd_inode_id     *id    = &oti->oti_id;
4303         struct inode            *child_inode = NULL;
4304         struct osd_object       *child = NULL;
4305         int                     rc;
4306         ENTRY;
4307
4308         if (!dt_object_exists(dt))
4309                 RETURN(-ENOENT);
4310
4311         LASSERT(osd_invariant(obj));
4312         LASSERT(!dt_object_remote(dt));
4313         LASSERT(th != NULL);
4314
4315         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4316
4317         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4318
4319         rc = osd_remote_fid(env, osd, fid);
4320         if (rc < 0) {
4321                 CERROR("%s: Can not find object "DFID" rc %d\n",
4322                        osd_name(osd), PFID(fid), rc);
4323                 RETURN(rc);
4324         }
4325
4326         if (rc == 1) {
4327                 /* Insert remote entry */
4328                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4329                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4330                         struct osd_thandle      *oh;
4331
4332                         /* If parent on remote MDT, we need put this object
4333                          * under AGENT */
4334                         oh = container_of(th, typeof(*oh), ot_super);
4335                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4336                         if (rc != 0) {
4337                                 CERROR("%s: add "DFID" error: rc = %d\n",
4338                                        osd_name(osd),
4339                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4340                                 RETURN(rc);
4341                         }
4342
4343                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4344                 } else {
4345                         child_inode = osd_create_local_agent_inode(env, osd,
4346                                         obj, fid, rec1->rec_type & S_IFMT, th);
4347                         if (IS_ERR(child_inode))
4348                                 RETURN(PTR_ERR(child_inode));
4349                 }
4350         } else {
4351                 /* Insert local entry */
4352                 child = osd_object_find(env, dt, fid);
4353                 if (IS_ERR(child)) {
4354                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4355                                osd_name(osd), PFID(fid),
4356                                id->oii_ino, id->oii_gen,
4357                                (int)PTR_ERR(child));
4358                         RETURN(PTR_ERR(child));
4359                 }
4360                 child_inode = igrab(child->oo_inode);
4361         }
4362
4363         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4364
4365         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4366                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4367
4368         iput(child_inode);
4369         if (child != NULL)
4370                 osd_object_put(env, child);
4371         LASSERT(osd_invariant(obj));
4372         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4373         RETURN(rc);
4374 }
4375
4376 /**
4377  *  Initialize osd Iterator for given osd index object.
4378  *
4379  *  \param  dt      osd index object
4380  */
4381
4382 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4383                                      struct dt_object *dt,
4384                                      __u32 unused)
4385 {
4386         struct osd_it_iam      *it;
4387         struct osd_object      *obj = osd_dt_obj(dt);
4388         struct lu_object       *lo  = &dt->do_lu;
4389         struct iam_path_descr  *ipd;
4390         struct iam_container   *bag = &obj->oo_dir->od_container;
4391
4392         if (!dt_object_exists(dt))
4393                 return ERR_PTR(-ENOENT);
4394
4395         OBD_ALLOC_PTR(it);
4396         if (it == NULL)
4397                 return ERR_PTR(-ENOMEM);
4398
4399         ipd = osd_it_ipd_get(env, bag);
4400         if (likely(ipd != NULL)) {
4401                 it->oi_obj = obj;
4402                 it->oi_ipd = ipd;
4403                 lu_object_get(lo);
4404                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4405                 return (struct dt_it *)it;
4406         } else {
4407                 OBD_FREE_PTR(it);
4408                 return ERR_PTR(-ENOMEM);
4409         }
4410 }
4411
4412 /**
4413  * free given Iterator.
4414  */
4415
4416 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4417 {
4418         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4419         struct osd_object       *obj = it->oi_obj;
4420
4421         iam_it_fini(&it->oi_it);
4422         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4423         lu_object_put(env, &obj->oo_dt.do_lu);
4424         OBD_FREE_PTR(it);
4425 }
4426
4427 /**
4428  *  Move Iterator to record specified by \a key
4429  *
4430  *  \param  di      osd iterator
4431  *  \param  key     key for index
4432  *
4433  *  \retval +ve  di points to record with least key not larger than key
4434  *  \retval  0   di points to exact matched key
4435  *  \retval -ve  failure
4436  */
4437
4438 static int osd_it_iam_get(const struct lu_env *env,
4439                           struct dt_it *di, const struct dt_key *key)
4440 {
4441         struct osd_thread_info  *oti = osd_oti_get(env);
4442         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4443
4444         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4445                 /* swab quota uid/gid */
4446                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4447                 key = (struct dt_key *)&oti->oti_quota_id;
4448         }
4449
4450         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4451 }
4452
4453 /**
4454  *  Release Iterator
4455  *
4456  *  \param  di      osd iterator
4457  */
4458 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4459 {
4460         struct osd_it_iam *it = (struct osd_it_iam *)di;
4461
4462         iam_it_put(&it->oi_it);
4463 }
4464
4465 /**
4466  *  Move iterator by one record
4467  *
4468  *  \param  di      osd iterator
4469  *
4470  *  \retval +1   end of container reached
4471  *  \retval  0   success
4472  *  \retval -ve  failure
4473  */
4474
4475 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4476 {
4477         struct osd_it_iam *it = (struct osd_it_iam *)di;
4478
4479         return iam_it_next(&it->oi_it);
4480 }
4481
4482 /**
4483  * Return pointer to the key under iterator.
4484  */
4485
4486 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4487                                  const struct dt_it *di)
4488 {
4489         struct osd_thread_info *oti = osd_oti_get(env);
4490         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4491         struct osd_object      *obj = it->oi_obj;
4492         struct dt_key          *key;
4493
4494         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4495
4496         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4497                 /* swab quota uid/gid */
4498                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4499                 key = (struct dt_key *)&oti->oti_quota_id;
4500         }
4501
4502         return key;
4503 }
4504
4505 /**
4506  * Return size of key under iterator (in bytes)
4507  */
4508
4509 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4510 {
4511         struct osd_it_iam *it = (struct osd_it_iam *)di;
4512
4513         return iam_it_key_size(&it->oi_it);
4514 }
4515
4516 static inline void
4517 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4518 {
4519         /* check if file type is required */
4520         if (ent->lde_attrs & LUDA_TYPE) {
4521                 struct luda_type *lt;
4522                 int align = sizeof(*lt) - 1;
4523
4524                 len = (len + align) & ~align;
4525                 lt = (struct luda_type *)(ent->lde_name + len);
4526                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4527         }
4528
4529         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4530 }
4531
4532 /**
4533  * build lu direct from backend fs dirent.
4534  */
4535
4536 static inline void
4537 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4538                    char *name, __u16 namelen, __u16 type, __u32 attr)
4539 {
4540         ent->lde_attrs = attr | LUDA_FID;
4541         fid_cpu_to_le(&ent->lde_fid, fid);
4542
4543         ent->lde_hash = cpu_to_le64(offset);
4544         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4545
4546         strncpy(ent->lde_name, name, namelen);
4547         ent->lde_name[namelen] = '\0';
4548         ent->lde_namelen = cpu_to_le16(namelen);
4549
4550         /* append lustre attributes */
4551         osd_it_append_attrs(ent, namelen, type);
4552 }
4553
4554 /**
4555  * Return pointer to the record under iterator.
4556  */
4557 static int osd_it_iam_rec(const struct lu_env *env,
4558                           const struct dt_it *di,
4559                           struct dt_rec *dtrec, __u32 attr)
4560 {
4561         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4562         struct osd_thread_info *info = osd_oti_get(env);
4563         ENTRY;
4564
4565         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4566                 const struct osd_fid_pack *rec;
4567                 struct lu_fid             *fid = &info->oti_fid;
4568                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4569                 char                      *name;
4570                 int                        namelen;
4571                 __u64                      hash;
4572                 int                        rc;
4573
4574                 name = (char *)iam_it_key_get(&it->oi_it);
4575                 if (IS_ERR(name))
4576                         RETURN(PTR_ERR(name));
4577
4578                 namelen = iam_it_key_size(&it->oi_it);
4579
4580                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4581                 if (IS_ERR(rec))
4582                         RETURN(PTR_ERR(rec));
4583
4584                 rc = osd_fid_unpack(fid, rec);
4585                 if (rc)
4586                         RETURN(rc);
4587
4588                 hash = iam_it_store(&it->oi_it);
4589
4590                 /* IAM does not store object type in IAM index (dir) */
4591                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4592                                    0, LUDA_FID);
4593         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4594                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4595                            (struct iam_rec *)dtrec);
4596                 osd_quota_unpack(it->oi_obj, dtrec);
4597         } else {
4598                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4599                            (struct iam_rec *)dtrec);
4600         }
4601
4602         RETURN(0);
4603 }
4604
4605 /**
4606  * Returns cookie for current Iterator position.
4607  */
4608 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4609 {
4610         struct osd_it_iam *it = (struct osd_it_iam *)di;
4611
4612         return iam_it_store(&it->oi_it);
4613 }
4614
4615 /**
4616  * Restore iterator from cookie.
4617  *
4618  * \param  di      osd iterator
4619  * \param  hash    Iterator location cookie
4620  *
4621  * \retval +ve  di points to record with least key not larger than key.
4622  * \retval  0   di points to exact matched key
4623  * \retval -ve  failure
4624  */
4625
4626 static int osd_it_iam_load(const struct lu_env *env,
4627                            const struct dt_it *di, __u64 hash)
4628 {
4629         struct osd_it_iam *it = (struct osd_it_iam *)di;
4630
4631         return iam_it_load(&it->oi_it, hash);
4632 }
4633
4634 static const struct dt_index_operations osd_index_iam_ops = {
4635         .dio_lookup         = osd_index_iam_lookup,
4636         .dio_declare_insert = osd_index_declare_iam_insert,
4637         .dio_insert         = osd_index_iam_insert,
4638         .dio_declare_delete = osd_index_declare_iam_delete,
4639         .dio_delete         = osd_index_iam_delete,
4640         .dio_it     = {
4641                 .init     = osd_it_iam_init,
4642                 .fini     = osd_it_iam_fini,
4643                 .get      = osd_it_iam_get,
4644                 .put      = osd_it_iam_put,
4645                 .next     = osd_it_iam_next,
4646                 .key      = osd_it_iam_key,
4647                 .key_size = osd_it_iam_key_size,
4648                 .rec      = osd_it_iam_rec,
4649                 .store    = osd_it_iam_store,
4650                 .load     = osd_it_iam_load
4651         }
4652 };
4653
4654
4655 /**
4656  * Creates or initializes iterator context.
4657  *
4658  * \retval struct osd_it_ea, iterator structure on success
4659  *
4660  */
4661 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4662                                     struct dt_object *dt,
4663                                     __u32 attr)
4664 {
4665         struct osd_object       *obj  = osd_dt_obj(dt);
4666         struct osd_thread_info  *info = osd_oti_get(env);
4667         struct osd_it_ea        *oie;
4668         struct file             *file;
4669         struct lu_object        *lo   = &dt->do_lu;
4670         struct dentry           *obj_dentry;
4671         ENTRY;
4672
4673         if (!dt_object_exists(dt) || obj->oo_destroyed)
4674                 RETURN(ERR_PTR(-ENOENT));
4675
4676         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4677         if (oie == NULL)
4678                 RETURN(ERR_PTR(-ENOMEM));
4679         obj_dentry = &oie->oie_dentry;
4680
4681         obj_dentry->d_inode = obj->oo_inode;
4682         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4683         obj_dentry->d_name.hash = 0;
4684
4685         oie->oie_rd_dirent       = 0;
4686         oie->oie_it_dirent       = 0;
4687         oie->oie_dirent          = NULL;
4688         if (unlikely(!info->oti_it_ea_buf_used)) {
4689                 oie->oie_buf = info->oti_it_ea_buf;
4690                 info->oti_it_ea_buf_used = 1;
4691         } else {
4692                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4693                 if (oie->oie_buf == NULL)
4694                         RETURN(ERR_PTR(-ENOMEM));
4695         }
4696         oie->oie_obj             = obj;
4697
4698         file = &oie->oie_file;
4699
4700         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4701         if (attr & LUDA_64BITHASH)
4702                 file->f_mode    = FMODE_64BITHASH;
4703         else
4704                 file->f_mode    = FMODE_32BITHASH;
4705         file->f_path.dentry     = obj_dentry;
4706         file->f_mapping         = obj->oo_inode->i_mapping;
4707         file->f_op              = obj->oo_inode->i_fop;
4708         set_file_inode(file, obj->oo_inode);
4709
4710         lu_object_get(lo);
4711         RETURN((struct dt_it *) oie);
4712 }
4713
4714 /**
4715  * Destroy or finishes iterator context.
4716  *
4717  * \param di iterator structure to be destroyed
4718  */
4719 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4720 {
4721         struct osd_thread_info  *info = osd_oti_get(env);
4722         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
4723         struct osd_object       *obj    = oie->oie_obj;
4724         struct inode            *inode  = obj->oo_inode;
4725
4726         ENTRY;
4727         oie->oie_file.f_op->release(inode, &oie->oie_file);
4728         lu_object_put(env, &obj->oo_dt.do_lu);
4729         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4730                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4731         else
4732                 info->oti_it_ea_buf_used = 0;
4733         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4734         EXIT;
4735 }
4736
4737 /**
4738  * It position the iterator at given key, so that next lookup continues from
4739  * that key Or it is similar to dio_it->load() but based on a key,
4740  * rather than file position.
4741  *
4742  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4743  * to the beginning.
4744  *
4745  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4746  */
4747 static int osd_it_ea_get(const struct lu_env *env,
4748                          struct dt_it *di, const struct dt_key *key)
4749 {
4750         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4751
4752         ENTRY;
4753         LASSERT(((const char *)key)[0] == '\0');
4754         it->oie_file.f_pos      = 0;
4755         it->oie_rd_dirent       = 0;
4756         it->oie_it_dirent       = 0;
4757         it->oie_dirent          = NULL;
4758
4759         RETURN(+1);
4760 }
4761
4762 /**
4763  * Does nothing
4764  */
4765 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4766 {
4767 }
4768
4769 struct osd_filldir_cbs {
4770 #ifdef HAVE_DIR_CONTEXT
4771         struct dir_context ctx;
4772 #endif
4773         struct osd_it_ea  *it;
4774 };
4775 /**
4776  * It is called internally by ->readdir(). It fills the
4777  * iterator's in-memory data structure with required
4778  * information i.e. name, namelen, rec_size etc.
4779  *
4780  * \param buf in which information to be filled in.
4781  * \param name name of the file in given dir
4782  *
4783  * \retval 0 on success
4784  * \retval 1 on buffer full
4785  */
4786 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4787                                loff_t offset, __u64 ino,
4788                                unsigned d_type)
4789 {
4790         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4791         struct osd_object       *obj  = it->oie_obj;
4792         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4793         struct lu_fid           *fid  = &ent->oied_fid;
4794         struct osd_fid_pack     *rec;
4795         ENTRY;
4796
4797         /* this should never happen */
4798         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4799                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4800                 RETURN(-EIO);
4801         }
4802
4803         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4804             OSD_IT_EA_BUFSIZE)
4805                 RETURN(1);
4806
4807         /* "." is just the object itself. */
4808         if (namelen == 1 && name[0] == '.') {
4809                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4810         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4811                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4812                 if (osd_fid_unpack(fid, rec) != 0)
4813                         fid_zero(fid);
4814         } else {
4815                 fid_zero(fid);
4816         }
4817         d_type &= ~LDISKFS_DIRENT_LUFID;
4818
4819         /* NOT export local root. */
4820         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4821                 ino = obj->oo_inode->i_ino;
4822                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4823         }
4824
4825         ent->oied_ino     = ino;
4826         ent->oied_off     = offset;
4827         ent->oied_namelen = namelen;
4828         ent->oied_type    = d_type;
4829
4830         memcpy(ent->oied_name, name, namelen);
4831
4832         it->oie_rd_dirent++;
4833         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4834         RETURN(0);
4835 }
4836
4837 /**
4838  * Calls ->readdir() to load a directory entry at a time
4839  * and stored it in iterator's in-memory data structure.
4840  *
4841  * \param di iterator's in memory structure
4842  *
4843  * \retval   0 on success
4844  * \retval -ve on error
4845  * \retval +1 reach the end of entry
4846  */
4847 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4848                                const struct dt_it *di)
4849 {
4850         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4851         struct osd_object  *obj   = it->oie_obj;
4852         struct inode       *inode = obj->oo_inode;
4853         struct htree_lock  *hlock = NULL;
4854         struct file        *filp  = &it->oie_file;
4855         int                 rc = 0;
4856         struct osd_filldir_cbs buf = {
4857 #ifdef HAVE_DIR_CONTEXT
4858                 .ctx.actor = osd_ldiskfs_filldir,
4859 #endif
4860                 .it = it
4861         };
4862
4863         ENTRY;
4864         it->oie_dirent = it->oie_buf;
4865         it->oie_rd_dirent = 0;
4866
4867         if (obj->oo_hl_head != NULL) {
4868                 hlock = osd_oti_get(env)->oti_hlock;
4869                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4870                                    inode, LDISKFS_HLOCK_READDIR);
4871         } else {
4872                 down_read(&obj->oo_ext_idx_sem);
4873         }
4874
4875 #ifdef HAVE_DIR_CONTEXT
4876         buf.ctx.pos = filp->f_pos;
4877         rc = inode->i_fop->iterate(filp, &buf.ctx);
4878         filp->f_pos = buf.ctx.pos;
4879 #else
4880         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4881 #endif
4882
4883         if (hlock != NULL)
4884                 ldiskfs_htree_unlock(hlock);
4885         else
4886                 up_read(&obj->oo_ext_idx_sem);
4887
4888         if (it->oie_rd_dirent == 0) {
4889                 /*If it does not get any dirent, it means it has been reached
4890                  *to the end of the dir */
4891                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4892                 if (rc == 0)
4893                         rc = 1;
4894         } else {
4895                 it->oie_dirent = it->oie_buf;
4896                 it->oie_it_dirent = 1;
4897         }
4898
4899         RETURN(rc);
4900 }
4901
4902 /**
4903  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4904  * to load a directory entry at a time and stored it in
4905  * iterator's in-memory data structure.
4906  *
4907  * \param di iterator's in memory structure
4908  *
4909  * \retval +ve iterator reached to end
4910  * \retval   0 iterator not reached to end
4911  * \retval -ve on error
4912  */
4913 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4914 {
4915         struct osd_it_ea *it = (struct osd_it_ea *)di;
4916         int rc;
4917
4918         ENTRY;
4919
4920         if (it->oie_it_dirent < it->oie_rd_dirent) {
4921                 it->oie_dirent =
4922                         (void *) it->oie_dirent +
4923                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4924                                        it->oie_dirent->oied_namelen);
4925                 it->oie_it_dirent++;
4926                 RETURN(0);
4927         } else {
4928                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4929                         rc = +1;
4930                 else
4931                         rc = osd_ldiskfs_it_fill(env, di);
4932         }
4933
4934         RETURN(rc);
4935 }
4936
4937 /**
4938  * Returns the key at current position from iterator's in memory structure.
4939  *
4940  * \param di iterator's in memory structure
4941  *
4942  * \retval key i.e. struct dt_key on success
4943  */
4944 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4945                                     const struct dt_it *di)
4946 {
4947         struct osd_it_ea *it = (struct osd_it_ea *)di;
4948
4949         return (struct dt_key *)it->oie_dirent->oied_name;
4950 }
4951
4952 /**
4953  * Returns the key's size at current position from iterator's in memory structure.
4954  *
4955  * \param di iterator's in memory structure
4956  *
4957  * \retval key_size i.e. struct dt_key on success
4958  */
4959 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4960 {
4961         struct osd_it_ea *it = (struct osd_it_ea *)di;
4962
4963         return it->oie_dirent->oied_namelen;
4964 }
4965
4966 static int
4967 osd_dirent_update(handle_t *jh, struct super_block *sb,
4968                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4969                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4970 {
4971         struct osd_fid_pack *rec;
4972         int                  rc;
4973         ENTRY;
4974
4975         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4976         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4977
4978         rc = ldiskfs_journal_get_write_access(jh, bh);
4979         if (rc != 0)
4980                 RETURN(rc);
4981
4982         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4983         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4984         rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
4985
4986         RETURN(rc);
4987 }
4988
4989 static inline int
4990 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4991 {
4992         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4993             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4994                 return 1;
4995         else
4996                 return 0;
4997 }
4998
4999 static inline int
5000 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5001 {
5002         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5003                  "dot_dotdot = %d\n", dot_dotdot);
5004
5005         if (LDISKFS_DIR_REC_LEN(de) >=
5006             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5007                 return 1;
5008         else
5009                 return 0;
5010 }
5011
5012 static int
5013 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5014                     struct inode *dir, struct inode *inode,
5015                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5016                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5017                     struct htree_lock *hlock)
5018 {
5019         struct dentry               *dentry;
5020         struct osd_fid_pack         *rec;
5021         struct ldiskfs_dentry_param *ldp;
5022         int                          rc;
5023         ENTRY;
5024
5025         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5026                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5027                 RETURN(0);
5028
5029         /* There is enough space to hold the FID-in-dirent. */
5030         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5031                                  dir->i_sb->s_blocksize)) {
5032                 rc = ldiskfs_journal_get_write_access(jh, bh);
5033                 if (rc != 0)
5034                         RETURN(rc);
5035
5036                 de->name[de->name_len] = 0;
5037                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5038                 rec->fp_len = sizeof(struct lu_fid) + 1;
5039                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5040                 de->file_type |= LDISKFS_DIRENT_LUFID;
5041
5042                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5043
5044                 RETURN(rc);
5045         }
5046
5047         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5048         if (rc != 0)
5049                 RETURN(rc);
5050
5051         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5052                                            ent->oied_namelen);
5053         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5054         osd_get_ldiskfs_dirent_param(ldp, fid);
5055         dentry->d_fsdata = (void *)ldp;
5056         ll_vfs_dq_init(dir);
5057         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5058         /* It is too bad, we cannot reinsert the name entry back.
5059          * That means we lose it! */
5060         if (rc != 0)
5061                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5062                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5063                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5064                        dir->i_ino, dir->i_generation,
5065                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5066
5067         RETURN(rc);
5068 }
5069
5070 static int
5071 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5072                         struct osd_it_ea *it, struct lu_fid *fid,
5073                         struct osd_inode_id *id, __u32 *attr)
5074 {
5075         struct osd_thread_info     *info        = osd_oti_get(env);
5076         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5077         struct osd_device          *dev         = osd_obj2dev(obj);
5078         struct super_block         *sb          = osd_sb(dev);
5079         const char                 *devname     =
5080                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5081         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5082         struct inode               *dir         = obj->oo_inode;
5083         struct htree_lock          *hlock       = NULL;
5084         struct buffer_head         *bh          = NULL;
5085         handle_t                   *jh          = NULL;
5086         struct ldiskfs_dir_entry_2 *de;
5087         struct dentry              *dentry;
5088         struct inode               *inode;
5089         int                         credits;
5090         int                         rc;
5091         int                         dot_dotdot  = 0;
5092         bool                        dirty       = false;
5093         ENTRY;
5094
5095         if (ent->oied_name[0] == '.') {
5096                 if (ent->oied_namelen == 1)
5097                         dot_dotdot = 1;
5098                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5099                         dot_dotdot = 2;
5100         }
5101
5102         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5103                                       ent->oied_namelen);
5104
5105         /* We need to ensure that the name entry is still valid.
5106          * Because it may be removed or renamed by other already.
5107          *
5108          * The unlink or rename operation will start journal before PDO lock,
5109          * so to avoid deadlock, here we need to start journal handle before
5110          * related PDO lock also. But because we do not know whether there
5111          * will be something to be repaired before PDO lock, we just start
5112          * journal without conditions.
5113          *
5114          * We may need to remove the name entry firstly, then insert back.
5115          * One credit is for user quota file update.
5116          * One credit is for group quota file update.
5117          * Two credits are for dirty inode. */
5118         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5119                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5120
5121 again:
5122         if (dev->od_dirent_journal != 0) {
5123                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5124                 if (IS_ERR(jh)) {
5125                         rc = PTR_ERR(jh);
5126                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5127                                "check_repair, dir = %lu/%u, credits = %d, "
5128                                "name = %.*s: rc = %d\n",
5129                                devname, dir->i_ino, dir->i_generation, credits,
5130                                ent->oied_namelen, ent->oied_name, rc);
5131                         RETURN(rc);
5132                 }
5133
5134                 if (obj->oo_hl_head != NULL) {
5135                         hlock = osd_oti_get(env)->oti_hlock;
5136                         /* "0" means exclusive lock for the whole directory.
5137                          * We need to prevent others access such name entry
5138                          * during the delete + insert. Neither HLOCK_ADD nor
5139                          * HLOCK_DEL cannot guarantee the atomicity. */
5140                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5141                 } else {
5142                         down_write(&obj->oo_ext_idx_sem);
5143                 }
5144         } else {
5145                 if (obj->oo_hl_head != NULL) {
5146                         hlock = osd_oti_get(env)->oti_hlock;
5147                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5148                                            LDISKFS_HLOCK_LOOKUP);
5149                 } else {
5150                         down_read(&obj->oo_ext_idx_sem);
5151                 }
5152         }
5153
5154         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5155         /* For dot/dotdot entry, if there is not enough space to hold the
5156          * FID-in-dirent, just keep them there. It only happens when the
5157          * device upgraded from 1.8 or restored from MDT file-level backup.
5158          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5159          * and needs to get FID from LMA when readdir, it will not affect the
5160          * performance much. */
5161         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5162             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5163                 *attr |= LUDA_IGNORE;
5164                 GOTO(out_journal, rc = 0);
5165         }
5166
5167         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5168         inode = osd_iget(info, dev, id);
5169         if (IS_ERR(inode)) {
5170                 rc = PTR_ERR(inode);
5171                 if (rc == -ENOENT || rc == -ESTALE)
5172                         rc = 1;
5173                 else
5174                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5175                                "check_repair, dir = %lu/%u, name = %.*s: "
5176                                "rc = %d\n",
5177                                devname, dir->i_ino, dir->i_generation,
5178                                ent->oied_namelen, ent->oied_name, rc);
5179
5180                 GOTO(out_journal, rc);
5181         }
5182
5183         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5184         if (rc == 0) {
5185                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5186
5187                 if (fid_is_sane(fid)) {
5188                         /* FID-in-dirent is valid. */
5189                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5190                                 GOTO(out_inode, rc = 0);
5191
5192                         /* Do not repair under dryrun mode. */
5193                         if (*attr & LUDA_VERIFY_DRYRUN) {
5194                                 *attr |= LUDA_REPAIR;
5195                                 GOTO(out_inode, rc = 0);
5196                         }
5197
5198                         if (dev->od_dirent_journal == 0) {
5199                                 iput(inode);
5200                                 brelse(bh);
5201                                 if (hlock != NULL)
5202                                         ldiskfs_htree_unlock(hlock);
5203                                 else
5204                                         up_read(&obj->oo_ext_idx_sem);
5205                                 dev->od_dirent_journal = 1;
5206                                 goto again;
5207                         }
5208
5209                         *fid = lma->lma_self_fid;
5210                         dirty = true;
5211                         /* Update the FID-in-dirent. */
5212                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5213                         if (rc == 0)
5214                                 *attr |= LUDA_REPAIR;
5215                         else
5216                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5217                                        "in the dirent, dir = %lu/%u, "
5218                                        "name = %.*s, "DFID": rc = %d\n",
5219                                        devname, dir->i_ino, dir->i_generation,
5220                                        ent->oied_namelen, ent->oied_name,
5221                                        PFID(fid), rc);
5222                 } else {
5223                         /* Do not repair under dryrun mode. */
5224                         if (*attr & LUDA_VERIFY_DRYRUN) {
5225                                 *fid = lma->lma_self_fid;
5226                                 *attr |= LUDA_REPAIR;
5227                                 GOTO(out_inode, rc = 0);
5228                         }
5229
5230                         if (dev->od_dirent_journal == 0) {
5231                                 iput(inode);
5232                                 brelse(bh);
5233                                 if (hlock != NULL)
5234                                         ldiskfs_htree_unlock(hlock);
5235                                 else
5236                                         up_read(&obj->oo_ext_idx_sem);
5237                                 dev->od_dirent_journal = 1;
5238                                 goto again;
5239                         }
5240
5241                         *fid = lma->lma_self_fid;
5242                         dirty = true;
5243                         /* Append the FID-in-dirent. */
5244                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5245                                                  fid, bh, de, hlock);
5246                         if (rc == 0)
5247                                 *attr |= LUDA_REPAIR;
5248                         else
5249                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5250                                        "after 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                 }
5256         } else if (rc == -ENODATA) {
5257                 /* Do not repair under dryrun mode. */
5258                 if (*attr & LUDA_VERIFY_DRYRUN) {
5259                         if (fid_is_sane(fid)) {
5260                                 *attr |= LUDA_REPAIR;
5261                         } else {
5262                                 lu_igif_build(fid, inode->i_ino,
5263                                               inode->i_generation);
5264                                 *attr |= LUDA_UPGRADE;
5265                         }
5266                         GOTO(out_inode, rc = 0);
5267                 }
5268
5269                 if (dev->od_dirent_journal == 0) {
5270                         iput(inode);
5271                         brelse(bh);
5272                         if (hlock != NULL)
5273                                 ldiskfs_htree_unlock(hlock);
5274                         else
5275                                 up_read(&obj->oo_ext_idx_sem);
5276                         dev->od_dirent_journal = 1;
5277                         goto again;
5278                 }
5279
5280                 dirty = true;
5281                 if (unlikely(fid_is_sane(fid))) {
5282                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5283                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5284                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5285                         if (rc == 0)
5286                                 *attr |= LUDA_REPAIR;
5287                         else
5288                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5289                                        "update dirent, dir = %lu/%u, "
5290                                        "name = %.*s, "DFID": rc = %d\n",
5291                                        devname, dir->i_ino, dir->i_generation,
5292                                        ent->oied_namelen, ent->oied_name,
5293                                        PFID(fid), rc);
5294                 } else {
5295                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5296                         /* It is probably IGIF object. Only aappend the
5297                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5298                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5299                                                  fid, bh, de, hlock);
5300                         if (rc == 0)
5301                                 *attr |= LUDA_UPGRADE;
5302                         else
5303                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5304                                        "after the dirent, dir = %lu/%u, "
5305                                        "name = %.*s, "DFID": rc = %d\n",
5306                                        devname, dir->i_ino, dir->i_generation,
5307                                        ent->oied_namelen, ent->oied_name,
5308                                        PFID(fid), rc);
5309                 }
5310         }
5311
5312         GOTO(out_inode, rc);
5313
5314 out_inode:
5315         iput(inode);
5316
5317 out_journal:
5318         brelse(bh);
5319         if (hlock != NULL) {
5320                 ldiskfs_htree_unlock(hlock);
5321         } else {
5322                 if (dev->od_dirent_journal != 0)
5323                         up_write(&obj->oo_ext_idx_sem);
5324                 else
5325                         up_read(&obj->oo_ext_idx_sem);
5326         }
5327         if (jh != NULL)
5328                 ldiskfs_journal_stop(jh);
5329         if (rc >= 0 && !dirty)
5330                 dev->od_dirent_journal = 0;
5331         return rc;
5332 }
5333
5334 /**
5335  * Returns the value at current position from iterator's in memory structure.
5336  *
5337  * \param di struct osd_it_ea, iterator's in memory structure
5338  * \param attr attr requested for dirent.
5339  * \param lde lustre dirent
5340  *
5341  * \retval   0 no error and \param lde has correct lustre dirent.
5342  * \retval -ve on error
5343  */
5344 static inline int osd_it_ea_rec(const struct lu_env *env,
5345                                 const struct dt_it *di,
5346                                 struct dt_rec *dtrec, __u32 attr)
5347 {
5348         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5349         struct osd_object      *obj   = it->oie_obj;
5350         struct osd_device      *dev   = osd_obj2dev(obj);
5351         struct osd_thread_info *oti   = osd_oti_get(env);
5352         struct osd_inode_id    *id    = &oti->oti_id;
5353         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5354         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5355         __u32                   ino   = it->oie_dirent->oied_ino;
5356         int                     rc    = 0;
5357         ENTRY;
5358
5359         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5360
5361         if (attr & LUDA_VERIFY) {
5362                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5363                         attr |= LUDA_IGNORE;
5364                         /* If the parent is on remote MDT, and there
5365                          * is no FID-in-dirent, then we have to get
5366                          * the parent FID from the linkEA.  */
5367                         if (!fid_is_sane(fid) &&
5368                             it->oie_dirent->oied_namelen == 2 &&
5369                             it->oie_dirent->oied_name[0] == '.' &&
5370                             it->oie_dirent->oied_name[1] == '.')
5371                                 osd_get_pfid_from_linkea(env, obj, fid);
5372                 } else {
5373                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5374                                                      &attr);
5375                 }
5376         } else {
5377                 attr &= ~LU_DIRENT_ATTRS_MASK;
5378                 if (!fid_is_sane(fid)) {
5379                         bool is_dotdot = false;
5380                         if (it->oie_dirent->oied_namelen == 2 &&
5381                             it->oie_dirent->oied_name[0] == '.' &&
5382                             it->oie_dirent->oied_name[1] == '.')
5383                                 is_dotdot = true;
5384                         /* If the parent is on remote MDT, and there
5385                          * is no FID-in-dirent, then we have to get
5386                          * the parent FID from the linkEA.  */
5387                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5388                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5389                         } else {
5390                                 if (is_dotdot == false &&
5391                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5392                                         RETURN(-ENOENT);
5393
5394                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5395                         }
5396                 } else {
5397                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5398                 }
5399         }
5400
5401         /* Pack the entry anyway, at least the offset is right. */
5402         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5403                            it->oie_dirent->oied_name,
5404                            it->oie_dirent->oied_namelen,
5405                            it->oie_dirent->oied_type, attr);
5406
5407         if (rc < 0)
5408                 RETURN(rc);
5409
5410         if (osd_remote_fid(env, dev, fid))
5411                 RETURN(0);
5412
5413         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5414                 osd_add_oi_cache(oti, dev, id, fid);
5415
5416         RETURN(rc > 0 ? 0 : rc);
5417 }
5418
5419 /**
5420  * Returns the record size size at current position.
5421  *
5422  * This function will return record(lu_dirent) size in bytes.
5423  *
5424  * \param[in] env       execution environment
5425  * \param[in] di        iterator's in memory structure
5426  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5427  *                      calculate the lu_dirent size.
5428  *
5429  * \retval      record size(in bytes & in memory) of the current lu_dirent
5430  *              entry.
5431  */
5432 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5433                               __u32 attr)
5434 {
5435         struct osd_it_ea *it = (struct osd_it_ea *)di;
5436
5437         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5438 }
5439
5440 /**
5441  * Returns a cookie for current position of the iterator head, so that
5442  * user can use this cookie to load/start the iterator next time.
5443  *
5444  * \param di iterator's in memory structure
5445  *
5446  * \retval cookie for current position, on success
5447  */
5448 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5449 {
5450         struct osd_it_ea *it = (struct osd_it_ea *)di;
5451
5452         return it->oie_dirent->oied_off;
5453 }
5454
5455 /**
5456  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5457  * to load a directory entry at a time and stored it i inn,
5458  * in iterator's in-memory data structure.
5459  *
5460  * \param di struct osd_it_ea, iterator's in memory structure
5461  *
5462  * \retval +ve on success
5463  * \retval -ve on error
5464  */
5465 static int osd_it_ea_load(const struct lu_env *env,
5466                           const struct dt_it *di, __u64 hash)
5467 {
5468         struct osd_it_ea *it = (struct osd_it_ea *)di;
5469         int rc;
5470
5471         ENTRY;
5472         it->oie_file.f_pos = hash;
5473
5474         rc =  osd_ldiskfs_it_fill(env, di);
5475         if (rc > 0)
5476                 rc = -ENODATA;
5477
5478         if (rc == 0)
5479                 rc = +1;
5480
5481         RETURN(rc);
5482 }
5483
5484 /**
5485  * Index lookup function for interoperability mode (b11826).
5486  *
5487  * \param key,  key i.e. file name to be searched
5488  *
5489  * \retval +ve, on success
5490  * \retval -ve, on error
5491  */
5492 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5493                                struct dt_rec *rec, const struct dt_key *key)
5494 {
5495         struct osd_object *obj = osd_dt_obj(dt);
5496         int rc = 0;
5497
5498         ENTRY;
5499
5500         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5501         LINVRNT(osd_invariant(obj));
5502
5503         rc = osd_ea_lookup_rec(env, obj, rec, key);
5504         if (rc == 0)
5505                 rc = +1;
5506         RETURN(rc);
5507 }
5508
5509 /**
5510  * Index and Iterator operations for interoperability
5511  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5512  */
5513 static const struct dt_index_operations osd_index_ea_ops = {
5514         .dio_lookup         = osd_index_ea_lookup,
5515         .dio_declare_insert = osd_index_declare_ea_insert,
5516         .dio_insert         = osd_index_ea_insert,
5517         .dio_declare_delete = osd_index_declare_ea_delete,
5518         .dio_delete         = osd_index_ea_delete,
5519         .dio_it     = {
5520                 .init     = osd_it_ea_init,
5521                 .fini     = osd_it_ea_fini,
5522                 .get      = osd_it_ea_get,
5523                 .put      = osd_it_ea_put,
5524                 .next     = osd_it_ea_next,
5525                 .key      = osd_it_ea_key,
5526                 .key_size = osd_it_ea_key_size,
5527                 .rec      = osd_it_ea_rec,
5528                 .rec_size = osd_it_ea_rec_size,
5529                 .store    = osd_it_ea_store,
5530                 .load     = osd_it_ea_load
5531         }
5532 };
5533
5534 static void *osd_key_init(const struct lu_context *ctx,
5535                           struct lu_context_key *key)
5536 {
5537         struct osd_thread_info *info;
5538
5539         OBD_ALLOC_PTR(info);
5540         if (info == NULL)
5541                 return ERR_PTR(-ENOMEM);
5542
5543         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5544         if (info->oti_it_ea_buf == NULL)
5545                 goto out_free_info;
5546
5547         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5548
5549         info->oti_hlock = ldiskfs_htree_lock_alloc();
5550         if (info->oti_hlock == NULL)
5551                 goto out_free_ea;
5552
5553         return info;
5554
5555  out_free_ea:
5556         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5557  out_free_info:
5558         OBD_FREE_PTR(info);
5559         return ERR_PTR(-ENOMEM);
5560 }
5561
5562 static void osd_key_fini(const struct lu_context *ctx,
5563                          struct lu_context_key *key, void* data)
5564 {
5565         struct osd_thread_info *info = data;
5566
5567         if (info->oti_inode != NULL)
5568                 OBD_FREE_PTR(info->oti_inode);
5569         if (info->oti_hlock != NULL)
5570                 ldiskfs_htree_lock_free(info->oti_hlock);
5571         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5572         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5573         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5574         lu_buf_free(&info->oti_big_buf);
5575         OBD_FREE_PTR(info);
5576 }
5577
5578 static void osd_key_exit(const struct lu_context *ctx,
5579                          struct lu_context_key *key, void *data)
5580 {
5581         struct osd_thread_info *info = data;
5582
5583         LASSERT(info->oti_r_locks == 0);
5584         LASSERT(info->oti_w_locks == 0);
5585         LASSERT(info->oti_txns    == 0);
5586 }
5587
5588 /* type constructor/destructor: osd_type_init, osd_type_fini */
5589 LU_TYPE_INIT_FINI(osd, &osd_key);
5590
5591 struct lu_context_key osd_key = {
5592         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5593         .lct_init = osd_key_init,
5594         .lct_fini = osd_key_fini,
5595         .lct_exit = osd_key_exit
5596 };
5597
5598
5599 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5600                            const char *name, struct lu_device *next)
5601 {
5602         struct osd_device *osd = osd_dev(d);
5603
5604         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5605             >= sizeof(osd->od_svname))
5606                 return -E2BIG;
5607         return osd_procfs_init(osd, name);
5608 }
5609
5610 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5611 {
5612         struct seq_server_site  *ss = osd_seq_site(osd);
5613         int                     rc;
5614         ENTRY;
5615
5616         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5617                 RETURN(0);
5618
5619         if (unlikely(ss == NULL))
5620                 RETURN(-ENODEV);
5621
5622         OBD_ALLOC_PTR(osd->od_cl_seq);
5623         if (osd->od_cl_seq == NULL)
5624                 RETURN(-ENOMEM);
5625
5626         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5627                              osd->od_svname, ss->ss_server_seq);
5628         if (rc != 0) {
5629                 OBD_FREE_PTR(osd->od_cl_seq);
5630                 osd->od_cl_seq = NULL;
5631                 RETURN(rc);
5632         }
5633
5634         if (ss->ss_node_id == 0) {
5635                 /* If the OSD on the sequence controller(MDT0), then allocate
5636                  * sequence here, otherwise allocate sequence after connected
5637                  * to MDT0 (see mdt_register_lwp_callback()). */
5638                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
5639                                    &osd->od_cl_seq->lcs_space, env);
5640         }
5641
5642         RETURN(rc);
5643 }
5644
5645 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5646 {
5647         if (osd->od_cl_seq == NULL)
5648                 return;
5649
5650         seq_client_fini(osd->od_cl_seq);
5651         OBD_FREE_PTR(osd->od_cl_seq);
5652         osd->od_cl_seq = NULL;
5653 }
5654
5655 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5656 {
5657         ENTRY;
5658
5659         /* shutdown quota slave instance associated with the device */
5660         if (o->od_quota_slave != NULL) {
5661                 qsd_fini(env, o->od_quota_slave);
5662                 o->od_quota_slave = NULL;
5663         }
5664
5665         osd_fid_fini(env, o);
5666
5667         RETURN(0);
5668 }
5669
5670 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5671 {
5672         ENTRY;
5673
5674         if (o->od_mnt != NULL) {
5675                 shrink_dcache_sb(osd_sb(o));
5676                 osd_sync(env, &o->od_dt_dev);
5677
5678                 mntput(o->od_mnt);
5679                 o->od_mnt = NULL;
5680         }
5681
5682         EXIT;
5683 }
5684
5685 static int osd_mount(const struct lu_env *env,
5686                      struct osd_device *o, struct lustre_cfg *cfg)
5687 {
5688         const char              *name  = lustre_cfg_string(cfg, 0);
5689         const char              *dev  = lustre_cfg_string(cfg, 1);
5690         const char              *opts;
5691         unsigned long            page, s_flags, lmd_flags = 0;
5692         struct page             *__page;
5693         struct file_system_type *type;
5694         char                    *options = NULL;
5695         char                    *str;
5696         struct osd_thread_info  *info = osd_oti_get(env);
5697         struct lu_fid           *fid = &info->oti_fid;
5698         struct inode            *inode;
5699         int                      rc = 0, force_over_128tb = 0;
5700         ENTRY;
5701
5702         if (o->od_mnt != NULL)
5703                 RETURN(0);
5704
5705         if (strlen(dev) >= sizeof(o->od_mntdev))
5706                 RETURN(-E2BIG);
5707         strcpy(o->od_mntdev, dev);
5708
5709         str = lustre_cfg_string(cfg, 2);
5710         s_flags = simple_strtoul(str, NULL, 0);
5711         str = strstr(str, ":");
5712         if (str)
5713                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5714         opts = lustre_cfg_string(cfg, 3);
5715 #ifdef __BIG_ENDIAN
5716         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5717                 CERROR("%s: device %s extents feature is not guaranteed to "
5718                        "work on big-endian systems. Use \"bigendian_extents\" "
5719                        "mount option to override.\n", name, dev);
5720                 RETURN(-EINVAL);
5721         }
5722 #endif
5723         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5724                 force_over_128tb = 1;
5725
5726         __page = alloc_page(GFP_IOFS);
5727         if (__page == NULL)
5728                 GOTO(out, rc = -ENOMEM);
5729         page = (unsigned long)page_address(__page);
5730         options = (char *)page;
5731         *options = '\0';
5732         if (opts != NULL) {
5733                 /* strip out the options for back compatiblity */
5734                 static char *sout[] = {
5735                         "mballoc",
5736                         "iopen",
5737                         "noiopen",
5738                         "iopen_nopriv",
5739                         "extents",
5740                         "noextents",
5741                         /* strip out option we processed in osd */
5742                         "bigendian_extents",
5743                         "force_over_128tb",
5744                         NULL
5745                 };
5746                 strcat(options, opts);
5747                 for (rc = 0, str = options; sout[rc]; ) {
5748                         char *op = strstr(str, sout[rc]);
5749                         if (op == NULL) {
5750                                 rc++;
5751                                 str = options;
5752                                 continue;
5753                         }
5754                         if (op == options || *(op - 1) == ',') {
5755                                 str = op + strlen(sout[rc]);
5756                                 if (*str == ',' || *str == '\0') {
5757                                         *str == ',' ? str++ : str;
5758                                         memmove(op, str, strlen(str) + 1);
5759                                 }
5760                         }
5761                         for (str = op; *str != ',' && *str != '\0'; str++)
5762                                 ;
5763                 }
5764         } else {
5765                 strncat(options, "user_xattr,acl", 14);
5766         }
5767
5768         /* Glom up mount options */
5769         if (*options != '\0')
5770                 strcat(options, ",");
5771         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5772
5773         type = get_fs_type("ldiskfs");
5774         if (!type) {
5775                 CERROR("%s: cannot find ldiskfs module\n", name);
5776                 GOTO(out, rc = -ENODEV);
5777         }
5778
5779         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5780         module_put(type->owner);
5781
5782         if (IS_ERR(o->od_mnt)) {
5783                 rc = PTR_ERR(o->od_mnt);
5784                 o->od_mnt = NULL;
5785                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5786                 GOTO(out, rc);
5787         }
5788
5789         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
5790             force_over_128tb == 0) {
5791                 CERROR("%s: device %s LDISKFS does not support filesystems "
5792                        "greater than 128TB and can cause data corruption. "
5793                        "Use \"force_over_128tb\" mount option to override.\n",
5794                        name, dev);
5795                 GOTO(out, rc = -EINVAL);
5796         }
5797
5798 #ifdef HAVE_DEV_SET_RDONLY
5799         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5800                 CERROR("%s: underlying device %s is marked as read-only. "
5801                        "Setup failed\n", name, dev);
5802                 GOTO(out_mnt, rc = -EROFS);
5803         }
5804 #endif
5805
5806         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5807                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5808                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5809                 GOTO(out_mnt, rc = -EINVAL);
5810         }
5811
5812 #ifdef LDISKFS_MOUNT_DIRDATA
5813         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5814                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5815                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5816         else if (!o->od_is_ost)
5817                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5818                       "enabling the dirdata feature. If you do not want to "
5819                       "downgrade to Lustre-1.x again, you can enable it via "
5820                       "'tune2fs -O dirdata device'\n", name, dev);
5821 #endif
5822         inode = osd_sb(o)->s_root->d_inode;
5823         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5824         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5825         if (rc != 0) {
5826                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5827                 GOTO(out_mnt, rc);
5828         }
5829
5830         if (lmd_flags & LMD_FLG_NOSCRUB)
5831                 o->od_noscrub = 1;
5832
5833         GOTO(out, rc = 0);
5834
5835 out_mnt:
5836         mntput(o->od_mnt);
5837         o->od_mnt = NULL;
5838
5839 out:
5840         if (__page)
5841                 __free_page(__page);
5842
5843         return rc;
5844 }
5845
5846 static struct lu_device *osd_device_fini(const struct lu_env *env,
5847                                          struct lu_device *d)
5848 {
5849         struct osd_device *o = osd_dev(d);
5850         ENTRY;
5851
5852         osd_shutdown(env, o);
5853         osd_procfs_fini(o);
5854         osd_scrub_cleanup(env, o);
5855         osd_obj_map_fini(o);
5856         osd_umount(env, o);
5857
5858         RETURN(NULL);
5859 }
5860
5861 static int osd_device_init0(const struct lu_env *env,
5862                             struct osd_device *o,
5863                             struct lustre_cfg *cfg)
5864 {
5865         struct lu_device        *l = osd2lu_dev(o);
5866         struct osd_thread_info *info;
5867         int                     rc;
5868         int                     cplen = 0;
5869
5870         /* if the module was re-loaded, env can loose its keys */
5871         rc = lu_env_refill((struct lu_env *) env);
5872         if (rc)
5873                 GOTO(out, rc);
5874         info = osd_oti_get(env);
5875         LASSERT(info);
5876
5877         l->ld_ops = &osd_lu_ops;
5878         o->od_dt_dev.dd_ops = &osd_dt_ops;
5879
5880         spin_lock_init(&o->od_osfs_lock);
5881         mutex_init(&o->od_otable_mutex);
5882
5883         o->od_read_cache = 1;
5884         o->od_writethrough_cache = 1;
5885         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5886
5887         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5888                         sizeof(o->od_svname));
5889         if (cplen >= sizeof(o->od_svname)) {
5890                 rc = -E2BIG;
5891                 GOTO(out, rc);
5892         }
5893
5894         o->od_index = -1; /* -1 means index is invalid */
5895         rc = server_name2index(o->od_svname, &o->od_index, NULL);
5896         if (rc == LDD_F_SV_TYPE_OST)
5897                 o->od_is_ost = 1;
5898
5899         o->od_full_scrub_ratio = OFSR_DEFAULT;
5900         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
5901         rc = osd_mount(env, o, cfg);
5902         if (rc != 0)
5903                 GOTO(out, rc);
5904
5905         rc = osd_obj_map_init(env, o);
5906         if (rc != 0)
5907                 GOTO(out_mnt, rc);
5908
5909         rc = lu_site_init(&o->od_site, l);
5910         if (rc != 0)
5911                 GOTO(out_compat, rc);
5912         o->od_site.ls_bottom_dev = l;
5913
5914         rc = lu_site_init_finish(&o->od_site);
5915         if (rc != 0)
5916                 GOTO(out_site, rc);
5917
5918         INIT_LIST_HEAD(&o->od_ios_list);
5919         /* setup scrub, including OI files initialization */
5920         rc = osd_scrub_setup(env, o);
5921         if (rc < 0)
5922                 GOTO(out_site, rc);
5923
5924         rc = osd_procfs_init(o, o->od_svname);
5925         if (rc != 0) {
5926                 CERROR("%s: can't initialize procfs: rc = %d\n",
5927                        o->od_svname, rc);
5928                 GOTO(out_scrub, rc);
5929         }
5930
5931         LASSERT(l->ld_site->ls_linkage.next != NULL);
5932         LASSERT(l->ld_site->ls_linkage.prev != NULL);
5933
5934         /* initialize quota slave instance */
5935         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5936                                      o->od_proc_entry);
5937         if (IS_ERR(o->od_quota_slave)) {
5938                 rc = PTR_ERR(o->od_quota_slave);
5939                 o->od_quota_slave = NULL;
5940                 GOTO(out_procfs, rc);
5941         }
5942
5943         RETURN(0);
5944
5945 out_procfs:
5946         osd_procfs_fini(o);
5947 out_scrub:
5948         osd_scrub_cleanup(env, o);
5949 out_site:
5950         lu_site_fini(&o->od_site);
5951 out_compat:
5952         osd_obj_map_fini(o);
5953 out_mnt:
5954         osd_umount(env, o);
5955 out:
5956         return rc;
5957 }
5958
5959 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5960                                           struct lu_device_type *t,
5961                                           struct lustre_cfg *cfg)
5962 {
5963         struct osd_device *o;
5964         int                rc;
5965
5966         OBD_ALLOC_PTR(o);
5967         if (o == NULL)
5968                 return ERR_PTR(-ENOMEM);
5969
5970         rc = dt_device_init(&o->od_dt_dev, t);
5971         if (rc == 0) {
5972                 /* Because the ctx might be revived in dt_device_init,
5973                  * refill the env here */
5974                 lu_env_refill((struct lu_env *)env);
5975                 rc = osd_device_init0(env, o, cfg);
5976                 if (rc)
5977                         dt_device_fini(&o->od_dt_dev);
5978         }
5979
5980         if (unlikely(rc != 0))
5981                 OBD_FREE_PTR(o);
5982
5983         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5984 }
5985
5986 static struct lu_device *osd_device_free(const struct lu_env *env,
5987                                          struct lu_device *d)
5988 {
5989         struct osd_device *o = osd_dev(d);
5990         ENTRY;
5991
5992         /* XXX: make osd top device in order to release reference */
5993         d->ld_site->ls_top_dev = d;
5994         lu_site_purge(env, d->ld_site, -1);
5995         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5996                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5997                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5998         }
5999         lu_site_fini(&o->od_site);
6000         dt_device_fini(&o->od_dt_dev);
6001         OBD_FREE_PTR(o);
6002         RETURN(NULL);
6003 }
6004
6005 static int osd_process_config(const struct lu_env *env,
6006                               struct lu_device *d, struct lustre_cfg *cfg)
6007 {
6008         struct osd_device               *o = osd_dev(d);
6009         int                             rc;
6010         ENTRY;
6011
6012         switch (cfg->lcfg_command) {
6013         case LCFG_SETUP:
6014                 rc = osd_mount(env, o, cfg);
6015                 break;
6016         case LCFG_CLEANUP:
6017                 lu_dev_del_linkage(d->ld_site, d);
6018                 rc = osd_shutdown(env, o);
6019                 break;
6020         case LCFG_PARAM:
6021                 LASSERT(&o->od_dt_dev);
6022                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6023                                               cfg, &o->od_dt_dev);
6024                 if (rc > 0 || rc == -ENOSYS)
6025                         rc = class_process_proc_param(PARAM_OST,
6026                                                       lprocfs_osd_obd_vars,
6027                                                       cfg, &o->od_dt_dev);
6028                 break;
6029         default:
6030                 rc = -ENOSYS;
6031         }
6032
6033         RETURN(rc);
6034 }
6035
6036 static int osd_recovery_complete(const struct lu_env *env,
6037                                  struct lu_device *d)
6038 {
6039         struct osd_device       *osd = osd_dev(d);
6040         int                      rc = 0;
6041         ENTRY;
6042
6043         if (osd->od_quota_slave == NULL)
6044                 RETURN(0);
6045
6046         /* start qsd instance on recovery completion, this notifies the quota
6047          * slave code that we are about to process new requests now */
6048         rc = qsd_start(env, osd->od_quota_slave);
6049         RETURN(rc);
6050 }
6051
6052 /*
6053  * we use exports to track all osd users
6054  */
6055 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6056                            struct obd_device *obd, struct obd_uuid *cluuid,
6057                            struct obd_connect_data *data, void *localdata)
6058 {
6059         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6060         struct lustre_handle  conn;
6061         int                   rc;
6062         ENTRY;
6063
6064         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6065
6066         rc = class_connect(&conn, obd, cluuid);
6067         if (rc)
6068                 RETURN(rc);
6069
6070         *exp = class_conn2export(&conn);
6071
6072         spin_lock(&osd->od_osfs_lock);
6073         osd->od_connects++;
6074         spin_unlock(&osd->od_osfs_lock);
6075
6076         RETURN(0);
6077 }
6078
6079 /*
6080  * once last export (we don't count self-export) disappeared
6081  * osd can be released
6082  */
6083 static int osd_obd_disconnect(struct obd_export *exp)
6084 {
6085         struct obd_device *obd = exp->exp_obd;
6086         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6087         int                rc, release = 0;
6088         ENTRY;
6089
6090         /* Only disconnect the underlying layers on the final disconnect. */
6091         spin_lock(&osd->od_osfs_lock);
6092         osd->od_connects--;
6093         if (osd->od_connects == 0)
6094                 release = 1;
6095         spin_unlock(&osd->od_osfs_lock);
6096
6097         rc = class_disconnect(exp); /* bz 9811 */
6098
6099         if (rc == 0 && release)
6100                 class_manual_cleanup(obd);
6101         RETURN(rc);
6102 }
6103
6104 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6105                        struct lu_device *dev)
6106 {
6107         struct osd_device       *osd    = osd_dev(dev);
6108         struct lr_server_data   *lsd    =
6109                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6110         int                      result = 0;
6111         ENTRY;
6112
6113         if (osd->od_quota_slave != NULL) {
6114                 /* set up quota slave objects */
6115                 result = qsd_prepare(env, osd->od_quota_slave);
6116                 if (result != 0)
6117                         RETURN(result);
6118         }
6119
6120         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6121 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6122                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6123                         osd->od_index_in_idif = 1;
6124                 } else {
6125                         osd->od_index_in_idif = 0;
6126                         result = osd_register_proc_index_in_idif(osd);
6127                         if (result != 0)
6128                                 RETURN(result);
6129                 }
6130 #else
6131                 osd->od_index_in_idif = 1;
6132 #endif
6133         }
6134
6135         result = osd_fid_init(env, osd);
6136
6137         RETURN(result);
6138 }
6139
6140 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6141                          struct lu_fid *fid, struct md_op_data *op_data)
6142 {
6143         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6144
6145         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6146 }
6147
6148 static const struct lu_object_operations osd_lu_obj_ops = {
6149         .loo_object_init      = osd_object_init,
6150         .loo_object_delete    = osd_object_delete,
6151         .loo_object_release   = osd_object_release,
6152         .loo_object_free      = osd_object_free,
6153         .loo_object_print     = osd_object_print,
6154         .loo_object_invariant = osd_object_invariant
6155 };
6156
6157 const struct lu_device_operations osd_lu_ops = {
6158         .ldo_object_alloc      = osd_object_alloc,
6159         .ldo_process_config    = osd_process_config,
6160         .ldo_recovery_complete = osd_recovery_complete,
6161         .ldo_prepare           = osd_prepare,
6162 };
6163
6164 static const struct lu_device_type_operations osd_device_type_ops = {
6165         .ldto_init = osd_type_init,
6166         .ldto_fini = osd_type_fini,
6167
6168         .ldto_start = osd_type_start,
6169         .ldto_stop  = osd_type_stop,
6170
6171         .ldto_device_alloc = osd_device_alloc,
6172         .ldto_device_free  = osd_device_free,
6173
6174         .ldto_device_init    = osd_device_init,
6175         .ldto_device_fini    = osd_device_fini
6176 };
6177
6178 static struct lu_device_type osd_device_type = {
6179         .ldt_tags     = LU_DEVICE_DT,
6180         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6181         .ldt_ops      = &osd_device_type_ops,
6182         .ldt_ctx_tags = LCT_LOCAL,
6183 };
6184
6185 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6186 {
6187         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6188         struct super_block *sb = osd_sb(osd);
6189
6190         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6191 }
6192
6193 /*
6194  * lprocfs legacy support.
6195  */
6196 static struct obd_ops osd_obd_device_ops = {
6197         .o_owner = THIS_MODULE,
6198         .o_connect      = osd_obd_connect,
6199         .o_disconnect   = osd_obd_disconnect,
6200         .o_fid_alloc    = osd_fid_alloc,
6201         .o_health_check = osd_health_check,
6202 };
6203
6204 static int __init osd_mod_init(void)
6205 {
6206         int rc;
6207
6208 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6209         /* please, try to keep osd_thread_info smaller than a page */
6210         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6211 #endif
6212
6213         osd_oi_mod_init();
6214
6215         rc = lu_kmem_init(ldiskfs_caches);
6216         if (rc)
6217                 return rc;
6218
6219         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6220                                  lprocfs_osd_module_vars,
6221                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6222         if (rc)
6223                 lu_kmem_fini(ldiskfs_caches);
6224         return rc;
6225 }
6226
6227 static void __exit osd_mod_exit(void)
6228 {
6229         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6230         lu_kmem_fini(ldiskfs_caches);
6231 }
6232
6233 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6234 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6235 MODULE_VERSION(LUSTRE_VERSION_STRING);
6236 MODULE_LICENSE("GPL");
6237
6238 module_init(osd_mod_init);
6239 module_exit(osd_mod_exit);