lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76
  77 int ldiskfs_pdo = 1;
  78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  79                 "ldiskfs with parallel directory operations");
  80
  81 int ldiskfs_track_declares_assert;
  82 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  83                 "LBUG during tracking of declares");
  84
  85 /* Slab to allocate dynlocks */
  86 struct kmem_cache *dynlock_cachep;
  87
  88 static struct lu_kmem_descr ldiskfs_caches[] = {
  89         {
  90                 .ckd_cache = &dynlock_cachep,
  91                 .ckd_name  = "dynlock_cache",
  92                 .ckd_size  = sizeof(struct dynlock_handle)
  93         },
  94         {
  95                 .ckd_cache = NULL
  96         }
  97 };
  98
  99 static const char dot[] = ".";
 100 static const char dotdot[] = "..";
 101 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 102
 103 static const struct lu_object_operations      osd_lu_obj_ops;
 104 static const struct dt_object_operations      osd_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ea_ops;
 106 static const struct dt_object_operations      osd_obj_otable_it_ops;
 107 static const struct dt_index_operations       osd_index_iam_ops;
 108 static const struct dt_index_operations       osd_index_ea_ops;
 109
 110 int osd_trans_declare_op2rb[] = {
 111         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 112         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 113         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 114         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 115         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 116         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 117         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 118         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 119         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 120         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 121         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 122         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 123 };
 124
 125 static int osd_has_index(const struct osd_object *obj)
 126 {
 127         return obj->oo_dt.do_index_ops != NULL;
 128 }
 129
 130 static int osd_object_invariant(const struct lu_object *l)
 131 {
 132         return osd_invariant(osd_obj(l));
 133 }
 134
 135 /*
 136  * Concurrency: doesn't matter
 137  */
 138
 139 /*
 140  * Concurrency: doesn't matter
 141  */
 142 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 143 {
 144         struct osd_thread_info *oti = osd_oti_get(env);
 145         return oti->oti_w_locks > 0 && o->oo_owner == env;
 146 }
 147
 148 /*
 149  * Concurrency: doesn't access mutable data
 150  */
 151 static int osd_root_get(const struct lu_env *env,
 152                         struct dt_device *dev, struct lu_fid *f)
 153 {
 154         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 155         return 0;
 156 }
 157
 158 /*
 159  * OSD object methods.
 160  */
 161
 162 /*
 163  * Concurrency: no concurrent access is possible that early in object
 164  * life-cycle.
 165  */
 166 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 167                                           const struct lu_object_header *hdr,
 168                                           struct lu_device *d)
 169 {
 170         struct osd_object *mo;
 171
 172         OBD_ALLOC_PTR(mo);
 173         if (mo != NULL) {
 174                 struct lu_object *l;
 175
 176                 l = &mo->oo_dt.do_lu;
 177                 dt_object_init(&mo->oo_dt, NULL, d);
 178                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 179                 l->lo_ops = &osd_lu_obj_ops;
 180                 init_rwsem(&mo->oo_sem);
 181                 init_rwsem(&mo->oo_ext_idx_sem);
 182                 spin_lock_init(&mo->oo_guard);
 183                 return l;
 184         } else {
 185                 return NULL;
 186         }
 187 }
 188
 189 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 190                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 191 {
 192         int rc;
 193
 194         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 195         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 196                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 197         if (rc > 0) {
 198                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 199                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 200                 rc = 0;
 201                 lustre_lma_swab(lma);
 202                 /* Check LMA compatibility */
 203                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 204                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 205                               "for fid = "DFID", ino = %lu\n",
 206                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 207                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 208                               PFID(&lma->lma_self_fid), inode->i_ino);
 209                         rc = -EOPNOTSUPP;
 210                 }
 211         } else if (rc == 0) {
 212                 rc = -ENODATA;
 213         }
 214
 215         return rc;
 216 }
 217
 218 /*
 219  * retrieve object from backend ext fs.
 220  **/
 221 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 222                        struct osd_inode_id *id)
 223 {
 224         struct inode *inode = NULL;
 225
 226         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 227         if (IS_ERR(inode)) {
 228                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 229                        id->oii_ino, PTR_ERR(inode));
 230         } else if (id->oii_gen != OSD_OII_NOGEN &&
 231                    inode->i_generation != id->oii_gen) {
 232                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 233                        "i_generation = %u\n",
 234                        id->oii_ino, id->oii_gen, inode->i_generation);
 235                 iput(inode);
 236                 inode = ERR_PTR(-ESTALE);
 237         } else if (inode->i_nlink == 0) {
 238                 /* due to parallel readdir and unlink,
 239                 * we can have dead inode here. */
 240                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 241                 iput(inode);
 242                 inode = ERR_PTR(-ESTALE);
 243         } else if (is_bad_inode(inode)) {
 244                 CWARN("%.16s: bad inode: ino = %u\n",
 245                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 246                 iput(inode);
 247                 inode = ERR_PTR(-ENOENT);
 248         } else {
 249                 if (id->oii_gen == OSD_OII_NOGEN)
 250                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 251
 252                 /* Do not update file c/mtime in ldiskfs.
 253                  * NB: we don't have any lock to protect this because we don't
 254                  * have reference on osd_object now, but contention with
 255                  * another lookup + attr_set can't happen in the tiny window
 256                  * between if (...) and set S_NOCMTIME. */
 257                 if (!(inode->i_flags & S_NOCMTIME))
 258                         inode->i_flags |= S_NOCMTIME;
 259         }
 260         return inode;
 261 }
 262
 263 static struct inode *
 264 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 265              struct osd_inode_id *id, struct lu_fid *fid)
 266 {
 267         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 268         struct inode            *inode;
 269         int                      rc;
 270
 271         inode = osd_iget(info, dev, id);
 272         if (IS_ERR(inode))
 273                 return inode;
 274
 275         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 276         if (rc == 0) {
 277                 *fid = lma->lma_self_fid;
 278         } else if (rc == -ENODATA) {
 279                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 280                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 281                 else
 282                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 283         } else {
 284                 iput(inode);
 285                 inode = ERR_PTR(rc);
 286         }
 287         return inode;
 288 }
 289
 290 static struct inode *osd_iget_check(struct osd_thread_info *info,
 291                                     struct osd_device *dev,
 292                                     const struct lu_fid *fid,
 293                                     struct osd_inode_id *id,
 294                                     bool in_oi)
 295 {
 296         struct inode    *inode;
 297         int              rc     = 0;
 298         ENTRY;
 299
 300         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 301         if (IS_ERR(inode)) {
 302                 rc = PTR_ERR(inode);
 303                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 304                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 305                                id->oii_ino, rc);
 306
 307                         GOTO(put, rc);
 308                 }
 309
 310                 goto check_oi;
 311         }
 312
 313         if (is_bad_inode(inode)) {
 314                 rc = -ENOENT;
 315                 if (!in_oi) {
 316                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 317
 318                         GOTO(put, rc);
 319                 }
 320
 321                 goto check_oi;
 322         }
 323
 324         if (id->oii_gen != OSD_OII_NOGEN &&
 325             inode->i_generation != id->oii_gen) {
 326                 rc = -ESTALE;
 327                 if (!in_oi) {
 328                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 329                                "oii_gen = %u, i_generation = %u\n",
 330                                id->oii_ino, id->oii_gen, inode->i_generation);
 331
 332                         GOTO(put, rc);
 333                 }
 334
 335                 goto check_oi;
 336         }
 337
 338         if (inode->i_nlink == 0) {
 339                 rc = -ENOENT;
 340                 if (!in_oi) {
 341                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 342
 343                         GOTO(put, rc);
 344                 }
 345
 346                 goto check_oi;
 347         }
 348
 349 check_oi:
 350         if (rc != 0) {
 351                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 352
 353                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 354                 /* XXX: There are some possible cases:
 355                  *      1. rc = 0.
 356                  *         Backup/restore caused the OI invalid.
 357                  *      2. rc = 0.
 358                  *         Someone unlinked the object but NOT removed
 359                  *         the OI mapping, such as mount target device
 360                  *         as ldiskfs, and modify something directly.
 361                  *      3. rc = -ENOENT.
 362                  *         Someone just removed the object between the
 363                  *         former oi_lookup and the iget. It is normal.
 364                  *      4. Other failure cases.
 365                  *
 366                  *      Generally, when the device is mounted, it will
 367                  *      auto check whether the system is restored from
 368                  *      file-level backup or not. We trust such detect
 369                  *      to distinguish the 1st case from the 2nd case. */
 370                 if (rc == 0) {
 371                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 372                             inode->i_generation == id->oii_gen)
 373                                 rc = -ENOENT;
 374                         else
 375                                 rc = -EREMCHG;
 376                 }
 377         } else {
 378                 if (id->oii_gen == OSD_OII_NOGEN)
 379                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 380
 381                 /* Do not update file c/mtime in ldiskfs.
 382                  * NB: we don't have any lock to protect this because we don't
 383                  * have reference on osd_object now, but contention with
 384                  * another lookup + attr_set can't happen in the tiny window
 385                  * between if (...) and set S_NOCMTIME. */
 386                 if (!(inode->i_flags & S_NOCMTIME))
 387                         inode->i_flags |= S_NOCMTIME;
 388         }
 389
 390         GOTO(put, rc);
 391
 392 put:
 393         if (rc != 0) {
 394                 if (!IS_ERR(inode))
 395                         iput(inode);
 396
 397                 inode = ERR_PTR(rc);
 398         }
 399
 400         return inode;
 401 }
 402
 403 /**
 404  * \retval +v: new filter_fid, does not contain self-fid
 405  * \retval 0:  filter_fid_old, contains self-fid
 406  * \retval -v: other failure cases
 407  */
 408 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 409                  struct dentry *dentry, struct lu_fid *fid)
 410 {
 411         struct filter_fid_old   *ff     = &info->oti_ff;
 412         struct ost_id           *ostid  = &info->oti_ostid;
 413         int                      rc;
 414
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 rc = 0;
 418                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 419                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 420                 /* XXX: should use real OST index in the future. LU-3569 */
 421                 ostid_to_fid(fid, ostid, 0);
 422         } else if (rc == sizeof(struct filter_fid)) {
 423                 rc = 1;
 424         } else if (rc >= 0) {
 425                 rc = -EINVAL;
 426         }
 427
 428         return rc;
 429 }
 430
 431 static int osd_lma_self_repair(struct osd_thread_info *info,
 432                                struct osd_device *osd, struct inode *inode,
 433                                const struct lu_fid *fid, __u32 compat)
 434 {
 435         handle_t *jh;
 436         int       rc;
 437
 438         LASSERT(current->journal_info == NULL);
 439
 440         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 441                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 442         if (IS_ERR(jh)) {
 443                 rc = PTR_ERR(jh);
 444                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 445                       osd_name(osd), rc);
 446                 return rc;
 447         }
 448
 449         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 450         if (rc != 0)
 451                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 452                       osd_name(osd), rc);
 453         ldiskfs_journal_stop(jh);
 454         return rc;
 455 }
 456
 457 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 458 {
 459         struct osd_thread_info  *info   = osd_oti_get(env);
 460         struct osd_device       *osd    = osd_obj2dev(obj);
 461         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 462         struct inode            *inode  = obj->oo_inode;
 463         struct dentry           *dentry = &info->oti_obj_dentry;
 464         struct lu_fid           *fid    = NULL;
 465         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 466         int                      rc;
 467         ENTRY;
 468
 469         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 470         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 471                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 472         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 473                 fid = &lma->lma_self_fid;
 474                 rc = osd_get_idif(info, inode, dentry, fid);
 475                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 476                         /* For the given OST-object, if it has neither LMA nor
 477                          * FID in XATTR_NAME_FID, then the given FID (which is
 478                          * contained in the @obj, from client RPC for locating
 479                          * the OST-object) is trusted. We use it to generate
 480                          * the LMA. */
 481                         osd_lma_self_repair(info, osd, inode, rfid,
 482                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 483                                 LMAC_FID_ON_OST : 0);
 484                         RETURN(0);
 485                 }
 486         }
 487
 488         if (unlikely(rc == -ENODATA))
 489                 RETURN(0);
 490
 491         if (rc < 0)
 492                 RETURN(rc);
 493
 494         if (rc > 0) {
 495                 rc = 0;
 496                 lustre_lma_swab(lma);
 497                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 498                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 499                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 500                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 501                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 502                               PFID(rfid), inode->i_ino);
 503                         rc = -EOPNOTSUPP;
 504                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 505                         fid = &lma->lma_self_fid;
 506                 }
 507         }
 508
 509         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 510                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 511                         struct ost_id   *oi   = &info->oti_ostid;
 512                         struct lu_fid   *fid1 = &info->oti_fid3;
 513                         __u32            idx  = fid_idif_ost_idx(rfid);
 514
 515                         /* For old IDIF, the OST index is not part of the IDIF,
 516                          * Means that different OSTs may have the same IDIFs.
 517                          * Under such case, we need to make some compatible
 518                          * check to make sure to trigger OI scrub properly. */
 519                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 520                                 /* Given @rfid is new, LMA is old. */
 521                                 fid_to_ostid(fid, oi);
 522                                 ostid_to_fid(fid1, oi, idx);
 523                                 if (lu_fid_eq(fid1, rfid)) {
 524                                         if (osd->od_lma_self_repair)
 525                                                 osd_lma_self_repair(info, osd,
 526                                                         inode, rfid,
 527                                                         LMAC_FID_ON_OST);
 528                                         RETURN(0);
 529                                 }
 530                         }
 531                 }
 532
 533                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 534                        osd_name(osd), PFID(rfid), PFID(fid));
 535                 rc = -EREMCHG;
 536         }
 537
 538         RETURN(rc);
 539 }
 540
 541 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 542                           const struct lu_fid *fid,
 543                           const struct lu_object_conf *conf)
 544 {
 545         struct osd_thread_info *info;
 546         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 547         struct osd_device      *dev;
 548         struct osd_idmap_cache *oic;
 549         struct osd_inode_id    *id;
 550         struct inode           *inode;
 551         struct osd_scrub       *scrub;
 552         struct scrub_file      *sf;
 553         int                     result;
 554         int                     saved  = 0;
 555         bool                    in_oi  = false;
 556         bool                    triggered = false;
 557         ENTRY;
 558
 559         LINVRNT(osd_invariant(obj));
 560         LASSERT(obj->oo_inode == NULL);
 561         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 562
 563         dev = osd_dev(ldev);
 564         scrub = &dev->od_scrub;
 565         sf = &scrub->os_file;
 566         info = osd_oti_get(env);
 567         LASSERT(info);
 568         oic = &info->oti_cache;
 569
 570         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 571                 RETURN(-ENOENT);
 572
 573         /* For the object is created as locking anchor, or for the object to
 574          * be created on disk. No need to osd_oi_lookup() at here because FID
 575          * shouldn't never be re-used, if it's really a duplicate FID from
 576          * unexpected reason, we should be able to detect it later by calling
 577          * do_create->osd_oi_insert(). */
 578         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 579                 GOTO(out, result = 0);
 580
 581         /* Search order: 1. per-thread cache. */
 582         if (lu_fid_eq(fid, &oic->oic_fid) &&
 583             likely(oic->oic_dev == dev)) {
 584                 id = &oic->oic_lid;
 585                 goto iget;
 586         }
 587
 588         id = &info->oti_id;
 589         if (!list_empty(&scrub->os_inconsistent_items)) {
 590                 /* Search order: 2. OI scrub pending list. */
 591                 result = osd_oii_lookup(dev, fid, id);
 592                 if (result == 0)
 593                         goto iget;
 594         }
 595
 596         /* Search order: 3. OI files. */
 597         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 598         if (result == -ENOENT) {
 599                 if (!fid_is_norm(fid) ||
 600                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 601                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 602                                       sf->sf_oi_bitmap))
 603                         GOTO(out, result = 0);
 604
 605                 goto trigger;
 606         }
 607
 608         if (result != 0)
 609                 GOTO(out, result);
 610
 611         in_oi = true;
 612
 613 iget:
 614         inode = osd_iget_check(info, dev, fid, id, in_oi);
 615         if (IS_ERR(inode)) {
 616                 result = PTR_ERR(inode);
 617                 if (result == -ENOENT || result == -ESTALE) {
 618                         if (!in_oi)
 619                                 fid_zero(&oic->oic_fid);
 620
 621                         GOTO(out, result = -ENOENT);
 622                 } else if (result == -EREMCHG) {
 623
 624 trigger:
 625                         if (!in_oi)
 626                                 fid_zero(&oic->oic_fid);
 627
 628                         if (unlikely(triggered))
 629                                 GOTO(out, result = saved);
 630
 631                         triggered = true;
 632                         if (thread_is_running(&scrub->os_thread)) {
 633                                 result = -EINPROGRESS;
 634                         } else if (!dev->od_noscrub) {
 635                                 result = osd_scrub_start(dev);
 636                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 637                                               "for "DFID", rc = %d [1]\n",
 638                                               osd_name(dev), PFID(fid),result);
 639                                 if (result == 0 || result == -EALREADY)
 640                                         result = -EINPROGRESS;
 641                                 else
 642                                         result = -EREMCHG;
 643                         }
 644
 645                         /* We still have chance to get the valid inode: for the
 646                          * object which is referenced by remote name entry, the
 647                          * object on the local MDT will be linked under the dir
 648                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 649                          *
 650                          * We do not know whether the object for the given FID
 651                          * is referenced by some remote name entry or not, and
 652                          * especially for DNE II, a multiple-linked object may
 653                          * have many name entries reside on many MDTs.
 654                          *
 655                          * To simplify the operation, OSD will not distinguish
 656                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 657                          * only happened for the RPC from other MDT during the
 658                          * OI scrub, or for the client side RPC with FID only,
 659                          * such as FID to path, or from old connected client. */
 660                         saved = result;
 661                         result = osd_lookup_in_remote_parent(info, dev,
 662                                                              fid, id);
 663                         if (result == 0) {
 664                                 in_oi = false;
 665                                 goto iget;
 666                         }
 667
 668                         result = saved;
 669                 }
 670
 671                 GOTO(out, result);
 672         }
 673
 674         obj->oo_inode = inode;
 675         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 676
 677         result = osd_check_lma(env, obj);
 678         if (result != 0) {
 679                 iput(inode);
 680                 obj->oo_inode = NULL;
 681                 if (result == -EREMCHG) {
 682                         if (!in_oi) {
 683                                 result = osd_oi_lookup(info, dev, fid, id,
 684                                                        OI_CHECK_FLD);
 685                                 if (result != 0) {
 686                                         fid_zero(&oic->oic_fid);
 687                                         GOTO(out, result);
 688                                 }
 689                         }
 690
 691                         goto trigger;
 692                 }
 693
 694                 GOTO(out, result);
 695         }
 696
 697         obj->oo_compat_dot_created = 1;
 698         obj->oo_compat_dotdot_created = 1;
 699
 700         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 701                 GOTO(out, result = 0);
 702
 703         LASSERT(obj->oo_hl_head == NULL);
 704         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 705         if (obj->oo_hl_head == NULL) {
 706                 obj->oo_inode = NULL;
 707                 iput(inode);
 708                 GOTO(out, result = -ENOMEM);
 709         }
 710         GOTO(out, result = 0);
 711
 712 out:
 713         LINVRNT(osd_invariant(obj));
 714         return result;
 715 }
 716
 717 /*
 718  * Concurrency: shouldn't matter.
 719  */
 720 static void osd_object_init0(struct osd_object *obj)
 721 {
 722         LASSERT(obj->oo_inode != NULL);
 723         obj->oo_dt.do_body_ops = &osd_body_ops;
 724         obj->oo_dt.do_lu.lo_header->loh_attr |=
 725                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 726 }
 727
 728 /*
 729  * Concurrency: no concurrent access is possible that early in object
 730  * life-cycle.
 731  */
 732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 733                            const struct lu_object_conf *conf)
 734 {
 735         struct osd_object *obj = osd_obj(l);
 736         int result;
 737
 738         LINVRNT(osd_invariant(obj));
 739
 740         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 741                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 742                 l->lo_header->loh_attr |= LOHA_EXISTS;
 743                 return 0;
 744         }
 745
 746         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 747         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 748         if (result == 0 && obj->oo_inode != NULL)
 749                 osd_object_init0(obj);
 750
 751         LINVRNT(osd_invariant(obj));
 752         return result;
 753 }
 754
 755 /*
 756  * Concurrency: no concurrent access is possible that late in object
 757  * life-cycle.
 758  */
 759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 760 {
 761         struct osd_object *obj = osd_obj(l);
 762
 763         LINVRNT(osd_invariant(obj));
 764
 765         dt_object_fini(&obj->oo_dt);
 766         if (obj->oo_hl_head != NULL)
 767                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 768         OBD_FREE_PTR(obj);
 769 }
 770
 771 /*
 772  * Concurrency: no concurrent access is possible that late in object
 773  * life-cycle.
 774  */
 775 static void osd_index_fini(struct osd_object *o)
 776 {
 777         struct iam_container *bag;
 778
 779         if (o->oo_dir != NULL) {
 780                 bag = &o->oo_dir->od_container;
 781                 if (o->oo_inode != NULL) {
 782                         if (bag->ic_object == o->oo_inode)
 783                                 iam_container_fini(bag);
 784                 }
 785                 OBD_FREE_PTR(o->oo_dir);
 786                 o->oo_dir = NULL;
 787         }
 788 }
 789
 790 /*
 791  * Concurrency: no concurrent access is possible that late in object
 792  * life-cycle (for all existing callers, that is. New callers have to provide
 793  * their own locking.)
 794  */
 795 static int osd_inode_unlinked(const struct inode *inode)
 796 {
 797         return inode->i_nlink == 0;
 798 }
 799
 800 enum {
 801         OSD_TXN_OI_DELETE_CREDITS    = 20,
 802         OSD_TXN_INODE_DELETE_CREDITS = 20
 803 };
 804
 805 /*
 806  * Journal
 807  */
 808
 809 #if OSD_THANDLE_STATS
 810 /**
 811  * Set time when the handle is allocated
 812  */
 813 static void osd_th_alloced(struct osd_thandle *oth)
 814 {
 815         oth->oth_alloced = cfs_time_current();
 816 }
 817
 818 /**
 819  * Set time when the handle started
 820  */
 821 static void osd_th_started(struct osd_thandle *oth)
 822 {
 823         oth->oth_started = cfs_time_current();
 824 }
 825
 826 /**
 827  * Helper function to convert time interval to microseconds packed in
 828  * long int.
 829  */
 830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 831 {
 832         struct timeval val;
 833
 834         cfs_duration_usec(cfs_time_sub(end, start), &val);
 835         return val.tv_sec * 1000000 + val.tv_usec;
 836 }
 837
 838 /**
 839  * Check whether the we deal with this handle for too long.
 840  */
 841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 842                                 cfs_time_t alloced, cfs_time_t started,
 843                                 cfs_time_t closed)
 844 {
 845         cfs_time_t now = cfs_time_current();
 846
 847         LASSERT(dev != NULL);
 848
 849         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 850                             interval_to_usec(alloced, started));
 851         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 852                             interval_to_usec(started, closed));
 853         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 854                             interval_to_usec(closed, now));
 855
 856         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 857                 CWARN("transaction handle %p was open for too long: "
 858                       "now "CFS_TIME_T" ,"
 859                       "alloced "CFS_TIME_T" ,"
 860                       "started "CFS_TIME_T" ,"
 861                       "closed "CFS_TIME_T"\n",
 862                       oth, now, alloced, started, closed);
 863                 libcfs_debug_dumpstack(NULL);
 864         }
 865 }
 866
 867 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 868 {                                                                       \
 869         cfs_time_t __closed = cfs_time_current();                       \
 870         cfs_time_t __alloced = oth->oth_alloced;                        \
 871         cfs_time_t __started = oth->oth_started;                        \
 872                                                                         \
 873         expr;                                                           \
 874         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 875 }
 876
 877 #else /* OSD_THANDLE_STATS */
 878
 879 #define osd_th_alloced(h)                  do {} while(0)
 880 #define osd_th_started(h)                  do {} while(0)
 881 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 882
 883 #endif /* OSD_THANDLE_STATS */
 884
 885 /*
 886  * Concurrency: doesn't access mutable data.
 887  */
 888 static int osd_param_is_not_sane(const struct osd_device *dev,
 889                                  const struct thandle *th)
 890 {
 891         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 892
 893         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 894 }
 895
 896 /*
 897  * Concurrency: shouldn't matter.
 898  */
 899 static void osd_trans_commit_cb(struct super_block *sb,
 900                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 901 {
 902         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 903         struct thandle     *th  = &oh->ot_super;
 904         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 905         struct dt_txn_commit_cb *dcb, *tmp;
 906
 907         LASSERT(oh->ot_handle == NULL);
 908
 909         if (error)
 910                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 911
 912         dt_txn_hook_commit(th);
 913
 914         /* call per-transaction callbacks if any */
 915         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 916                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 917                          "commit callback entry: magic=%x name='%s'\n",
 918                          dcb->dcb_magic, dcb->dcb_name);
 919                 list_del_init(&dcb->dcb_linkage);
 920                 dcb->dcb_func(NULL, th, dcb, error);
 921         }
 922
 923         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 924         lu_device_put(lud);
 925         th->th_dev = NULL;
 926
 927         lu_context_exit(&th->th_ctx);
 928         lu_context_fini(&th->th_ctx);
 929         thandle_put(th);
 930 }
 931
 932 static struct thandle *osd_trans_create(const struct lu_env *env,
 933                                         struct dt_device *d)
 934 {
 935         struct osd_thread_info  *oti = osd_oti_get(env);
 936         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 937         struct osd_thandle      *oh;
 938         struct thandle          *th;
 939         ENTRY;
 940
 941         /* on pending IO in this thread should left from prev. request */
 942         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 943
 944         th = ERR_PTR(-ENOMEM);
 945         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 946         if (oh != NULL) {
 947                 oh->ot_quota_trans = &oti->oti_quota_trans;
 948                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 949                 th = &oh->ot_super;
 950                 th->th_dev = d;
 951                 th->th_result = 0;
 952                 th->th_tags = LCT_TX_HANDLE;
 953                 oh->ot_credits = 0;
 954                 atomic_set(&th->th_refc, 1);
 955                 th->th_alloc_size = sizeof(*oh);
 956                 oti->oti_dev = osd_dt_dev(d);
 957                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 958                 osd_th_alloced(oh);
 959
 960                 memset(oti->oti_declare_ops, 0,
 961                                         sizeof(oti->oti_declare_ops));
 962                 memset(oti->oti_declare_ops_rb, 0,
 963                                         sizeof(oti->oti_declare_ops_rb));
 964                 memset(oti->oti_declare_ops_cred, 0,
 965                                         sizeof(oti->oti_declare_ops_cred));
 966                 oti->oti_rollback = false;
 967         }
 968         RETURN(th);
 969 }
 970
 971 /*
 972  * Concurrency: shouldn't matter.
 973  */
 974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 975                     struct thandle *th)
 976 {
 977         struct osd_thread_info *oti = osd_oti_get(env);
 978         struct osd_device  *dev = osd_dt_dev(d);
 979         handle_t           *jh;
 980         struct osd_thandle *oh;
 981         int rc;
 982
 983         ENTRY;
 984
 985         LASSERT(current->journal_info == NULL);
 986
 987         oh = container_of0(th, struct osd_thandle, ot_super);
 988         LASSERT(oh != NULL);
 989         LASSERT(oh->ot_handle == NULL);
 990
 991         rc = dt_txn_hook_start(env, d, th);
 992         if (rc != 0)
 993                 GOTO(out, rc);
 994
 995         if (unlikely(osd_param_is_not_sane(dev, th))) {
 996                 static unsigned long last_printed;
 997                 static int last_credits;
 998
 999                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1001                       oh->ot_credits,
1002                       osd_journal(dev)->j_max_transaction_buffers);
1003                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1004                       oti->oti_declare_ops[OSD_OT_CREATE],
1005                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006                       oti->oti_declare_ops[OSD_OT_DESTROY],
1007                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_WRITE],
1015                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016                       oti->oti_declare_ops[OSD_OT_PUNCH],
1017                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops[OSD_OT_QUOTA],
1019                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1021                       oti->oti_declare_ops[OSD_OT_INSERT],
1022                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023                       oti->oti_declare_ops[OSD_OT_DELETE],
1024                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1027                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1029                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1030
1031                 if (last_credits != oh->ot_credits &&
1032                     time_after(jiffies, last_printed + 60 * HZ)) {
1033                         libcfs_debug_dumpstack(NULL);
1034                         last_credits = oh->ot_credits;
1035                         last_printed = jiffies;
1036                 }
1037                 /* XXX Limit the credits to 'max_transaction_buffers', and
1038                  *     let the underlying filesystem to catch the error if
1039                  *     we really need so many credits.
1040                  *
1041                  *     This should be removed when we can calculate the
1042                  *     credits precisely. */
1043                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1044         }
1045
1046         /*
1047          * XXX temporary stuff. Some abstraction layer should
1048          * be used.
1049          */
1050         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1051         osd_th_started(oh);
1052         if (!IS_ERR(jh)) {
1053                 oh->ot_handle = jh;
1054                 LASSERT(oti->oti_txns == 0);
1055                 lu_context_init(&th->th_ctx, th->th_tags);
1056                 lu_context_enter(&th->th_ctx);
1057
1058                 lu_device_get(&d->dd_lu_dev);
1059                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1060                               "osd-tx", th);
1061                 oti->oti_txns++;
1062                 rc = 0;
1063         } else {
1064                 rc = PTR_ERR(jh);
1065         }
1066 out:
1067         RETURN(rc);
1068 }
1069
1070 static int osd_seq_exists(const struct lu_env *env,
1071                               struct osd_device *osd, obd_seq seq)
1072 {
1073         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1074         struct seq_server_site  *ss = osd_seq_site(osd);
1075         int                     rc;
1076         ENTRY;
1077
1078         LASSERT(ss != NULL);
1079         LASSERT(ss->ss_server_fld != NULL);
1080
1081         rc = osd_fld_lookup(env, osd, seq, range);
1082         if (rc != 0) {
1083                 if (rc != -ENOENT)
1084                         CERROR("%s: can't lookup FLD sequence "LPX64
1085                                ": rc = %d\n", osd_name(osd), seq, rc);
1086                 RETURN(0);
1087         }
1088
1089         RETURN(ss->ss_node_id == range->lsr_index);
1090 }
1091
1092 /*
1093  * Concurrency: shouldn't matter.
1094  */
1095 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1096                           struct thandle *th)
1097 {
1098         int                     rc = 0;
1099         struct osd_thandle     *oh;
1100         struct osd_thread_info *oti = osd_oti_get(env);
1101         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1102         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1103         struct lquota_trans    *qtrans;
1104         ENTRY;
1105
1106         oh = container_of0(th, struct osd_thandle, ot_super);
1107
1108         qtrans = oh->ot_quota_trans;
1109         oh->ot_quota_trans = NULL;
1110
1111         if (oh->ot_handle != NULL) {
1112                 handle_t *hdl = oh->ot_handle;
1113
1114                 /*
1115                  * add commit callback
1116                  * notice we don't do this in osd_trans_start()
1117                  * as underlying transaction can change during truncate
1118                  */
1119                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1120                                          &oh->ot_jcb);
1121
1122                 LASSERT(oti->oti_txns == 1);
1123                 oti->oti_txns--;
1124                 rc = dt_txn_hook_stop(env, th);
1125                 if (rc != 0)
1126                         CERROR("Failure in transaction hook: %d\n", rc);
1127
1128                 /* hook functions might modify th_sync */
1129                 hdl->h_sync = th->th_sync;
1130
1131                 oh->ot_handle = NULL;
1132                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1133                                   rc = ldiskfs_journal_stop(hdl));
1134                 if (rc != 0)
1135                         CERROR("Failure to stop transaction: %d\n", rc);
1136         } else {
1137                 thandle_put(&oh->ot_super);
1138         }
1139
1140         /* inform the quota slave device that the transaction is stopping */
1141         qsd_op_end(env, qsd, qtrans);
1142
1143         /* as we want IO to journal and data IO be concurrent, we don't block
1144          * awaiting data IO completion in osd_do_bio(), instead we wait here
1145          * once transaction is submitted to the journal. all reqular requests
1146          * don't do direct IO (except read/write), thus this wait_event becomes
1147          * no-op for them.
1148          *
1149          * IMPORTANT: we have to wait till any IO submited by the thread is
1150          * completed otherwise iobuf may be corrupted by different request
1151          */
1152         wait_event(iobuf->dr_wait,
1153                        atomic_read(&iobuf->dr_numreqs) == 0);
1154         osd_fini_iobuf(oti->oti_dev, iobuf);
1155         if (!rc)
1156                 rc = iobuf->dr_error;
1157
1158         RETURN(rc);
1159 }
1160
1161 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 {
1163         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1164                                                ot_super);
1165
1166         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1167         LASSERT(&dcb->dcb_func != NULL);
1168         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1169
1170         return 0;
1171 }
1172
1173 /*
1174  * Called just before object is freed. Releases all resources except for
1175  * object itself (that is released by osd_object_free()).
1176  *
1177  * Concurrency: no concurrent access is possible that late in object
1178  * life-cycle.
1179  */
1180 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 {
1182         struct osd_object *obj   = osd_obj(l);
1183         struct inode      *inode = obj->oo_inode;
1184
1185         LINVRNT(osd_invariant(obj));
1186
1187         /*
1188          * If object is unlinked remove fid->ino mapping from object index.
1189          */
1190
1191         osd_index_fini(obj);
1192         if (inode != NULL) {
1193                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1194                 qid_t                    uid = i_uid_read(inode);
1195                 qid_t                    gid = i_gid_read(inode);
1196
1197                 iput(inode);
1198                 obj->oo_inode = NULL;
1199
1200                 if (qsd != NULL) {
1201                         struct osd_thread_info  *info = osd_oti_get(env);
1202                         struct lquota_id_info   *qi = &info->oti_qi;
1203
1204                         /* Release granted quota to master if necessary */
1205                         qi->lqi_id.qid_uid = uid;
1206                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207
1208                         qi->lqi_id.qid_uid = gid;
1209                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1210                 }
1211         }
1212 }
1213
1214 /*
1215  * Concurrency: ->loo_object_release() is called under site spin-lock.
1216  */
1217 static void osd_object_release(const struct lu_env *env,
1218                                struct lu_object *l)
1219 {
1220 }
1221
1222 /*
1223  * Concurrency: shouldn't matter.
1224  */
1225 static int osd_object_print(const struct lu_env *env, void *cookie,
1226                             lu_printer_t p, const struct lu_object *l)
1227 {
1228         struct osd_object *o = osd_obj(l);
1229         struct iam_descr  *d;
1230
1231         if (o->oo_dir != NULL)
1232                 d = o->oo_dir->od_container.ic_descr;
1233         else
1234                 d = NULL;
1235         return (*p)(env, cookie,
1236                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237                     o, o->oo_inode,
1238                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1239                     o->oo_inode ? o->oo_inode->i_generation : 0,
1240                     d ? d->id_ops->id_name : "plain");
1241 }
1242
1243 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1244
1245 /*
1246  * Concurrency: shouldn't matter.
1247  */
1248 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1249                struct obd_statfs *sfs)
1250 {
1251         struct osd_device  *osd = osd_dt_dev(d);
1252         struct super_block *sb = osd_sb(osd);
1253         struct kstatfs     *ksfs;
1254         int result = 0;
1255
1256         if (unlikely(osd->od_mnt == NULL))
1257                 return -EINPROGRESS;
1258
1259         /* osd_lproc.c call this without env, allocate ksfs for that case */
1260         if (unlikely(env == NULL)) {
1261                 OBD_ALLOC_PTR(ksfs);
1262                 if (ksfs == NULL)
1263                         return -ENOMEM;
1264         } else {
1265                 ksfs = &osd_oti_get(env)->oti_ksfs;
1266         }
1267
1268         spin_lock(&osd->od_osfs_lock);
1269         result = sb->s_op->statfs(sb->s_root, ksfs);
1270         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1271                 statfs_pack(sfs, ksfs);
1272                 if (sb->s_flags & MS_RDONLY)
1273                         sfs->os_state = OS_STATE_READONLY;
1274         }
1275
1276         spin_unlock(&osd->od_osfs_lock);
1277
1278         if (unlikely(env == NULL))
1279                 OBD_FREE_PTR(ksfs);
1280
1281         /* Reserve a small amount of space for local objects like last_rcvd,
1282          * llog, quota files, ... */
1283         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1284                 sfs->os_bavail = 0;
1285         } else {
1286                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1287                 /** Take out metadata overhead for indirect blocks */
1288                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1289         }
1290
1291         return result;
1292 }
1293
1294 /**
1295  * Estimate space needed for file creations. We assume the largest filename
1296  * which is 2^64 - 1, hence a filename of 20 chars.
1297  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298  */
1299 #ifdef __LDISKFS_DIR_REC_LEN
1300 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #else
1302 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1303 #endif
1304
1305 /*
1306  * Concurrency: doesn't access mutable data.
1307  */
1308 static void osd_conf_get(const struct lu_env *env,
1309                          const struct dt_device *dev,
1310                          struct dt_device_param *param)
1311 {
1312         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1313         int                ea_overhead;
1314
1315         /*
1316          * XXX should be taken from not-yet-existing fs abstraction layer.
1317          */
1318         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1319         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1320         param->ddp_block_shift  = sb->s_blocksize_bits;
1321         param->ddp_mount_type     = LDD_MT_LDISKFS;
1322         param->ddp_maxbytes       = sb->s_maxbytes;
1323         /* Overhead estimate should be fairly accurate, so we really take a tiny
1324          * error margin which also avoids fragmenting the filesystem too much */
1325         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1326         /* inode are statically allocated, so per-inode space consumption
1327          * is the space consumed by the directory entry */
1328         param->ddp_inodespace     = PER_OBJ_USAGE;
1329         /* per-fragment overhead to be used by the client code */
1330         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1331         param->ddp_mntopts      = 0;
1332         if (test_opt(sb, XATTR_USER))
1333                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1334         if (test_opt(sb, POSIX_ACL))
1335                 param->ddp_mntopts |= MNTOPT_ACL;
1336
1337         /* LOD might calculate the max stripe count based on max_ea_size,
1338          * so we need take account in the overhead as well,
1339          * xattr_header + magic + xattr_entry_head */
1340         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1341                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342
1343 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1344         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1345                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1346                                                                 ea_overhead;
1347         else
1348 #endif
1349                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1350 }
1351
1352 /*
1353  * Concurrency: shouldn't matter.
1354  */
1355 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 {
1357         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1358         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1359 }
1360
1361 /**
1362  * Start commit for OSD device.
1363  *
1364  * An implementation of dt_commit_async method for OSD device.
1365  * Asychronously starts underlayng fs sync and thereby a transaction
1366  * commit.
1367  *
1368  * \param env environment
1369  * \param d dt device
1370  *
1371  * \see dt_device_operations
1372  */
1373 static int osd_commit_async(const struct lu_env *env,
1374                             struct dt_device *d)
1375 {
1376         struct super_block *s = osd_sb(osd_dt_dev(d));
1377         ENTRY;
1378
1379         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1380         RETURN(s->s_op->sync_fs(s, 0));
1381 }
1382
1383 /*
1384  * Concurrency: shouldn't matter.
1385  */
1386
1387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 {
1389         struct super_block *sb = osd_sb(osd_dt_dev(d));
1390         struct block_device *dev = sb->s_bdev;
1391 #ifdef HAVE_DEV_SET_RDONLY
1392         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1393         int rc = 0;
1394 #else
1395         int rc = -EOPNOTSUPP;
1396 #endif
1397         ENTRY;
1398
1399 #ifdef HAVE_DEV_SET_RDONLY
1400         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401
1402         if (jdev && (jdev != dev)) {
1403                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404                        (long)jdev);
1405                 dev_set_rdonly(jdev);
1406         }
1407         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1408         dev_set_rdonly(dev);
1409 #else
1410         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1411                osd_dt_dev(d)->od_svname, (long)dev, rc);
1412 #endif
1413         RETURN(rc);
1414 }
1415
1416 /*
1417  * Concurrency: serialization provided by callers.
1418  */
1419 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1420                               int mode, unsigned long timeout, __u32 alg,
1421                               struct lustre_capa_key *keys)
1422 {
1423         struct osd_device *dev = osd_dt_dev(d);
1424         ENTRY;
1425
1426         dev->od_fl_capa = mode;
1427         dev->od_capa_timeout = timeout;
1428         dev->od_capa_alg = alg;
1429         dev->od_capa_keys = keys;
1430         RETURN(0);
1431 }
1432
1433 /**
1434  * Note: we do not count into QUOTA here.
1435  * If we mount with --data_journal we may need more.
1436  */
1437 const int osd_dto_credits_noquota[DTO_NR] = {
1438         /**
1439          * Insert.
1440          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1441          * SINGLEDATA_TRANS_BLOCKS(8)
1442          * XXX Note: maybe iam need more, since iam have more level than
1443          *           EXT3 htree.
1444          */
1445         [DTO_INDEX_INSERT]  = 16,
1446         /**
1447          * Delete
1448          * just modify a single entry, probably merge few within a block
1449          */
1450         [DTO_INDEX_DELETE]  = 1,
1451         /**
1452          * Used for OI scrub
1453          */
1454         [DTO_INDEX_UPDATE]  = 16,
1455         /**
1456          * 4(inode, inode bits, groups, GDT)
1457          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1458          */
1459         [DTO_OBJECT_CREATE] = 4,
1460         /**
1461          * 4(inode, inode bits, groups, GDT)
1462          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1463          */
1464         [DTO_OBJECT_DELETE] = 4,
1465         /**
1466          * Attr set credits (inode)
1467          */
1468         [DTO_ATTR_SET_BASE] = 1,
1469         /**
1470          * Xattr set. The same as xattr of EXT3.
1471          * DATA_TRANS_BLOCKS(14)
1472          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1473          * are also counted in. Do not know why?
1474          */
1475         [DTO_XATTR_SET]     = 14,
1476         /**
1477          * credits for inode change during write.
1478          */
1479         [DTO_WRITE_BASE]    = 3,
1480         /**
1481          * credits for single block write.
1482          */
1483         [DTO_WRITE_BLOCK]   = 14,
1484         /**
1485          * Attr set credits for chown.
1486          * This is extra credits for setattr, and it is null without quota
1487          */
1488         [DTO_ATTR_SET_CHOWN] = 0
1489 };
1490
1491 static const struct dt_device_operations osd_dt_ops = {
1492         .dt_root_get       = osd_root_get,
1493         .dt_statfs         = osd_statfs,
1494         .dt_trans_create   = osd_trans_create,
1495         .dt_trans_start    = osd_trans_start,
1496         .dt_trans_stop     = osd_trans_stop,
1497         .dt_trans_cb_add   = osd_trans_cb_add,
1498         .dt_conf_get       = osd_conf_get,
1499         .dt_sync           = osd_sync,
1500         .dt_ro             = osd_ro,
1501         .dt_commit_async   = osd_commit_async,
1502         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1503 };
1504
1505 static void osd_object_read_lock(const struct lu_env *env,
1506                                  struct dt_object *dt, unsigned role)
1507 {
1508         struct osd_object *obj = osd_dt_obj(dt);
1509         struct osd_thread_info *oti = osd_oti_get(env);
1510
1511         LINVRNT(osd_invariant(obj));
1512
1513         LASSERT(obj->oo_owner != env);
1514         down_read_nested(&obj->oo_sem, role);
1515
1516         LASSERT(obj->oo_owner == NULL);
1517         oti->oti_r_locks++;
1518 }
1519
1520 static void osd_object_write_lock(const struct lu_env *env,
1521                                   struct dt_object *dt, unsigned role)
1522 {
1523         struct osd_object *obj = osd_dt_obj(dt);
1524         struct osd_thread_info *oti = osd_oti_get(env);
1525
1526         LINVRNT(osd_invariant(obj));
1527
1528         LASSERT(obj->oo_owner != env);
1529         down_write_nested(&obj->oo_sem, role);
1530
1531         LASSERT(obj->oo_owner == NULL);
1532         obj->oo_owner = env;
1533         oti->oti_w_locks++;
1534 }
1535
1536 static void osd_object_read_unlock(const struct lu_env *env,
1537                                    struct dt_object *dt)
1538 {
1539         struct osd_object *obj = osd_dt_obj(dt);
1540         struct osd_thread_info *oti = osd_oti_get(env);
1541
1542         LINVRNT(osd_invariant(obj));
1543
1544         LASSERT(oti->oti_r_locks > 0);
1545         oti->oti_r_locks--;
1546         up_read(&obj->oo_sem);
1547 }
1548
1549 static void osd_object_write_unlock(const struct lu_env *env,
1550                                     struct dt_object *dt)
1551 {
1552         struct osd_object *obj = osd_dt_obj(dt);
1553         struct osd_thread_info *oti = osd_oti_get(env);
1554
1555         LINVRNT(osd_invariant(obj));
1556
1557         LASSERT(obj->oo_owner == env);
1558         LASSERT(oti->oti_w_locks > 0);
1559         oti->oti_w_locks--;
1560         obj->oo_owner = NULL;
1561         up_write(&obj->oo_sem);
1562 }
1563
1564 static int osd_object_write_locked(const struct lu_env *env,
1565                                    struct dt_object *dt)
1566 {
1567         struct osd_object *obj = osd_dt_obj(dt);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         return obj->oo_owner == env;
1572 }
1573
1574 static int capa_is_sane(const struct lu_env *env,
1575                         struct osd_device *dev,
1576                         struct lustre_capa *capa,
1577                         struct lustre_capa_key *keys)
1578 {
1579         struct osd_thread_info *oti = osd_oti_get(env);
1580         struct lustre_capa *tcapa = &oti->oti_capa;
1581         struct obd_capa *oc;
1582         int i, rc = 0;
1583         ENTRY;
1584
1585         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586         if (oc) {
1587                 if (capa_is_expired(oc)) {
1588                         DEBUG_CAPA(D_ERROR, capa, "expired");
1589                         rc = -ESTALE;
1590                 }
1591                 capa_put(oc);
1592                 RETURN(rc);
1593         }
1594
1595         if (capa_is_expired_sec(capa)) {
1596                 DEBUG_CAPA(D_ERROR, capa, "expired");
1597                 RETURN(-ESTALE);
1598         }
1599
1600         spin_lock(&capa_lock);
1601         for (i = 0; i < 2; i++) {
1602                 if (keys[i].lk_keyid == capa->lc_keyid) {
1603                         oti->oti_capa_key = keys[i];
1604                         break;
1605                 }
1606         }
1607         spin_unlock(&capa_lock);
1608
1609         if (i == 2) {
1610                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1619                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1620                 RETURN(-EACCES);
1621         }
1622
1623         oc = capa_add(dev->od_capa_hash, capa);
1624         capa_put(oc);
1625
1626         RETURN(0);
1627 }
1628
1629 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1630                     struct lustre_capa *capa, __u64 opc)
1631 {
1632         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1633         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1634         struct lu_capainfo *lci;
1635         int rc;
1636
1637         if (!osd->od_fl_capa)
1638                 return 0;
1639
1640         if (capa == BYPASS_CAPA)
1641                 return 0;
1642
1643         lci = lu_capainfo_get(env);
1644         if (unlikely(lci == NULL))
1645                 return 0;
1646
1647         if (lci->lci_auth == LC_ID_NONE)
1648                 return 0;
1649
1650         if (capa == NULL) {
1651                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1652                        osd_name(osd), PFID(fid), -EACCES);
1653                 return -EACCES;
1654         }
1655
1656         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1657                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1658                            PFID(fid));
1659                 return -EACCES;
1660         }
1661
1662         if (!capa_opc_supported(capa, opc)) {
1663                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1664                 return -EACCES;
1665         }
1666
1667         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668         if (rc != 0) {
1669                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1670                 return -EACCES;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static struct timespec *osd_inode_time(const struct lu_env *env,
1677                                        struct inode *inode, __u64 seconds)
1678 {
1679         struct osd_thread_info  *oti = osd_oti_get(env);
1680         struct timespec         *t   = &oti->oti_time;
1681
1682         t->tv_sec = seconds;
1683         t->tv_nsec = 0;
1684         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1685         return t;
1686 }
1687
1688
1689 static void osd_inode_getattr(const struct lu_env *env,
1690                               struct inode *inode, struct lu_attr *attr)
1691 {
1692         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1693                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1694                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1695                            LA_TYPE;
1696
1697         attr->la_atime   = LTIME_S(inode->i_atime);
1698         attr->la_mtime   = LTIME_S(inode->i_mtime);
1699         attr->la_ctime   = LTIME_S(inode->i_ctime);
1700         attr->la_mode    = inode->i_mode;
1701         attr->la_size    = i_size_read(inode);
1702         attr->la_blocks  = inode->i_blocks;
1703         attr->la_uid     = i_uid_read(inode);
1704         attr->la_gid     = i_gid_read(inode);
1705         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1706         attr->la_nlink   = inode->i_nlink;
1707         attr->la_rdev    = inode->i_rdev;
1708         attr->la_blksize = 1 << inode->i_blkbits;
1709         attr->la_blkbits = inode->i_blkbits;
1710 }
1711
1712 static int osd_attr_get(const struct lu_env *env,
1713                         struct dt_object *dt,
1714                         struct lu_attr *attr,
1715                         struct lustre_capa *capa)
1716 {
1717         struct osd_object *obj = osd_dt_obj(dt);
1718
1719         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1720         LINVRNT(osd_invariant(obj));
1721
1722         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1723                 return -EACCES;
1724
1725         spin_lock(&obj->oo_guard);
1726         osd_inode_getattr(env, obj->oo_inode, attr);
1727         spin_unlock(&obj->oo_guard);
1728         return 0;
1729 }
1730
1731 static int osd_declare_attr_set(const struct lu_env *env,
1732                                 struct dt_object *dt,
1733                                 const struct lu_attr *attr,
1734                                 struct thandle *handle)
1735 {
1736         struct osd_thandle     *oh;
1737         struct osd_object      *obj;
1738         struct osd_thread_info *info = osd_oti_get(env);
1739         struct lquota_id_info  *qi = &info->oti_qi;
1740         qid_t                   uid;
1741         qid_t                   gid;
1742         long long               bspace;
1743         int                     rc = 0;
1744         bool                    allocated;
1745         ENTRY;
1746
1747         LASSERT(dt != NULL);
1748         LASSERT(handle != NULL);
1749
1750         obj = osd_dt_obj(dt);
1751         LASSERT(osd_invariant(obj));
1752
1753         oh = container_of0(handle, struct osd_thandle, ot_super);
1754         LASSERT(oh->ot_handle == NULL);
1755
1756         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1757                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1758
1759         if (attr == NULL || obj->oo_inode == NULL)
1760                 RETURN(rc);
1761
1762         bspace   = obj->oo_inode->i_blocks;
1763         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1764         bspace   = toqb(bspace);
1765
1766         /* Changing ownership is always preformed by super user, it should not
1767          * fail with EDQUOT.
1768          *
1769          * We still need to call the osd_declare_qid() to calculate the journal
1770          * credits for updating quota accounting files and to trigger quota
1771          * space adjustment once the operation is completed.*/
1772         if ((attr->la_valid & LA_UID) != 0 &&
1773              attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
1774                 qi->lqi_type = USRQUOTA;
1775
1776                 /* inode accounting */
1777                 qi->lqi_is_blk = false;
1778
1779                 /* one more inode for the new owner ... */
1780                 qi->lqi_id.qid_uid = attr->la_uid;
1781                 qi->lqi_space      = 1;
1782                 allocated = (attr->la_uid == 0) ? true : false;
1783                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1784                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1785                         rc = 0;
1786                 if (rc)
1787                         RETURN(rc);
1788
1789                 /* and one less inode for the current uid */
1790                 qi->lqi_id.qid_uid = uid;
1791                 qi->lqi_space      = -1;
1792                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1793                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1794                         rc = 0;
1795                 if (rc)
1796                         RETURN(rc);
1797
1798                 /* block accounting */
1799                 qi->lqi_is_blk = true;
1800
1801                 /* more blocks for the new owner ... */
1802                 qi->lqi_id.qid_uid = attr->la_uid;
1803                 qi->lqi_space      = bspace;
1804                 allocated = (attr->la_uid == 0) ? true : false;
1805                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1806                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1807                         rc = 0;
1808                 if (rc)
1809                         RETURN(rc);
1810
1811                 /* and finally less blocks for the current owner */
1812                 qi->lqi_id.qid_uid = uid;
1813                 qi->lqi_space      = -bspace;
1814                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1815                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1816                         rc = 0;
1817                 if (rc)
1818                         RETURN(rc);
1819         }
1820
1821         if (attr->la_valid & LA_GID &&
1822             attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
1823                 qi->lqi_type = GRPQUOTA;
1824
1825                 /* inode accounting */
1826                 qi->lqi_is_blk = false;
1827
1828                 /* one more inode for the new group owner ... */
1829                 qi->lqi_id.qid_gid = attr->la_gid;
1830                 qi->lqi_space      = 1;
1831                 allocated = (attr->la_gid == 0) ? true : false;
1832                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1833                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1834                         rc = 0;
1835                 if (rc)
1836                         RETURN(rc);
1837
1838                 /* and one less inode for the current gid */
1839                 qi->lqi_id.qid_gid = gid;
1840                 qi->lqi_space      = -1;
1841                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1842                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1843                         rc = 0;
1844                 if (rc)
1845                         RETURN(rc);
1846
1847                 /* block accounting */
1848                 qi->lqi_is_blk = true;
1849
1850                 /* more blocks for the new owner ... */
1851                 qi->lqi_id.qid_gid = attr->la_gid;
1852                 qi->lqi_space      = bspace;
1853                 allocated = (attr->la_gid == 0) ? true : false;
1854                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1855                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1856                         rc = 0;
1857                 if (rc)
1858                         RETURN(rc);
1859
1860                 /* and finally less blocks for the current owner */
1861                 qi->lqi_id.qid_gid = gid;
1862                 qi->lqi_space      = -bspace;
1863                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1864                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1865                         rc = 0;
1866                 if (rc)
1867                         RETURN(rc);
1868         }
1869
1870         RETURN(rc);
1871 }
1872
1873 static int osd_inode_setattr(const struct lu_env *env,
1874                              struct inode *inode, const struct lu_attr *attr)
1875 {
1876         __u64 bits = attr->la_valid;
1877
1878         /* Only allow set size for regular file */
1879         if (!S_ISREG(inode->i_mode))
1880                 bits &= ~(LA_SIZE | LA_BLOCKS);
1881
1882         if (bits == 0)
1883                 return 0;
1884
1885         if (bits & LA_ATIME)
1886                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1887         if (bits & LA_CTIME)
1888                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1889         if (bits & LA_MTIME)
1890                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1891         if (bits & LA_SIZE) {
1892                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1893                 i_size_write(inode, attr->la_size);
1894         }
1895
1896 #if 0
1897         /* OSD should not change "i_blocks" which is used by quota.
1898          * "i_blocks" should be changed by ldiskfs only. */
1899         if (bits & LA_BLOCKS)
1900                 inode->i_blocks = attr->la_blocks;
1901 #endif
1902         if (bits & LA_MODE)
1903                 inode->i_mode = (inode->i_mode & S_IFMT) |
1904                                 (attr->la_mode & ~S_IFMT);
1905         if (bits & LA_UID)
1906                 i_uid_write(inode, attr->la_uid);
1907         if (bits & LA_GID)
1908                 i_gid_write(inode, attr->la_gid);
1909         if (bits & LA_NLINK)
1910                 set_nlink(inode, attr->la_nlink);
1911         if (bits & LA_RDEV)
1912                 inode->i_rdev = attr->la_rdev;
1913
1914         if (bits & LA_FLAGS) {
1915                 /* always keep S_NOCMTIME */
1916                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1917                                  S_NOCMTIME;
1918         }
1919         return 0;
1920 }
1921
1922 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1923 {
1924         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1925             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1926                 struct iattr    iattr;
1927                 int             rc;
1928
1929                 iattr.ia_valid = 0;
1930                 if (attr->la_valid & LA_UID)
1931                         iattr.ia_valid |= ATTR_UID;
1932                 if (attr->la_valid & LA_GID)
1933                         iattr.ia_valid |= ATTR_GID;
1934                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1935                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1936
1937                 rc = ll_vfs_dq_transfer(inode, &iattr);
1938                 if (rc) {
1939                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1940                                "enforcement enabled on the ldiskfs filesystem?",
1941                                inode->i_sb->s_id, rc);
1942                         return rc;
1943                 }
1944         }
1945         return 0;
1946 }
1947
1948 static int osd_attr_set(const struct lu_env *env,
1949                         struct dt_object *dt,
1950                         const struct lu_attr *attr,
1951                         struct thandle *handle,
1952                         struct lustre_capa *capa)
1953 {
1954         struct osd_object *obj = osd_dt_obj(dt);
1955         struct inode      *inode;
1956         int rc;
1957
1958         LASSERT(handle != NULL);
1959         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1960         LASSERT(osd_invariant(obj));
1961
1962         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1963                 return -EACCES;
1964
1965         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1966
1967         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1968                 struct osd_thread_info  *oti  = osd_oti_get(env);
1969                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1970                 struct lu_fid           *fid1 = &oti->oti_fid;
1971                 struct osd_inode_id     *id   = &oti->oti_id;
1972                 struct iam_path_descr   *ipd;
1973                 struct iam_container    *bag;
1974                 struct osd_thandle      *oh;
1975                 int                      rc;
1976
1977                 fid_cpu_to_be(fid1, fid0);
1978                 memset(id, 1, sizeof(*id));
1979                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1980                                   fid0)->oi_dir.od_container;
1981                 ipd = osd_idx_ipd_get(env, bag);
1982                 if (unlikely(ipd == NULL))
1983                         RETURN(-ENOMEM);
1984
1985                 oh = container_of0(handle, struct osd_thandle, ot_super);
1986                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1987                                 (const struct iam_rec *)id, ipd);
1988                 osd_ipd_put(env, bag, ipd);
1989                 return(rc > 0 ? 0 : rc);
1990         }
1991
1992         inode = obj->oo_inode;
1993         ll_vfs_dq_init(inode);
1994
1995         rc = osd_quota_transfer(inode, attr);
1996         if (rc)
1997                 return rc;
1998
1999         spin_lock(&obj->oo_guard);
2000         rc = osd_inode_setattr(env, inode, attr);
2001         spin_unlock(&obj->oo_guard);
2002
2003         if (!rc)
2004                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2005         return rc;
2006 }
2007
2008 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2009                                     struct osd_object *obj,
2010                                     const char *name, const int namelen)
2011 {
2012         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2013 }
2014
2015 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2016                       umode_t mode, struct dt_allocation_hint *hint,
2017                       struct thandle *th)
2018 {
2019         int result;
2020         struct osd_device  *osd = osd_obj2dev(obj);
2021         struct osd_thandle *oth;
2022         struct dt_object   *parent = NULL;
2023         struct inode       *inode;
2024
2025         LINVRNT(osd_invariant(obj));
2026         LASSERT(obj->oo_inode == NULL);
2027         LASSERT(obj->oo_hl_head == NULL);
2028
2029         if (S_ISDIR(mode) && ldiskfs_pdo) {
2030                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2031                 if (obj->oo_hl_head == NULL)
2032                         return -ENOMEM;
2033         }
2034
2035         oth = container_of(th, struct osd_thandle, ot_super);
2036         LASSERT(oth->ot_handle->h_transaction != NULL);
2037
2038         if (hint && hint->dah_parent)
2039                 parent = hint->dah_parent;
2040
2041         inode = ldiskfs_create_inode(oth->ot_handle,
2042                                      parent ? osd_dt_obj(parent)->oo_inode :
2043                                               osd_sb(osd)->s_root->d_inode,
2044                                      mode);
2045         if (!IS_ERR(inode)) {
2046                 /* Do not update file c/mtime in ldiskfs.
2047                  * NB: don't need any lock because no contention at this
2048                  * early stage */
2049                 inode->i_flags |= S_NOCMTIME;
2050
2051                 /* For new created object, it must be consistent,
2052                  * and it is unnecessary to scrub against it. */
2053                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2054                 obj->oo_inode = inode;
2055                 result = 0;
2056         } else {
2057                 if (obj->oo_hl_head != NULL) {
2058                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2059                         obj->oo_hl_head = NULL;
2060                 }
2061                 result = PTR_ERR(inode);
2062         }
2063         LINVRNT(osd_invariant(obj));
2064         return result;
2065 }
2066
2067 enum {
2068         OSD_NAME_LEN = 255
2069 };
2070
2071 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2072                      struct lu_attr *attr,
2073                      struct dt_allocation_hint *hint,
2074                      struct dt_object_format *dof,
2075                      struct thandle *th)
2076 {
2077         int result;
2078         struct osd_thandle *oth;
2079         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2080
2081         LASSERT(S_ISDIR(attr->la_mode));
2082
2083         oth = container_of(th, struct osd_thandle, ot_super);
2084         LASSERT(oth->ot_handle->h_transaction != NULL);
2085         result = osd_mkfile(info, obj, mode, hint, th);
2086
2087         return result;
2088 }
2089
2090 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2091                         struct lu_attr *attr,
2092                         struct dt_allocation_hint *hint,
2093                         struct dt_object_format *dof,
2094                         struct thandle *th)
2095 {
2096         int result;
2097         struct osd_thandle *oth;
2098         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2099
2100         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2101
2102         LASSERT(S_ISREG(attr->la_mode));
2103
2104         oth = container_of(th, struct osd_thandle, ot_super);
2105         LASSERT(oth->ot_handle->h_transaction != NULL);
2106
2107         result = osd_mkfile(info, obj, mode, hint, th);
2108         if (result == 0) {
2109                 LASSERT(obj->oo_inode != NULL);
2110                 if (feat->dif_flags & DT_IND_VARKEY)
2111                         result = iam_lvar_create(obj->oo_inode,
2112                                                  feat->dif_keysize_max,
2113                                                  feat->dif_ptrsize,
2114                                                  feat->dif_recsize_max,
2115                                                  oth->ot_handle);
2116                 else
2117                         result = iam_lfix_create(obj->oo_inode,
2118                                                  feat->dif_keysize_max,
2119                                                  feat->dif_ptrsize,
2120                                                  feat->dif_recsize_max,
2121                                                  oth->ot_handle);
2122
2123         }
2124         return result;
2125 }
2126
2127 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2128                      struct lu_attr *attr,
2129                      struct dt_allocation_hint *hint,
2130                      struct dt_object_format *dof,
2131                      struct thandle *th)
2132 {
2133         LASSERT(S_ISREG(attr->la_mode));
2134         return osd_mkfile(info, obj, (attr->la_mode &
2135                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2136 }
2137
2138 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2139                      struct lu_attr *attr,
2140                      struct dt_allocation_hint *hint,
2141                      struct dt_object_format *dof,
2142                      struct thandle *th)
2143 {
2144         LASSERT(S_ISLNK(attr->la_mode));
2145         return osd_mkfile(info, obj, (attr->la_mode &
2146                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2147 }
2148
2149 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2150                      struct lu_attr *attr,
2151                      struct dt_allocation_hint *hint,
2152                      struct dt_object_format *dof,
2153                      struct thandle *th)
2154 {
2155         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2156         int result;
2157
2158         LINVRNT(osd_invariant(obj));
2159         LASSERT(obj->oo_inode == NULL);
2160         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2161                 S_ISFIFO(mode) || S_ISSOCK(mode));
2162
2163         result = osd_mkfile(info, obj, mode, hint, th);
2164         if (result == 0) {
2165                 LASSERT(obj->oo_inode != NULL);
2166                 /*
2167                  * This inode should be marked dirty for i_rdev.  Currently
2168                  * that is done in the osd_attr_init().
2169                  */
2170                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2171                                    attr->la_rdev);
2172         }
2173         LINVRNT(osd_invariant(obj));
2174         return result;
2175 }
2176
2177 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2178                               struct lu_attr *,
2179                               struct dt_allocation_hint *hint,
2180                               struct dt_object_format *dof,
2181                               struct thandle *);
2182
2183 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2184 {
2185         osd_obj_type_f result;
2186
2187         switch (type) {
2188         case DFT_DIR:
2189                 result = osd_mkdir;
2190                 break;
2191         case DFT_REGULAR:
2192                 result = osd_mkreg;
2193                 break;
2194         case DFT_SYM:
2195                 result = osd_mksym;
2196                 break;
2197         case DFT_NODE:
2198                 result = osd_mknod;
2199                 break;
2200         case DFT_INDEX:
2201                 result = osd_mk_index;
2202                 break;
2203
2204         default:
2205                 LBUG();
2206                 break;
2207         }
2208         return result;
2209 }
2210
2211
2212 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2213                         struct dt_object *parent, struct dt_object *child,
2214                         umode_t child_mode)
2215 {
2216         LASSERT(ah);
2217
2218         ah->dah_parent = parent;
2219         ah->dah_mode = child_mode;
2220 }
2221
2222 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2223                           struct lu_attr *attr, struct dt_object_format *dof)
2224 {
2225         struct inode   *inode = obj->oo_inode;
2226         __u64           valid = attr->la_valid;
2227         int             result;
2228
2229         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2230
2231         if (dof->dof_type != DFT_NODE)
2232                 attr->la_valid &= ~LA_RDEV;
2233         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2234                 attr->la_valid &= ~LA_ATIME;
2235         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2236                 attr->la_valid &= ~LA_CTIME;
2237         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2238                 attr->la_valid &= ~LA_MTIME;
2239
2240         result = osd_quota_transfer(inode, attr);
2241         if (result)
2242                 return;
2243
2244         if (attr->la_valid != 0) {
2245                 result = osd_inode_setattr(info->oti_env, inode, attr);
2246                 /*
2247                  * The osd_inode_setattr() should always succeed here.  The
2248                  * only error that could be returned is EDQUOT when we are
2249                  * trying to change the UID or GID of the inode. However, this
2250                  * should not happen since quota enforcement is no longer
2251                  * enabled on ldiskfs (lquota takes care of it).
2252                  */
2253                 LASSERTF(result == 0, "%d", result);
2254                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2255         }
2256
2257         attr->la_valid = valid;
2258 }
2259
2260 /**
2261  * Helper function for osd_object_create()
2262  *
2263  * \retval 0, on success
2264  */
2265 static int __osd_object_create(struct osd_thread_info *info,
2266                                struct osd_object *obj, struct lu_attr *attr,
2267                                struct dt_allocation_hint *hint,
2268                                struct dt_object_format *dof,
2269                                struct thandle *th)
2270 {
2271         int     result;
2272         __u32   umask;
2273
2274         /* we drop umask so that permissions we pass are not affected */
2275         umask = current->fs->umask;
2276         current->fs->umask = 0;
2277
2278         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2279                                                   th);
2280         if (result == 0) {
2281                 osd_attr_init(info, obj, attr, dof);
2282                 osd_object_init0(obj);
2283                 /* bz 24037 */
2284                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2285                         unlock_new_inode(obj->oo_inode);
2286         }
2287
2288         /* restore previous umask value */
2289         current->fs->umask = umask;
2290
2291         return result;
2292 }
2293
2294 /**
2295  * Helper function for osd_object_create()
2296  *
2297  * \retval 0, on success
2298  */
2299 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2300                            const struct lu_fid *fid, struct thandle *th)
2301 {
2302         struct osd_thread_info *info = osd_oti_get(env);
2303         struct osd_inode_id    *id   = &info->oti_id;
2304         struct osd_device      *osd  = osd_obj2dev(obj);
2305         struct osd_thandle     *oh;
2306
2307         LASSERT(obj->oo_inode != NULL);
2308
2309         oh = container_of0(th, struct osd_thandle, ot_super);
2310         LASSERT(oh->ot_handle);
2311
2312         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2313         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2314 }
2315
2316 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2317                    obd_seq seq, struct lu_seq_range *range)
2318 {
2319         struct seq_server_site  *ss = osd_seq_site(osd);
2320
2321         if (fid_seq_is_idif(seq)) {
2322                 fld_range_set_ost(range);
2323                 range->lsr_index = idif_ost_idx(seq);
2324                 return 0;
2325         }
2326
2327         if (!fid_seq_in_fldb(seq)) {
2328                 fld_range_set_mdt(range);
2329                 if (ss != NULL)
2330                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2331                          * at all */
2332                         range->lsr_index = ss->ss_node_id;
2333                 return 0;
2334         }
2335
2336         LASSERT(ss != NULL);
2337         fld_range_set_any(range);
2338         /* OSD will only do local fld lookup */
2339         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2340 }
2341
2342 /*
2343  * Concurrency: no external locking is necessary.
2344  */
2345 static int osd_declare_object_create(const struct lu_env *env,
2346                                      struct dt_object *dt,
2347                                      struct lu_attr *attr,
2348                                      struct dt_allocation_hint *hint,
2349                                      struct dt_object_format *dof,
2350                                      struct thandle *handle)
2351 {
2352         struct osd_thandle      *oh;
2353         int                      rc;
2354         ENTRY;
2355
2356         LASSERT(handle != NULL);
2357
2358         oh = container_of0(handle, struct osd_thandle, ot_super);
2359         LASSERT(oh->ot_handle == NULL);
2360
2361         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2362                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2363         /* Reuse idle OI block may cause additional one OI block
2364          * to be changed. */
2365         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2366                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2367
2368         /* If this is directory, then we expect . and .. to be inserted as
2369          * well. The one directory block always needs to be created for the
2370          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2371          * block), there is no danger of needing a tree for the first block.
2372          */
2373         if (attr && S_ISDIR(attr->la_mode)) {
2374                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2375                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2376                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2377         }
2378
2379         if (!attr)
2380                 RETURN(0);
2381
2382         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2383                                    false, false, NULL, false);
2384         if (rc != 0)
2385                 RETURN(rc);
2386
2387         RETURN(rc);
2388 }
2389
2390 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2391                              struct lu_attr *attr,
2392                              struct dt_allocation_hint *hint,
2393                              struct dt_object_format *dof,
2394                              struct thandle *th)
2395 {
2396         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2397         struct osd_object      *obj    = osd_dt_obj(dt);
2398         struct osd_thread_info *info   = osd_oti_get(env);
2399         int result;
2400
2401         ENTRY;
2402
2403         LINVRNT(osd_invariant(obj));
2404         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2405         LASSERT(osd_write_locked(env, obj));
2406         LASSERT(th != NULL);
2407
2408         if (unlikely(fid_is_acct(fid)))
2409                 /* Quota files can't be created from the kernel any more,
2410                  * 'tune2fs -O quota' will take care of creating them */
2411                 RETURN(-EPERM);
2412
2413         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2414         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2415
2416         result = __osd_object_create(info, obj, attr, hint, dof, th);
2417         if (result == 0)
2418                 result = __osd_oi_insert(env, obj, fid, th);
2419
2420         LASSERT(ergo(result == 0,
2421                      dt_object_exists(dt) && !dt_object_remote(dt)));
2422
2423         LASSERT(osd_invariant(obj));
2424         RETURN(result);
2425 }
2426
2427 /**
2428  * Called to destroy on-disk representation of the object
2429  *
2430  * Concurrency: must be locked
2431  */
2432 static int osd_declare_object_destroy(const struct lu_env *env,
2433                                       struct dt_object *dt,
2434                                       struct thandle *th)
2435 {
2436         struct osd_object  *obj = osd_dt_obj(dt);
2437         struct inode       *inode = obj->oo_inode;
2438         struct osd_thandle *oh;
2439         int                 rc;
2440         ENTRY;
2441
2442         oh = container_of0(th, struct osd_thandle, ot_super);
2443         LASSERT(oh->ot_handle == NULL);
2444         LASSERT(inode);
2445
2446         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2447                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2448         /* Recycle idle OI leaf may cause additional three OI blocks
2449          * to be changed. */
2450         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2451                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2452         /* one less inode */
2453         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2454                                    -1, oh, false, true, NULL, false);
2455         if (rc)
2456                 RETURN(rc);
2457         /* data to be truncated */
2458         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2459                                    0, oh, true, true, NULL, false);
2460         RETURN(rc);
2461 }
2462
2463 static int osd_object_destroy(const struct lu_env *env,
2464                               struct dt_object *dt,
2465                               struct thandle *th)
2466 {
2467         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2468         struct osd_object      *obj = osd_dt_obj(dt);
2469         struct inode           *inode = obj->oo_inode;
2470         struct osd_device      *osd = osd_obj2dev(obj);
2471         struct osd_thandle     *oh;
2472         int                     result;
2473         ENTRY;
2474
2475         oh = container_of0(th, struct osd_thandle, ot_super);
2476         LASSERT(oh->ot_handle);
2477         LASSERT(inode);
2478         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2479
2480         if (unlikely(fid_is_acct(fid)))
2481                 RETURN(-EPERM);
2482
2483         if (S_ISDIR(inode->i_mode)) {
2484                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2485                         inode->i_nlink == 2);
2486                 /* it will check/delete the inode from remote parent,
2487                  * how to optimize it? unlink performance impaction XXX */
2488                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2489                 if (result != 0 && result != -ENOENT) {
2490                         CERROR("%s: delete inode "DFID": rc = %d\n",
2491                                osd_name(osd), PFID(fid), result);
2492                 }
2493                 spin_lock(&obj->oo_guard);
2494                 clear_nlink(inode);
2495                 spin_unlock(&obj->oo_guard);
2496                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2497         }
2498
2499         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2500
2501         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2502                                OI_CHECK_FLD);
2503
2504         /* XXX: add to ext3 orphan list */
2505         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2506
2507         /* not needed in the cache anymore */
2508         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2509
2510         RETURN(0);
2511 }
2512
2513 /**
2514  * Put the fid into lustre_mdt_attrs, and then place the structure
2515  * inode's ea. This fid should not be altered during the life time
2516  * of the inode.
2517  *
2518  * \retval +ve, on success
2519  * \retval -ve, on error
2520  *
2521  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2522  */
2523 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2524                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2525 {
2526         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2527         int                      rc;
2528         ENTRY;
2529
2530         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2531                 RETURN(0);
2532
2533         lustre_lma_init(lma, fid, compat, incompat);
2534         lustre_lma_swab(lma);
2535
2536         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2537                              XATTR_CREATE);
2538         /* LMA may already exist, but we need to check that all the
2539          * desired compat/incompat flags have been added. */
2540         if (unlikely(rc == -EEXIST)) {
2541                 if (compat == 0 && incompat == 0)
2542                         RETURN(0);
2543
2544                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2545                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2546                                      LMA_OLD_SIZE);
2547                 if (rc <= 0)
2548                         RETURN(-EINVAL);
2549
2550                 lustre_lma_swab(lma);
2551                 if (!(~lma->lma_compat & compat) &&
2552                     !(~lma->lma_incompat & incompat))
2553                         RETURN(0);
2554
2555                 lma->lma_compat |= compat;
2556                 lma->lma_incompat |= incompat;
2557                 lustre_lma_swab(lma);
2558                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2559                                      sizeof(*lma), XATTR_REPLACE);
2560         }
2561
2562         RETURN(rc);
2563 }
2564
2565 /**
2566  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2567  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2568  * To have compatilibility with 1.8 ldiskfs driver we need to have
2569  * magic number at start of fid data.
2570  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2571  * its inmemory API.
2572  */
2573 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2574                                   const struct dt_rec *fid)
2575 {
2576         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2577             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2578                 param->edp_magic = 0;
2579                 return;
2580         }
2581
2582         param->edp_magic = LDISKFS_LUFID_MAGIC;
2583         param->edp_len =  sizeof(struct lu_fid) + 1;
2584         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2585 }
2586
2587 /**
2588  * Try to read the fid from inode ea into dt_rec.
2589  *
2590  * \param fid object fid.
2591  *
2592  * \retval 0 on success
2593  */
2594 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2595                           __u32 ino, struct lu_fid *fid,
2596                           struct osd_inode_id *id)
2597 {
2598         struct osd_thread_info *info  = osd_oti_get(env);
2599         struct inode           *inode;
2600         ENTRY;
2601
2602         osd_id_gen(id, ino, OSD_OII_NOGEN);
2603         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2604         if (IS_ERR(inode))
2605                 RETURN(PTR_ERR(inode));
2606
2607         iput(inode);
2608         RETURN(0);
2609 }
2610
2611 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2612                                         struct inode *dir,
2613                                         struct inode  *parent_dir,
2614                                         const struct dt_rec *dot_fid,
2615                                         const struct dt_rec *dot_dot_fid,
2616                                         struct osd_thandle *oth)
2617 {
2618         struct ldiskfs_dentry_param *dot_ldp;
2619         struct ldiskfs_dentry_param *dot_dot_ldp;
2620
2621         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2622         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2623
2624         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2625         dot_ldp->edp_magic = 0;
2626         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2627                                         dir, dot_ldp, dot_dot_ldp);
2628 }
2629
2630 /**
2631  * Create an local agent inode for remote entry
2632  */
2633 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2634                                                   struct osd_device *osd,
2635                                                   struct osd_object *pobj,
2636                                                   const struct lu_fid *fid,
2637                                                   struct thandle *th)
2638 {
2639         struct osd_thread_info  *info = osd_oti_get(env);
2640         struct inode            *local;
2641         struct osd_thandle      *oh;
2642         int                     rc;
2643         ENTRY;
2644
2645         LASSERT(th);
2646         oh = container_of(th, struct osd_thandle, ot_super);
2647         LASSERT(oh->ot_handle->h_transaction != NULL);
2648
2649         /* FIXME: Insert index api needs to know the mode of
2650          * the remote object. Just use S_IFDIR for now */
2651         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2652         if (IS_ERR(local)) {
2653                 CERROR("%s: create local error %d\n", osd_name(osd),
2654                        (int)PTR_ERR(local));
2655                 RETURN(local);
2656         }
2657
2658         /* Set special LMA flag for local agent inode */
2659         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2660         if (rc != 0) {
2661                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2662                        osd_name(osd), PFID(fid), rc);
2663                 RETURN(ERR_PTR(rc));
2664         }
2665
2666         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2667                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2668                 (const struct dt_rec *)fid, oh);
2669         if (rc != 0) {
2670                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2671                         osd_name(osd), PFID(fid), rc);
2672                 RETURN(ERR_PTR(rc));
2673         }
2674
2675         RETURN(local);
2676 }
2677
2678 /**
2679  * Delete local agent inode for remote entry
2680  */
2681 static int osd_delete_local_agent_inode(const struct lu_env *env,
2682                                         struct osd_device *osd,
2683                                         const struct lu_fid *fid,
2684                                         __u32 ino, struct osd_thandle *oh)
2685 {
2686         struct osd_thread_info  *oti = osd_oti_get(env);
2687         struct osd_inode_id     *id = &oti->oti_id;
2688         struct inode            *inode;
2689         ENTRY;
2690
2691         id->oii_ino = le32_to_cpu(ino);
2692         id->oii_gen = OSD_OII_NOGEN;
2693         inode = osd_iget(oti, osd, id);
2694         if (IS_ERR(inode)) {
2695                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2696                        PFID(fid), id->oii_ino, id->oii_gen);
2697                 RETURN(PTR_ERR(inode));
2698         }
2699
2700         clear_nlink(inode);
2701         mark_inode_dirty(inode);
2702         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2703                 osd_name(osd), PFID(fid), inode->i_ino);
2704         iput(inode);
2705         RETURN(0);
2706 }
2707
2708 /**
2709  * OSD layer object create function for interoperability mode (b11826).
2710  * This is mostly similar to osd_object_create(). Only difference being, fid is
2711  * inserted into inode ea here.
2712  *
2713  * \retval   0, on success
2714  * \retval -ve, on error
2715  */
2716 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2717                                 struct lu_attr *attr,
2718                                 struct dt_allocation_hint *hint,
2719                                 struct dt_object_format *dof,
2720                                 struct thandle *th)
2721 {
2722         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2723         struct osd_object      *obj    = osd_dt_obj(dt);
2724         struct osd_thread_info *info   = osd_oti_get(env);
2725         int                     result;
2726
2727         ENTRY;
2728
2729         LASSERT(osd_invariant(obj));
2730         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2731         LASSERT(osd_write_locked(env, obj));
2732         LASSERT(th != NULL);
2733
2734         if (unlikely(fid_is_acct(fid)))
2735                 /* Quota files can't be created from the kernel any more,
2736                  * 'tune2fs -O quota' will take care of creating them */
2737                 RETURN(-EPERM);
2738
2739         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2740         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2741
2742         result = __osd_object_create(info, obj, attr, hint, dof, th);
2743         if (result == 0)
2744                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2745                                 fid_is_on_ost(info, osd_obj2dev(obj),
2746                                               fid, OI_CHECK_FLD) ?
2747                                 LMAC_FID_ON_OST : 0, 0);
2748
2749         if (result == 0)
2750                 result = __osd_oi_insert(env, obj, fid, th);
2751
2752         LASSERT(ergo(result == 0,
2753                      dt_object_exists(dt) && !dt_object_remote(dt)));
2754         LINVRNT(osd_invariant(obj));
2755         RETURN(result);
2756 }
2757
2758 static int osd_declare_object_ref_add(const struct lu_env *env,
2759                                       struct dt_object *dt,
2760                                       struct thandle *handle)
2761 {
2762         struct osd_thandle       *oh;
2763
2764         /* it's possible that object doesn't exist yet */
2765         LASSERT(handle != NULL);
2766
2767         oh = container_of0(handle, struct osd_thandle, ot_super);
2768         LASSERT(oh->ot_handle == NULL);
2769
2770         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2771                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2772
2773         return 0;
2774 }
2775
2776 /*
2777  * Concurrency: @dt is write locked.
2778  */
2779 static int osd_object_ref_add(const struct lu_env *env,
2780                               struct dt_object *dt, struct thandle *th)
2781 {
2782         struct osd_object  *obj = osd_dt_obj(dt);
2783         struct inode       *inode = obj->oo_inode;
2784         struct osd_thandle *oh;
2785         int                 rc = 0;
2786
2787         LINVRNT(osd_invariant(obj));
2788         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2789         LASSERT(osd_write_locked(env, obj));
2790         LASSERT(th != NULL);
2791
2792         oh = container_of0(th, struct osd_thandle, ot_super);
2793         LASSERT(oh->ot_handle != NULL);
2794
2795         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2796
2797         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2798                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2799         /*
2800          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2801          * (65000) subdirectories by storing "1" in i_nlink if the link count
2802          * would otherwise overflow. Directory tranversal tools understand
2803          * that (st_nlink == 1) indicates that the filesystem dose not track
2804          * hard links count on the directory, and will not abort subdirectory
2805          * scanning early once (st_nlink - 2) subdirs have been found.
2806          *
2807          * This also has to properly handle the case of inodes with nlink == 0
2808          * in case they are being linked into the PENDING directory
2809          */
2810 #ifdef I_LINKABLE
2811         /* This is necessary to increment from i_nlink == 0 */
2812         spin_lock(&inode->i_lock);
2813         inode->i_state |= I_LINKABLE;
2814         spin_unlock(&inode->i_lock);
2815 #endif
2816
2817         spin_lock(&obj->oo_guard);
2818         ldiskfs_inc_count(oh->ot_handle, inode);
2819         if (!S_ISDIR(inode->i_mode))
2820                 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2821         spin_unlock(&obj->oo_guard);
2822
2823         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2824         LINVRNT(osd_invariant(obj));
2825
2826         return rc;
2827 }
2828
2829 static int osd_declare_object_ref_del(const struct lu_env *env,
2830                                       struct dt_object *dt,
2831                                       struct thandle *handle)
2832 {
2833         struct osd_thandle *oh;
2834
2835         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2836         LASSERT(handle != NULL);
2837
2838         oh = container_of0(handle, struct osd_thandle, ot_super);
2839         LASSERT(oh->ot_handle == NULL);
2840
2841         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2842                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2843
2844         return 0;
2845 }
2846
2847 /*
2848  * Concurrency: @dt is write locked.
2849  */
2850 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2851                               struct thandle *th)
2852 {
2853         struct osd_object       *obj = osd_dt_obj(dt);
2854         struct inode            *inode = obj->oo_inode;
2855         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2856         struct osd_thandle      *oh;
2857
2858         LINVRNT(osd_invariant(obj));
2859         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2860         LASSERT(osd_write_locked(env, obj));
2861         LASSERT(th != NULL);
2862
2863         oh = container_of0(th, struct osd_thandle, ot_super);
2864         LASSERT(oh->ot_handle != NULL);
2865
2866         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2867
2868         spin_lock(&obj->oo_guard);
2869         /* That can be result of upgrade from old Lustre version and
2870          * applied only to local files.  Just skip this ref_del call.
2871          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2872         if (inode->i_nlink == 0) {
2873                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2874                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2875                              ", maybe an upgraded file? (LU-3915)\n",
2876                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2877                 spin_unlock(&obj->oo_guard);
2878                 return 0;
2879         }
2880
2881         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2882                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2883
2884         ldiskfs_dec_count(oh->ot_handle, inode);
2885         spin_unlock(&obj->oo_guard);
2886
2887         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2888         LINVRNT(osd_invariant(obj));
2889
2890         return 0;
2891 }
2892
2893 /*
2894  * Get the 64-bit version for an inode.
2895  */
2896 static int osd_object_version_get(const struct lu_env *env,
2897                                   struct dt_object *dt, dt_obj_version_t *ver)
2898 {
2899         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2900
2901         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2902                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2903         *ver = LDISKFS_I(inode)->i_fs_version;
2904         return 0;
2905 }
2906
2907 /*
2908  * Concurrency: @dt is read locked.
2909  */
2910 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2911                          struct lu_buf *buf, const char *name,
2912                          struct lustre_capa *capa)
2913 {
2914         struct osd_object      *obj    = osd_dt_obj(dt);
2915         struct inode           *inode  = obj->oo_inode;
2916         struct osd_thread_info *info   = osd_oti_get(env);
2917         struct dentry          *dentry = &info->oti_obj_dentry;
2918
2919         /* version get is not real XATTR but uses xattr API */
2920         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2921                 /* for version we are just using xattr API but change inode
2922                  * field instead */
2923                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2924                 osd_object_version_get(env, dt, buf->lb_buf);
2925                 return sizeof(dt_obj_version_t);
2926         }
2927
2928         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2929         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2930
2931         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2932                 return -EACCES;
2933
2934         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2935 }
2936
2937
2938 static int osd_declare_xattr_set(const struct lu_env *env,
2939                                  struct dt_object *dt,
2940                                  const struct lu_buf *buf, const char *name,
2941                                  int fl, struct thandle *handle)
2942 {
2943         struct osd_thandle *oh;
2944         int credits;
2945
2946         LASSERT(handle != NULL);
2947
2948         oh = container_of0(handle, struct osd_thandle, ot_super);
2949         LASSERT(oh->ot_handle == NULL);
2950
2951         /* optimistic optimization: LMA is set first and usually fit inode */
2952         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2953                 if (dt_object_exists(dt))
2954                         credits = 0;
2955                 else
2956                         credits = 1;
2957         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2958                 credits = 1;
2959         } else {
2960                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2961                 struct super_block *sb = osd_sb(osd);
2962                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2963                 if (buf && buf->lb_len > sb->s_blocksize) {
2964                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2965                                         sb->s_blocksize_bits;
2966                 }
2967         }
2968
2969         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2970
2971         return 0;
2972 }
2973
2974 /*
2975  * Set the 64-bit version for object
2976  */
2977 static void osd_object_version_set(const struct lu_env *env,
2978                                    struct dt_object *dt,
2979                                    dt_obj_version_t *new_version)
2980 {
2981         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2982
2983         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2984                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2985
2986         LDISKFS_I(inode)->i_fs_version = *new_version;
2987         /** Version is set after all inode operations are finished,
2988          *  so we should mark it dirty here */
2989         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2990 }
2991
2992 /*
2993  * Concurrency: @dt is write locked.
2994  */
2995 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2996                          const struct lu_buf *buf, const char *name, int fl,
2997                          struct thandle *handle, struct lustre_capa *capa)
2998 {
2999         struct osd_object      *obj      = osd_dt_obj(dt);
3000         struct inode           *inode    = obj->oo_inode;
3001         struct osd_thread_info *info     = osd_oti_get(env);
3002         int                     fs_flags = 0;
3003         ENTRY;
3004
3005         LASSERT(handle != NULL);
3006
3007         /* version set is not real XATTR */
3008         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3009                 /* for version we are just using xattr API but change inode
3010                  * field instead */
3011                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3012                 osd_object_version_set(env, dt, buf->lb_buf);
3013                 return sizeof(dt_obj_version_t);
3014         }
3015
3016         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3017                 return -EACCES;
3018
3019         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3020                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3021
3022         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3023         if (fl & LU_XATTR_REPLACE)
3024                 fs_flags |= XATTR_REPLACE;
3025
3026         if (fl & LU_XATTR_CREATE)
3027                 fs_flags |= XATTR_CREATE;
3028
3029         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3030                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3031                 int                      rc;
3032
3033                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3034                 if (rc != 0)
3035                         RETURN(rc);
3036
3037                 lma->lma_incompat |= LMAI_STRIPED;
3038                 lustre_lma_swab(lma);
3039                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3040                                      sizeof(*lma), XATTR_REPLACE);
3041                 if (rc != 0)
3042                         RETURN(rc);
3043         }
3044
3045         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3046                                fs_flags);
3047 }
3048
3049 /*
3050  * Concurrency: @dt is read locked.
3051  */
3052 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3053                           struct lu_buf *buf, struct lustre_capa *capa)
3054 {
3055         struct osd_object      *obj    = osd_dt_obj(dt);
3056         struct inode           *inode  = obj->oo_inode;
3057         struct osd_thread_info *info   = osd_oti_get(env);
3058         struct dentry          *dentry = &info->oti_obj_dentry;
3059
3060         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3061         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3062
3063         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3064                 return -EACCES;
3065
3066         dentry->d_inode = inode;
3067         dentry->d_sb = inode->i_sb;
3068         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3069 }
3070
3071 static int osd_declare_xattr_del(const struct lu_env *env,
3072                                  struct dt_object *dt, const char *name,
3073                                  struct thandle *handle)
3074 {
3075         struct osd_thandle *oh;
3076
3077         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3078         LASSERT(handle != NULL);
3079
3080         oh = container_of0(handle, struct osd_thandle, ot_super);
3081         LASSERT(oh->ot_handle == NULL);
3082
3083         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3084                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3085
3086         return 0;
3087 }
3088
3089 /*
3090  * Concurrency: @dt is write locked.
3091  */
3092 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3093                          const char *name, struct thandle *handle,
3094                          struct lustre_capa *capa)
3095 {
3096         struct osd_object      *obj    = osd_dt_obj(dt);
3097         struct inode           *inode  = obj->oo_inode;
3098         struct osd_thread_info *info   = osd_oti_get(env);
3099         struct dentry          *dentry = &info->oti_obj_dentry;
3100         int                     rc;
3101
3102         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3103         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3104         LASSERT(handle != NULL);
3105
3106         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3107                 return -EACCES;
3108
3109         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3110
3111         ll_vfs_dq_init(inode);
3112         dentry->d_inode = inode;
3113         dentry->d_sb = inode->i_sb;
3114         rc = inode->i_op->removexattr(dentry, name);
3115         return rc;
3116 }
3117
3118 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3119                                      struct dt_object *dt,
3120                                      struct lustre_capa *old, __u64 opc)
3121 {
3122         struct osd_thread_info *info = osd_oti_get(env);
3123         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3124         struct osd_object *obj = osd_dt_obj(dt);
3125         struct osd_device *osd = osd_obj2dev(obj);
3126         struct lustre_capa_key *key = &info->oti_capa_key;
3127         struct lustre_capa *capa = &info->oti_capa;
3128         struct obd_capa *oc;
3129         struct lu_capainfo *lci;
3130         int rc;
3131         ENTRY;
3132
3133         if (!osd->od_fl_capa)
3134                 RETURN(ERR_PTR(-ENOENT));
3135
3136         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3137         LINVRNT(osd_invariant(obj));
3138
3139         /* renewal sanity check */
3140         if (old && osd_object_auth(env, dt, old, opc))
3141                 RETURN(ERR_PTR(-EACCES));
3142
3143         lci = lu_capainfo_get(env);
3144         if (unlikely(lci == NULL))
3145                 RETURN(ERR_PTR(-ENOENT));
3146
3147         switch (lci->lci_auth) {
3148         case LC_ID_NONE:
3149                 RETURN(NULL);
3150         case LC_ID_PLAIN:
3151                 capa->lc_uid = i_uid_read(obj->oo_inode);
3152                 capa->lc_gid = i_gid_read(obj->oo_inode);
3153                 capa->lc_flags = LC_ID_PLAIN;
3154                 break;
3155         case LC_ID_CONVERT: {
3156                 __u32 d[4], s[4];
3157
3158                 s[0] = i_uid_read(obj->oo_inode);
3159                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3160                 s[2] = i_uid_read(obj->oo_inode);
3161                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3162                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3163                 if (unlikely(rc))
3164                         RETURN(ERR_PTR(rc));
3165
3166                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3167                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3168                 capa->lc_flags = LC_ID_CONVERT;
3169                 break;
3170         }
3171         default:
3172                 RETURN(ERR_PTR(-EINVAL));
3173         }
3174
3175         capa->lc_fid = *fid;
3176         capa->lc_opc = opc;
3177         capa->lc_flags |= osd->od_capa_alg << 24;
3178         capa->lc_timeout = osd->od_capa_timeout;
3179         capa->lc_expiry = 0;
3180
3181         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3182         if (oc) {
3183                 LASSERT(!capa_is_expired(oc));
3184                 RETURN(oc);
3185         }
3186
3187         spin_lock(&capa_lock);
3188         *key = osd->od_capa_keys[1];
3189         spin_unlock(&capa_lock);
3190
3191         capa->lc_keyid = key->lk_keyid;
3192         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3193
3194         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3195         if (rc) {
3196                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3197                 RETURN(ERR_PTR(rc));
3198         }
3199
3200         oc = capa_add(osd->od_capa_hash, capa);
3201         RETURN(oc);
3202 }
3203
3204 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3205                            __u64 start, __u64 end)
3206 {
3207         struct osd_object       *obj    = osd_dt_obj(dt);
3208         struct inode            *inode  = obj->oo_inode;
3209         struct osd_thread_info  *info   = osd_oti_get(env);
3210         struct dentry           *dentry = &info->oti_obj_dentry;
3211         struct file             *file   = &info->oti_file;
3212         int                     rc;
3213
3214         ENTRY;
3215
3216         dentry->d_inode = inode;
3217         dentry->d_sb = inode->i_sb;
3218         file->f_dentry = dentry;
3219         file->f_mapping = inode->i_mapping;
3220         file->f_op = inode->i_fop;
3221         set_file_inode(file, inode);
3222
3223 #ifdef HAVE_FILE_FSYNC_4ARGS
3224         rc = file->f_op->fsync(file, start, end, 0);
3225 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3226         mutex_lock(&inode->i_mutex);
3227         rc = file->f_op->fsync(file, 0);
3228         mutex_unlock(&inode->i_mutex);
3229 #else
3230         mutex_lock(&inode->i_mutex);
3231         rc = file->f_op->fsync(file, dentry, 0);
3232         mutex_unlock(&inode->i_mutex);
3233 #endif
3234
3235         RETURN(rc);
3236 }
3237
3238 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3239                         void **data)
3240 {
3241         struct osd_object *obj = osd_dt_obj(dt);
3242         ENTRY;
3243
3244         *data = (void *)obj->oo_inode;
3245         RETURN(0);
3246 }
3247
3248 /*
3249  * Index operations.
3250  */
3251
3252 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3253                            const struct dt_index_features *feat)
3254 {
3255         struct iam_descr *descr;
3256
3257         if (osd_object_is_root(o))
3258                 return feat == &dt_directory_features;
3259
3260         LASSERT(o->oo_dir != NULL);
3261
3262         descr = o->oo_dir->od_container.ic_descr;
3263         if (feat == &dt_directory_features) {
3264                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3265                         return 1;
3266                 else
3267                         return 0;
3268         } else {
3269                 return
3270                         feat->dif_keysize_min <= descr->id_key_size &&
3271                         descr->id_key_size <= feat->dif_keysize_max &&
3272                         feat->dif_recsize_min <= descr->id_rec_size &&
3273                         descr->id_rec_size <= feat->dif_recsize_max &&
3274                         !(feat->dif_flags & (DT_IND_VARKEY |
3275                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3276                         ergo(feat->dif_flags & DT_IND_UPDATE,
3277                              1 /* XXX check that object (and file system) is
3278                                 * writable */);
3279         }
3280 }
3281
3282 static int osd_iam_container_init(const struct lu_env *env,
3283                                   struct osd_object *obj,
3284                                   struct osd_directory *dir)
3285 {
3286         struct iam_container *bag = &dir->od_container;
3287         int result;
3288
3289         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3290         if (result != 0)
3291                 return result;
3292
3293         result = iam_container_setup(bag);
3294         if (result == 0)
3295                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3296         else
3297                 iam_container_fini(bag);
3298
3299         return result;
3300 }
3301
3302
3303 /*
3304  * Concurrency: no external locking is necessary.
3305  */
3306 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3307                          const struct dt_index_features *feat)
3308 {
3309         int                      result;
3310         int                      skip_iam = 0;
3311         struct osd_object       *obj = osd_dt_obj(dt);
3312
3313         LINVRNT(osd_invariant(obj));
3314
3315         if (osd_object_is_root(obj)) {
3316                 dt->do_index_ops = &osd_index_ea_ops;
3317                 result = 0;
3318         } else if (feat == &dt_directory_features) {
3319                 dt->do_index_ops = &osd_index_ea_ops;
3320                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3321                         result = 0;
3322                 else
3323                         result = -ENOTDIR;
3324                 skip_iam = 1;
3325         } else if (unlikely(feat == &dt_otable_features)) {
3326                 dt->do_index_ops = &osd_otable_ops;
3327                 return 0;
3328         } else if (unlikely(feat == &dt_acct_features)) {
3329                 dt->do_index_ops = &osd_acct_index_ops;
3330                 result = 0;
3331                 skip_iam = 1;
3332         } else if (!osd_has_index(obj)) {
3333                 struct osd_directory *dir;
3334
3335                 OBD_ALLOC_PTR(dir);
3336                 if (dir != NULL) {
3337
3338                         spin_lock(&obj->oo_guard);
3339                         if (obj->oo_dir == NULL)
3340                                 obj->oo_dir = dir;
3341                         else
3342                                 /*
3343                                  * Concurrent thread allocated container data.
3344                                  */
3345                                 OBD_FREE_PTR(dir);
3346                         spin_unlock(&obj->oo_guard);
3347                         /*
3348                          * Now, that we have container data, serialize its
3349                          * initialization.
3350                          */
3351                         down_write(&obj->oo_ext_idx_sem);
3352                         /*
3353                          * recheck under lock.
3354                          */
3355                         if (!osd_has_index(obj))
3356                                 result = osd_iam_container_init(env, obj,
3357                                                                 obj->oo_dir);
3358                         else
3359                                 result = 0;
3360                         up_write(&obj->oo_ext_idx_sem);
3361                 } else {
3362                         result = -ENOMEM;
3363                 }
3364         } else {
3365                 result = 0;
3366         }
3367
3368         if (result == 0 && skip_iam == 0) {
3369                 if (!osd_iam_index_probe(env, obj, feat))
3370                         result = -ENOTDIR;
3371         }
3372         LINVRNT(osd_invariant(obj));
3373
3374         if (result == 0 && is_quota_glb_feat(feat) &&
3375             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3376                 result = osd_quota_migration(env, dt, feat);
3377
3378         return result;
3379 }
3380
3381 static int osd_otable_it_attr_get(const struct lu_env *env,
3382                                  struct dt_object *dt,
3383                                  struct lu_attr *attr,
3384                                  struct lustre_capa *capa)
3385 {
3386         attr->la_valid = 0;
3387         return 0;
3388 }
3389
3390 static const struct dt_object_operations osd_obj_ops = {
3391         .do_read_lock         = osd_object_read_lock,
3392         .do_write_lock        = osd_object_write_lock,
3393         .do_read_unlock       = osd_object_read_unlock,
3394         .do_write_unlock      = osd_object_write_unlock,
3395         .do_write_locked      = osd_object_write_locked,
3396         .do_attr_get          = osd_attr_get,
3397         .do_declare_attr_set  = osd_declare_attr_set,
3398         .do_attr_set          = osd_attr_set,
3399         .do_ah_init           = osd_ah_init,
3400         .do_declare_create    = osd_declare_object_create,
3401         .do_create            = osd_object_create,
3402         .do_declare_destroy   = osd_declare_object_destroy,
3403         .do_destroy           = osd_object_destroy,
3404         .do_index_try         = osd_index_try,
3405         .do_declare_ref_add   = osd_declare_object_ref_add,
3406         .do_ref_add           = osd_object_ref_add,
3407         .do_declare_ref_del   = osd_declare_object_ref_del,
3408         .do_ref_del           = osd_object_ref_del,
3409         .do_xattr_get         = osd_xattr_get,
3410         .do_declare_xattr_set = osd_declare_xattr_set,
3411         .do_xattr_set         = osd_xattr_set,
3412         .do_declare_xattr_del = osd_declare_xattr_del,
3413         .do_xattr_del         = osd_xattr_del,
3414         .do_xattr_list        = osd_xattr_list,
3415         .do_capa_get          = osd_capa_get,
3416         .do_object_sync       = osd_object_sync,
3417         .do_data_get          = osd_data_get,
3418 };
3419
3420 /**
3421  * dt_object_operations for interoperability mode
3422  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3423  */
3424 static const struct dt_object_operations osd_obj_ea_ops = {
3425         .do_read_lock         = osd_object_read_lock,
3426         .do_write_lock        = osd_object_write_lock,
3427         .do_read_unlock       = osd_object_read_unlock,
3428         .do_write_unlock      = osd_object_write_unlock,
3429         .do_write_locked      = osd_object_write_locked,
3430         .do_attr_get          = osd_attr_get,
3431         .do_declare_attr_set  = osd_declare_attr_set,
3432         .do_attr_set          = osd_attr_set,
3433         .do_ah_init           = osd_ah_init,
3434         .do_declare_create    = osd_declare_object_create,
3435         .do_create            = osd_object_ea_create,
3436         .do_declare_destroy   = osd_declare_object_destroy,
3437         .do_destroy           = osd_object_destroy,
3438         .do_index_try         = osd_index_try,
3439         .do_declare_ref_add   = osd_declare_object_ref_add,
3440         .do_ref_add           = osd_object_ref_add,
3441         .do_declare_ref_del   = osd_declare_object_ref_del,
3442         .do_ref_del           = osd_object_ref_del,
3443         .do_xattr_get         = osd_xattr_get,
3444         .do_declare_xattr_set = osd_declare_xattr_set,
3445         .do_xattr_set         = osd_xattr_set,
3446         .do_declare_xattr_del = osd_declare_xattr_del,
3447         .do_xattr_del         = osd_xattr_del,
3448         .do_xattr_list        = osd_xattr_list,
3449         .do_capa_get          = osd_capa_get,
3450         .do_object_sync       = osd_object_sync,
3451         .do_data_get          = osd_data_get,
3452 };
3453
3454 static const struct dt_object_operations osd_obj_otable_it_ops = {
3455         .do_attr_get    = osd_otable_it_attr_get,
3456         .do_index_try   = osd_index_try,
3457 };
3458
3459 static int osd_index_declare_iam_delete(const struct lu_env *env,
3460                                         struct dt_object *dt,
3461                                         const struct dt_key *key,
3462                                         struct thandle *handle)
3463 {
3464         struct osd_thandle    *oh;
3465
3466         oh = container_of0(handle, struct osd_thandle, ot_super);
3467         LASSERT(oh->ot_handle == NULL);
3468
3469         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3470                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3471
3472         return 0;
3473 }
3474
3475 /**
3476  *      delete a (key, value) pair from index \a dt specified by \a key
3477  *
3478  *      \param  dt      osd index object
3479  *      \param  key     key for index
3480  *      \param  rec     record reference
3481  *      \param  handle  transaction handler
3482  *
3483  *      \retval  0  success
3484  *      \retval -ve   failure
3485  */
3486
3487 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3488                                 const struct dt_key *key,
3489                                 struct thandle *handle,
3490                                 struct lustre_capa *capa)
3491 {
3492         struct osd_thread_info *oti = osd_oti_get(env);
3493         struct osd_object      *obj = osd_dt_obj(dt);
3494         struct osd_thandle     *oh;
3495         struct iam_path_descr  *ipd;
3496         struct iam_container   *bag = &obj->oo_dir->od_container;
3497         int                     rc;
3498
3499         ENTRY;
3500
3501         LINVRNT(osd_invariant(obj));
3502         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3503         LASSERT(bag->ic_object == obj->oo_inode);
3504         LASSERT(handle != NULL);
3505
3506         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3507                 RETURN(-EACCES);
3508
3509         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3510
3511         ipd = osd_idx_ipd_get(env, bag);
3512         if (unlikely(ipd == NULL))
3513                 RETURN(-ENOMEM);
3514
3515         oh = container_of0(handle, struct osd_thandle, ot_super);
3516         LASSERT(oh->ot_handle != NULL);
3517         LASSERT(oh->ot_handle->h_transaction != NULL);
3518
3519         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3520                 /* swab quota uid/gid provided by caller */
3521                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3522                 key = (const struct dt_key *)&oti->oti_quota_id;
3523         }
3524
3525         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3526         osd_ipd_put(env, bag, ipd);
3527         LINVRNT(osd_invariant(obj));
3528         RETURN(rc);
3529 }
3530
3531 static int osd_index_declare_ea_delete(const struct lu_env *env,
3532                                        struct dt_object *dt,
3533                                        const struct dt_key *key,
3534                                        struct thandle *handle)
3535 {
3536         struct osd_thandle *oh;
3537         struct inode       *inode;
3538         int                 rc;
3539         ENTRY;
3540
3541         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3542         LASSERT(handle != NULL);
3543
3544         oh = container_of0(handle, struct osd_thandle, ot_super);
3545         LASSERT(oh->ot_handle == NULL);
3546
3547         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3548                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3549
3550         inode = osd_dt_obj(dt)->oo_inode;
3551         LASSERT(inode);
3552
3553         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3554                                    0, oh, true, true, NULL, false);
3555         RETURN(rc);
3556 }
3557
3558 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3559                                           struct dt_rec *fid)
3560 {
3561         struct osd_fid_pack *rec;
3562         int                  rc = -ENODATA;
3563
3564         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3565                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3566                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3567         }
3568         return rc;
3569 }
3570
3571 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3572                           struct lu_fid *fid)
3573 {
3574         struct seq_server_site  *ss = osd_seq_site(osd);
3575         ENTRY;
3576
3577         /* FID seqs not in FLDB, must be local seq */
3578         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3579                 RETURN(0);
3580
3581         /* If FLD is not being initialized yet, it only happens during the
3582          * initialization, likely during mgs initialization, and we assume
3583          * this is local FID. */
3584         if (ss == NULL || ss->ss_server_fld == NULL)
3585                 RETURN(0);
3586
3587         /* Only check the local FLDB here */
3588         if (osd_seq_exists(env, osd, fid_seq(fid)))
3589                 RETURN(0);
3590
3591         RETURN(1);
3592 }
3593
3594 /**
3595  * Index delete function for interoperability mode (b11826).
3596  * It will remove the directory entry added by osd_index_ea_insert().
3597  * This entry is needed to maintain name->fid mapping.
3598  *
3599  * \param key,  key i.e. file entry to be deleted
3600  *
3601  * \retval   0, on success
3602  * \retval -ve, on error
3603  */
3604 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3605                                const struct dt_key *key,
3606                                struct thandle *handle,
3607                                struct lustre_capa *capa)
3608 {
3609         struct osd_object          *obj    = osd_dt_obj(dt);
3610         struct inode               *dir    = obj->oo_inode;
3611         struct dentry              *dentry;
3612         struct osd_thandle         *oh;
3613         struct ldiskfs_dir_entry_2 *de = NULL;
3614         struct buffer_head         *bh;
3615         struct htree_lock          *hlock = NULL;
3616         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3617         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3618         int                        rc;
3619         ENTRY;
3620
3621         LINVRNT(osd_invariant(obj));
3622         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3623         LASSERT(handle != NULL);
3624
3625         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3626
3627         oh = container_of(handle, struct osd_thandle, ot_super);
3628         LASSERT(oh->ot_handle != NULL);
3629         LASSERT(oh->ot_handle->h_transaction != NULL);
3630
3631         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3632                 RETURN(-EACCES);
3633
3634         ll_vfs_dq_init(dir);
3635         dentry = osd_child_dentry_get(env, obj,
3636                                       (char *)key, strlen((char *)key));
3637
3638         if (obj->oo_hl_head != NULL) {
3639                 hlock = osd_oti_get(env)->oti_hlock;
3640                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3641                                    dir, LDISKFS_HLOCK_DEL);
3642         } else {
3643                 down_write(&obj->oo_ext_idx_sem);
3644         }
3645
3646         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3647         if (bh) {
3648                 __u32 ino = 0;
3649
3650                 /* If this is not the ".." entry, it might be a remote DNE
3651                  * entry and  we need to check if the FID is for a remote
3652                  * MDT.  If the FID is  not in the directory entry (e.g.
3653                  * upgraded 1.8 filesystem without dirdata enabled) then
3654                  * we need to get the FID from the LMA. For a remote directory
3655                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3656                  * case.
3657                  *
3658                  * Delete the entry before the agent inode in order to
3659                  * simplify error handling.  At worst an error after deleting
3660                  * the entry first might leak the agent inode afterward. The
3661                  * reverse would need filesystem abort in case of error deleting
3662                  * the entry after the agent had been removed, or leave a
3663                  * dangling entry pointing at a random inode. */
3664                 if (strcmp((char *)key, dotdot) != 0) {
3665                         LASSERT(de != NULL);
3666                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3667                         /* If Fid is not in dentry, try to get it from LMA */
3668                         if (rc == -ENODATA) {
3669                                 struct osd_inode_id *id;
3670                                 struct inode *inode;
3671
3672                                 /* Before trying to get fid from the inode,
3673                                  * check whether the inode is valid.
3674                                  *
3675                                  * If the inode has been deleted, do not go
3676                                  * ahead to do osd_ea_fid_get, which will set
3677                                  * the inode to bad inode, which might cause
3678                                  * the inode to be deleted uncorrectly */
3679                                 inode = ldiskfs_iget(osd_sb(osd),
3680                                                      le32_to_cpu(de->inode));
3681                                 if (IS_ERR(inode)) {
3682                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3683                                                "error.\n", osd_name(osd),
3684                                                PFID(fid));
3685                                         rc = PTR_ERR(inode);
3686                                 } else {
3687                                         if (likely(inode->i_nlink != 0)) {
3688                                                 id = &osd_oti_get(env)->oti_id;
3689                                                 rc = osd_ea_fid_get(env, obj,
3690                                                         le32_to_cpu(de->inode),
3691                                                                     fid, id);
3692                                         } else {
3693                                                 CDEBUG(D_INFO, "%s: %u "DFID
3694                                                        "deleted.\n",
3695                                                        osd_name(osd),
3696                                                        le32_to_cpu(de->inode),
3697                                                        PFID(fid));
3698                                                 rc = -ESTALE;
3699                                         }
3700                                         iput(inode);
3701                                 }
3702                         }
3703                         if (rc == 0 &&
3704                             unlikely(osd_remote_fid(env, osd, fid)))
3705                                 /* Need to delete agent inode */
3706                                 ino = le32_to_cpu(de->inode);
3707                 }
3708                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3709                 brelse(bh);
3710                 if (rc == 0 && unlikely(ino != 0)) {
3711                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3712                                                           oh);
3713                         if (rc != 0)
3714                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3715                                        osd_name(osd), PFID(fid), rc);
3716                 }
3717         } else {
3718                 rc = -ENOENT;
3719         }
3720         if (hlock != NULL)
3721                 ldiskfs_htree_unlock(hlock);
3722         else
3723                 up_write(&obj->oo_ext_idx_sem);
3724
3725         if (rc != 0)
3726                 GOTO(out, rc);
3727
3728         /* For inode on the remote MDT, .. will point to
3729          * /Agent directory, Check whether it needs to delete
3730          * from agent directory */
3731         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3732                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3733                                                    oh);
3734                 if (rc != 0 && rc != -ENOENT) {
3735                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3736                                osd_name(osd), PFID(fid), rc);
3737                 }
3738
3739                 if (rc == -ENOENT)
3740                         rc = 0;
3741
3742                 GOTO(out, rc);
3743         }
3744 out:
3745
3746         LASSERT(osd_invariant(obj));
3747         RETURN(rc);
3748 }
3749
3750 /**
3751  *      Lookup index for \a key and copy record to \a rec.
3752  *
3753  *      \param  dt      osd index object
3754  *      \param  key     key for index
3755  *      \param  rec     record reference
3756  *
3757  *      \retval  +ve  success : exact mach
3758  *      \retval  0    return record with key not greater than \a key
3759  *      \retval -ve   failure
3760  */
3761 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3762                                 struct dt_rec *rec, const struct dt_key *key,
3763                                 struct lustre_capa *capa)
3764 {
3765         struct osd_object      *obj = osd_dt_obj(dt);
3766         struct iam_path_descr  *ipd;
3767         struct iam_container   *bag = &obj->oo_dir->od_container;
3768         struct osd_thread_info *oti = osd_oti_get(env);
3769         struct iam_iterator    *it = &oti->oti_idx_it;
3770         struct iam_rec         *iam_rec;
3771         int                     rc;
3772
3773         ENTRY;
3774
3775         LASSERT(osd_invariant(obj));
3776         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3777         LASSERT(bag->ic_object == obj->oo_inode);
3778
3779         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3780                 RETURN(-EACCES);
3781
3782         ipd = osd_idx_ipd_get(env, bag);
3783         if (IS_ERR(ipd))
3784                 RETURN(-ENOMEM);
3785
3786         /* got ipd now we can start iterator. */
3787         iam_it_init(it, bag, 0, ipd);
3788
3789         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3790                 /* swab quota uid/gid provided by caller */
3791                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3792                 key = (const struct dt_key *)&oti->oti_quota_id;
3793         }
3794
3795         rc = iam_it_get(it, (struct iam_key *)key);
3796         if (rc >= 0) {
3797                 if (S_ISDIR(obj->oo_inode->i_mode))
3798                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3799                 else
3800                         iam_rec = (struct iam_rec *) rec;
3801
3802                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3803
3804                 if (S_ISDIR(obj->oo_inode->i_mode))
3805                         osd_fid_unpack((struct lu_fid *) rec,
3806                                        (struct osd_fid_pack *)iam_rec);
3807                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3808                         osd_quota_unpack(obj, rec);
3809         }
3810
3811         iam_it_put(it);
3812         iam_it_fini(it);
3813         osd_ipd_put(env, bag, ipd);
3814
3815         LINVRNT(osd_invariant(obj));
3816
3817         RETURN(rc);
3818 }
3819
3820 static int osd_index_declare_iam_insert(const struct lu_env *env,
3821                                         struct dt_object *dt,
3822                                         const struct dt_rec *rec,
3823                                         const struct dt_key *key,
3824                                         struct thandle *handle)
3825 {
3826         struct osd_thandle *oh;
3827
3828         LASSERT(handle != NULL);
3829
3830         oh = container_of0(handle, struct osd_thandle, ot_super);
3831         LASSERT(oh->ot_handle == NULL);
3832
3833         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3834                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3835
3836         return 0;
3837 }
3838
3839 /**
3840  *      Inserts (key, value) pair in \a dt index object.
3841  *
3842  *      \param  dt      osd index object
3843  *      \param  key     key for index
3844  *      \param  rec     record reference
3845  *      \param  th      transaction handler
3846  *
3847  *      \retval  0  success
3848  *      \retval -ve failure
3849  */
3850 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3851                                 const struct dt_rec *rec,
3852                                 const struct dt_key *key, struct thandle *th,
3853                                 struct lustre_capa *capa, int ignore_quota)
3854 {
3855         struct osd_object     *obj = osd_dt_obj(dt);
3856         struct iam_path_descr *ipd;
3857         struct osd_thandle    *oh;
3858         struct iam_container  *bag = &obj->oo_dir->od_container;
3859         struct osd_thread_info *oti = osd_oti_get(env);
3860         struct iam_rec         *iam_rec;
3861         int                     rc;
3862
3863         ENTRY;
3864
3865         LINVRNT(osd_invariant(obj));
3866         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3867         LASSERT(bag->ic_object == obj->oo_inode);
3868         LASSERT(th != NULL);
3869
3870         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3871                 RETURN(-EACCES);
3872
3873         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3874
3875         ipd = osd_idx_ipd_get(env, bag);
3876         if (unlikely(ipd == NULL))
3877                 RETURN(-ENOMEM);
3878
3879         oh = container_of0(th, struct osd_thandle, ot_super);
3880         LASSERT(oh->ot_handle != NULL);
3881         LASSERT(oh->ot_handle->h_transaction != NULL);
3882         if (S_ISDIR(obj->oo_inode->i_mode)) {
3883                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3884                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3885         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3886                 /* pack quota uid/gid */
3887                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3888                 key = (const struct dt_key *)&oti->oti_quota_id;
3889                 /* pack quota record */
3890                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3891                 iam_rec = (struct iam_rec *)rec;
3892         } else {
3893                 iam_rec = (struct iam_rec *)rec;
3894         }
3895
3896         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3897                         iam_rec, ipd);
3898         osd_ipd_put(env, bag, ipd);
3899         LINVRNT(osd_invariant(obj));
3900         RETURN(rc);
3901 }
3902
3903 /**
3904  * Calls ldiskfs_add_entry() to add directory entry
3905  * into the directory. This is required for
3906  * interoperability mode (b11826)
3907  *
3908  * \retval   0, on success
3909  * \retval -ve, on error
3910  */
3911 static int __osd_ea_add_rec(struct osd_thread_info *info,
3912                             struct osd_object *pobj, struct inode  *cinode,
3913                             const char *name, const struct dt_rec *fid,
3914                             struct htree_lock *hlock, struct thandle *th)
3915 {
3916         struct ldiskfs_dentry_param *ldp;
3917         struct dentry               *child;
3918         struct osd_thandle          *oth;
3919         int                          rc;
3920
3921         oth = container_of(th, struct osd_thandle, ot_super);
3922         LASSERT(oth->ot_handle != NULL);
3923         LASSERT(oth->ot_handle->h_transaction != NULL);
3924         LASSERT(pobj->oo_inode);
3925
3926         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3927         if (unlikely(pobj->oo_inode ==
3928                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3929                 ldp->edp_magic = 0;
3930         else
3931                 osd_get_ldiskfs_dirent_param(ldp, fid);
3932         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3933         child->d_fsdata = (void *)ldp;
3934         ll_vfs_dq_init(pobj->oo_inode);
3935         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3936
3937         RETURN(rc);
3938 }
3939
3940 /**
3941  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3942  * into the directory.Also sets flags into osd object to
3943  * indicate dot and dotdot are created. This is required for
3944  * interoperability mode (b11826)
3945  *
3946  * \param dir   directory for dot and dotdot fixup.
3947  * \param obj   child object for linking
3948  *
3949  * \retval   0, on success
3950  * \retval -ve, on error
3951  */
3952 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3953                               struct osd_object *dir,
3954                               struct inode  *parent_dir, const char *name,
3955                               const struct dt_rec *dot_fid,
3956                               const struct dt_rec *dot_dot_fid,
3957                               struct thandle *th)
3958 {
3959         struct inode                *inode = dir->oo_inode;
3960         struct osd_thandle          *oth;
3961         int result = 0;
3962
3963         oth = container_of(th, struct osd_thandle, ot_super);
3964         LASSERT(oth->ot_handle->h_transaction != NULL);
3965         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3966
3967         if (strcmp(name, dot) == 0) {
3968                 if (dir->oo_compat_dot_created) {
3969                         result = -EEXIST;
3970                 } else {
3971                         LASSERT(inode == parent_dir);
3972                         dir->oo_compat_dot_created = 1;
3973                         result = 0;
3974                 }
3975         } else if (strcmp(name, dotdot) == 0) {
3976                 if (!dir->oo_compat_dot_created)
3977                         return -EINVAL;
3978                 /* in case of rename, dotdot is already created */
3979                 if (dir->oo_compat_dotdot_created) {
3980                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3981                                                 dot_dot_fid, NULL, th);
3982                 }
3983
3984                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3985                                                 parent_dir, dot_fid,
3986                                                 dot_dot_fid, oth);
3987                 if (result == 0)
3988                         dir->oo_compat_dotdot_created = 1;
3989         }
3990
3991         return result;
3992 }
3993
3994
3995 /**
3996  * It will call the appropriate osd_add* function and return the
3997  * value, return by respective functions.
3998  */
3999 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4000                           struct inode *cinode, const char *name,
4001                           const struct dt_rec *fid, struct thandle *th)
4002 {
4003         struct osd_thread_info *info   = osd_oti_get(env);
4004         struct htree_lock      *hlock;
4005         int                     rc;
4006
4007         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4008
4009         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4010                                                    name[2] =='\0'))) {
4011                 if (hlock != NULL) {
4012                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4013                                            pobj->oo_inode, 0);
4014                 } else {
4015                         down_write(&pobj->oo_ext_idx_sem);
4016                 }
4017                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4018                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
4019                                         fid, th);
4020         } else {
4021                 if (hlock != NULL) {
4022                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4023                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4024                 } else {
4025                         down_write(&pobj->oo_ext_idx_sem);
4026                 }
4027
4028                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4029                         struct lu_fid *tfid = &info->oti_fid;
4030
4031                         *tfid = *(const struct lu_fid *)fid;
4032                         tfid->f_ver = ~0;
4033                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4034                                               (const struct dt_rec *)tfid,
4035                                               hlock, th);
4036                 } else {
4037                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4038                                               hlock, th);
4039                 }
4040         }
4041         if (hlock != NULL)
4042                 ldiskfs_htree_unlock(hlock);
4043         else
4044                 up_write(&pobj->oo_ext_idx_sem);
4045
4046         return rc;
4047 }
4048
4049 static void
4050 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4051                       struct osd_idmap_cache *oic)
4052 {
4053         struct osd_scrub    *scrub = &dev->od_scrub;
4054         struct lu_fid       *fid   = &oic->oic_fid;
4055         struct osd_inode_id *id    = &oti->oti_id;
4056         int                  once  = 0;
4057         int                  rc;
4058         ENTRY;
4059
4060         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4061                 RETURN_EXIT;
4062
4063 again:
4064         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4065         if (rc != 0 && rc != -ENOENT)
4066                 RETURN_EXIT;
4067
4068         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4069                 RETURN_EXIT;
4070
4071         if (thread_is_running(&scrub->os_thread)) {
4072                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4073                 /* There is race condition between osd_oi_lookup and OI scrub.
4074                  * The OI scrub finished just after osd_oi_lookup() failure.
4075                  * Under such case, it is unnecessary to trigger OI scrub again,
4076                  * but try to call osd_oi_lookup() again. */
4077                 if (unlikely(rc == -EAGAIN))
4078                         goto again;
4079
4080                 RETURN_EXIT;
4081         }
4082
4083         if (!dev->od_noscrub && ++once == 1) {
4084                 rc = osd_scrub_start(dev);
4085                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4086                               ", rc = %d [2]\n",
4087                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4088                               PFID(fid), rc);
4089                 if (rc == 0)
4090                         goto again;
4091         }
4092
4093         EXIT;
4094 }
4095
4096 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4097                                struct osd_device *dev,
4098                                struct osd_idmap_cache *oic,
4099                                struct lu_fid *fid, __u32 ino)
4100 {
4101         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4102         struct inode            *inode;
4103         int                      rc;
4104
4105         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4106         inode = osd_iget(oti, dev, &oic->oic_lid);
4107         if (IS_ERR(inode)) {
4108                 fid_zero(&oic->oic_fid);
4109                 return PTR_ERR(inode);
4110         }
4111
4112         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4113         iput(inode);
4114         if (rc != 0)
4115                 fid_zero(&oic->oic_fid);
4116         else
4117                 *fid = oic->oic_fid = lma->lma_self_fid;
4118         return rc;
4119 }
4120
4121 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4122                      struct osd_inode_id *id, const struct lu_fid *fid)
4123 {
4124         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4125                id->oii_ino, id->oii_gen, info);
4126         info->oti_cache.oic_lid = *id;
4127         info->oti_cache.oic_fid = *fid;
4128         info->oti_cache.oic_dev = osd;
4129
4130         return 0;
4131 }
4132
4133 /**
4134  * Calls ->lookup() to find dentry. From dentry get inode and
4135  * read inode's ea to get fid. This is required for  interoperability
4136  * mode (b11826)
4137  *
4138  * \retval   0, on success
4139  * \retval -ve, on error
4140  */
4141 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4142                              struct dt_rec *rec, const struct dt_key *key)
4143 {
4144         struct inode               *dir    = obj->oo_inode;
4145         struct dentry              *dentry;
4146         struct ldiskfs_dir_entry_2 *de;
4147         struct buffer_head         *bh;
4148         struct lu_fid              *fid = (struct lu_fid *) rec;
4149         struct htree_lock          *hlock = NULL;
4150         int                         ino;
4151         int                         rc;
4152         ENTRY;
4153
4154         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4155
4156         dentry = osd_child_dentry_get(env, obj,
4157                                       (char *)key, strlen((char *)key));
4158
4159         if (obj->oo_hl_head != NULL) {
4160                 hlock = osd_oti_get(env)->oti_hlock;
4161                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4162                                    dir, LDISKFS_HLOCK_LOOKUP);
4163         } else {
4164                 down_read(&obj->oo_ext_idx_sem);
4165         }
4166
4167         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4168         if (bh) {
4169                 struct osd_thread_info *oti = osd_oti_get(env);
4170                 struct osd_inode_id *id = &oti->oti_id;
4171                 struct osd_idmap_cache *oic = &oti->oti_cache;
4172                 struct osd_device *dev = osd_obj2dev(obj);
4173                 struct osd_scrub *scrub = &dev->od_scrub;
4174                 struct scrub_file *sf = &scrub->os_file;
4175
4176                 ino = le32_to_cpu(de->inode);
4177                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4178                         brelse(bh);
4179                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4180                         GOTO(out, rc);
4181                 }
4182
4183                 rc = osd_get_fid_from_dentry(de, rec);
4184
4185                 /* done with de, release bh */
4186                 brelse(bh);
4187                 if (rc != 0)
4188                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
4189                 else
4190                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4191                 if (rc != 0) {
4192                         fid_zero(&oic->oic_fid);
4193                         GOTO(out, rc);
4194                 }
4195
4196                 if (osd_remote_fid(env, dev, fid))
4197                         GOTO(out, rc = 0);
4198
4199                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4200                                       fid);
4201                 if (rc != 0)
4202                         GOTO(out, rc);
4203                 if ((scrub->os_pos_current <= ino) &&
4204                     ((sf->sf_flags & SF_INCONSISTENT) ||
4205                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4206                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4207                                       sf->sf_oi_bitmap)))
4208                         osd_consistency_check(oti, dev, oic);
4209         } else {
4210                 rc = -ENOENT;
4211         }
4212
4213         GOTO(out, rc);
4214
4215 out:
4216         if (hlock != NULL)
4217                 ldiskfs_htree_unlock(hlock);
4218         else
4219                 up_read(&obj->oo_ext_idx_sem);
4220         return rc;
4221 }
4222
4223 /**
4224  * Find the osd object for given fid.
4225  *
4226  * \param fid need to find the osd object having this fid
4227  *
4228  * \retval osd_object on success
4229  * \retval        -ve on error
4230  */
4231 struct osd_object *osd_object_find(const struct lu_env *env,
4232                                    struct dt_object *dt,
4233                                    const struct lu_fid *fid)
4234 {
4235         struct lu_device  *ludev = dt->do_lu.lo_dev;
4236         struct osd_object *child = NULL;
4237         struct lu_object  *luch;
4238         struct lu_object  *lo;
4239
4240         /*
4241          * at this point topdev might not exist yet
4242          * (i.e. MGS is preparing profiles). so we can
4243          * not rely on topdev and instead lookup with
4244          * our device passed as topdev. this can't work
4245          * if the object isn't cached yet (as osd doesn't
4246          * allocate lu_header). IOW, the object must be
4247          * in the cache, otherwise lu_object_alloc() crashes
4248          * -bzzz
4249          */
4250         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4251                                  ludev : ludev->ld_site->ls_top_dev,
4252                                  fid, NULL);
4253         if (!IS_ERR(luch)) {
4254                 if (lu_object_exists(luch)) {
4255                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4256                         if (lo != NULL)
4257                                 child = osd_obj(lo);
4258                         else
4259                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4260                                                 "lu_object can't be located"
4261                                                 DFID"\n", PFID(fid));
4262
4263                         if (child == NULL) {
4264                                 lu_object_put(env, luch);
4265                                 CERROR("Unable to get osd_object\n");
4266                                 child = ERR_PTR(-ENOENT);
4267                         }
4268                 } else {
4269                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4270                                         "lu_object does not exists "DFID"\n",
4271                                         PFID(fid));
4272                         lu_object_put(env, luch);
4273                         child = ERR_PTR(-ENOENT);
4274                 }
4275         } else {
4276                 child = ERR_CAST(luch);
4277         }
4278
4279         return child;
4280 }
4281
4282 /**
4283  * Put the osd object once done with it.
4284  *
4285  * \param obj osd object that needs to be put
4286  */
4287 static inline void osd_object_put(const struct lu_env *env,
4288                                   struct osd_object *obj)
4289 {
4290         lu_object_put(env, &obj->oo_dt.do_lu);
4291 }
4292
4293 static int osd_index_declare_ea_insert(const struct lu_env *env,
4294                                        struct dt_object *dt,
4295                                        const struct dt_rec *rec,
4296                                        const struct dt_key *key,
4297                                        struct thandle *handle)
4298 {
4299         struct osd_thandle      *oh;
4300         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4301         struct lu_fid           *fid = (struct lu_fid *)rec;
4302         int                     rc;
4303         ENTRY;
4304
4305         LASSERT(!dt_object_remote(dt));
4306         LASSERT(handle != NULL);
4307
4308         oh = container_of0(handle, struct osd_thandle, ot_super);
4309         LASSERT(oh->ot_handle == NULL);
4310
4311         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4312                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4313
4314         if (osd_dt_obj(dt)->oo_inode != NULL) {
4315                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4316
4317                 /* We ignore block quota on meta pool (MDTs), so needn't
4318                  * calculate how many blocks will be consumed by this index
4319                  * insert */
4320                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4321                                            i_gid_read(inode), 0,
4322                                            oh, true, true, NULL, false);
4323         }
4324
4325         if (fid == NULL)
4326                 RETURN(0);
4327
4328         rc = osd_remote_fid(env, osd, fid);
4329         if (rc <= 0)
4330                 RETURN(rc);
4331
4332         rc = 0;
4333
4334         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4335                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4336         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4337                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4338         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4339                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4340
4341         RETURN(rc);
4342 }
4343
4344 /**
4345  * Index add function for interoperability mode (b11826).
4346  * It will add the directory entry.This entry is needed to
4347  * maintain name->fid mapping.
4348  *
4349  * \param key it is key i.e. file entry to be inserted
4350  * \param rec it is value of given key i.e. fid
4351  *
4352  * \retval   0, on success
4353  * \retval -ve, on error
4354  */
4355 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4356                                const struct dt_rec *rec,
4357                                const struct dt_key *key, struct thandle *th,
4358                                struct lustre_capa *capa, int ignore_quota)
4359 {
4360         struct osd_object       *obj = osd_dt_obj(dt);
4361         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4362         struct lu_fid           *fid = (struct lu_fid *) rec;
4363         const char              *name = (const char *)key;
4364         struct osd_thread_info  *oti   = osd_oti_get(env);
4365         struct osd_inode_id     *id    = &oti->oti_id;
4366         struct inode            *child_inode = NULL;
4367         struct osd_object       *child = NULL;
4368         int                     rc;
4369         ENTRY;
4370
4371         LASSERT(osd_invariant(obj));
4372         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4373         LASSERT(th != NULL);
4374
4375         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4376
4377         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4378                 RETURN(-EACCES);
4379
4380         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4381
4382         rc = osd_remote_fid(env, osd, fid);
4383         if (rc < 0) {
4384                 CERROR("%s: Can not find object "DFID" rc %d\n",
4385                        osd_name(osd), PFID(fid), rc);
4386                 RETURN(rc);
4387         }
4388
4389         if (rc == 1) {
4390                 /* Insert remote entry */
4391                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4392                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4393                         struct osd_thandle      *oh;
4394
4395                         /* If parent on remote MDT, we need put this object
4396                          * under AGENT */
4397                         oh = container_of(th, typeof(*oh), ot_super);
4398                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4399                         if (rc != 0) {
4400                                 CERROR("%s: add "DFID" error: rc = %d\n",
4401                                        osd_name(osd),
4402                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4403                                 RETURN(rc);
4404                         }
4405
4406                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4407                 } else {
4408                         child_inode = osd_create_local_agent_inode(env, osd,
4409                                                                    obj, fid,
4410                                                                    th);
4411                         if (IS_ERR(child_inode))
4412                                 RETURN(PTR_ERR(child_inode));
4413                 }
4414         } else {
4415                 /* Insert local entry */
4416                 child = osd_object_find(env, dt, fid);
4417                 if (IS_ERR(child)) {
4418                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4419                                osd_name(osd), PFID(fid),
4420                                id->oii_ino, id->oii_gen,
4421                                (int)PTR_ERR(child));
4422                         RETURN(PTR_ERR(child));
4423                 }
4424                 child_inode = igrab(child->oo_inode);
4425         }
4426
4427         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4428
4429         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4430                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4431
4432         iput(child_inode);
4433         if (child != NULL)
4434                 osd_object_put(env, child);
4435         LASSERT(osd_invariant(obj));
4436         RETURN(rc);
4437 }
4438
4439 /**
4440  *  Initialize osd Iterator for given osd index object.
4441  *
4442  *  \param  dt      osd index object
4443  */
4444
4445 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4446                                      struct dt_object *dt,
4447                                      __u32 unused,
4448                                      struct lustre_capa *capa)
4449 {
4450         struct osd_it_iam      *it;
4451         struct osd_thread_info *oti = osd_oti_get(env);
4452         struct osd_object      *obj = osd_dt_obj(dt);
4453         struct lu_object       *lo  = &dt->do_lu;
4454         struct iam_path_descr  *ipd;
4455         struct iam_container   *bag = &obj->oo_dir->od_container;
4456
4457         LASSERT(lu_object_exists(lo));
4458
4459         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4460                 return ERR_PTR(-EACCES);
4461
4462         it = &oti->oti_it;
4463         ipd = osd_it_ipd_get(env, bag);
4464         if (likely(ipd != NULL)) {
4465                 it->oi_obj = obj;
4466                 it->oi_ipd = ipd;
4467                 lu_object_get(lo);
4468                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4469                 return (struct dt_it *)it;
4470         }
4471         return ERR_PTR(-ENOMEM);
4472 }
4473
4474 /**
4475  * free given Iterator.
4476  */
4477
4478 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4479 {
4480         struct osd_it_iam *it = (struct osd_it_iam *)di;
4481         struct osd_object *obj = it->oi_obj;
4482
4483         iam_it_fini(&it->oi_it);
4484         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4485         lu_object_put(env, &obj->oo_dt.do_lu);
4486 }
4487
4488 /**
4489  *  Move Iterator to record specified by \a key
4490  *
4491  *  \param  di      osd iterator
4492  *  \param  key     key for index
4493  *
4494  *  \retval +ve  di points to record with least key not larger than key
4495  *  \retval  0   di points to exact matched key
4496  *  \retval -ve  failure
4497  */
4498
4499 static int osd_it_iam_get(const struct lu_env *env,
4500                           struct dt_it *di, const struct dt_key *key)
4501 {
4502         struct osd_thread_info  *oti = osd_oti_get(env);
4503         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4504
4505         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4506                 /* swab quota uid/gid */
4507                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4508                 key = (struct dt_key *)&oti->oti_quota_id;
4509         }
4510
4511         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4512 }
4513
4514 /**
4515  *  Release Iterator
4516  *
4517  *  \param  di      osd iterator
4518  */
4519 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4520 {
4521         struct osd_it_iam *it = (struct osd_it_iam *)di;
4522
4523         iam_it_put(&it->oi_it);
4524 }
4525
4526 /**
4527  *  Move iterator by one record
4528  *
4529  *  \param  di      osd iterator
4530  *
4531  *  \retval +1   end of container reached
4532  *  \retval  0   success
4533  *  \retval -ve  failure
4534  */
4535
4536 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4537 {
4538         struct osd_it_iam *it = (struct osd_it_iam *)di;
4539
4540         return iam_it_next(&it->oi_it);
4541 }
4542
4543 /**
4544  * Return pointer to the key under iterator.
4545  */
4546
4547 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4548                                  const struct dt_it *di)
4549 {
4550         struct osd_thread_info *oti = osd_oti_get(env);
4551         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4552         struct osd_object      *obj = it->oi_obj;
4553         struct dt_key          *key;
4554
4555         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4556
4557         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4558                 /* swab quota uid/gid */
4559                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4560                 key = (struct dt_key *)&oti->oti_quota_id;
4561         }
4562
4563         return key;
4564 }
4565
4566 /**
4567  * Return size of key under iterator (in bytes)
4568  */
4569
4570 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4571 {
4572         struct osd_it_iam *it = (struct osd_it_iam *)di;
4573
4574         return iam_it_key_size(&it->oi_it);
4575 }
4576
4577 static inline void
4578 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4579 {
4580         /* check if file type is required */
4581         if (ent->lde_attrs & LUDA_TYPE) {
4582                 struct luda_type *lt;
4583                 int align = sizeof(*lt) - 1;
4584
4585                 len = (len + align) & ~align;
4586                 lt = (struct luda_type *)(ent->lde_name + len);
4587                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4588         }
4589
4590         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4591 }
4592
4593 /**
4594  * build lu direct from backend fs dirent.
4595  */
4596
4597 static inline void
4598 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4599                    char *name, __u16 namelen, __u16 type, __u32 attr)
4600 {
4601         ent->lde_attrs = attr | LUDA_FID;
4602         fid_cpu_to_le(&ent->lde_fid, fid);
4603
4604         ent->lde_hash = cpu_to_le64(offset);
4605         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4606
4607         strncpy(ent->lde_name, name, namelen);
4608         ent->lde_name[namelen] = '\0';
4609         ent->lde_namelen = cpu_to_le16(namelen);
4610
4611         /* append lustre attributes */
4612         osd_it_append_attrs(ent, namelen, type);
4613 }
4614
4615 /**
4616  * Return pointer to the record under iterator.
4617  */
4618 static int osd_it_iam_rec(const struct lu_env *env,
4619                           const struct dt_it *di,
4620                           struct dt_rec *dtrec, __u32 attr)
4621 {
4622         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4623         struct osd_thread_info *info = osd_oti_get(env);
4624         ENTRY;
4625
4626         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4627                 const struct osd_fid_pack *rec;
4628                 struct lu_fid             *fid = &info->oti_fid;
4629                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4630                 char                      *name;
4631                 int                        namelen;
4632                 __u64                      hash;
4633                 int                        rc;
4634
4635                 name = (char *)iam_it_key_get(&it->oi_it);
4636                 if (IS_ERR(name))
4637                         RETURN(PTR_ERR(name));
4638
4639                 namelen = iam_it_key_size(&it->oi_it);
4640
4641                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4642                 if (IS_ERR(rec))
4643                         RETURN(PTR_ERR(rec));
4644
4645                 rc = osd_fid_unpack(fid, rec);
4646                 if (rc)
4647                         RETURN(rc);
4648
4649                 hash = iam_it_store(&it->oi_it);
4650
4651                 /* IAM does not store object type in IAM index (dir) */
4652                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4653                                    0, LUDA_FID);
4654         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4655                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4656                            (struct iam_rec *)dtrec);
4657                 osd_quota_unpack(it->oi_obj, dtrec);
4658         } else {
4659                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4660                            (struct iam_rec *)dtrec);
4661         }
4662
4663         RETURN(0);
4664 }
4665
4666 /**
4667  * Returns cookie for current Iterator position.
4668  */
4669 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4670 {
4671         struct osd_it_iam *it = (struct osd_it_iam *)di;
4672
4673         return iam_it_store(&it->oi_it);
4674 }
4675
4676 /**
4677  * Restore iterator from cookie.
4678  *
4679  * \param  di      osd iterator
4680  * \param  hash    Iterator location cookie
4681  *
4682  * \retval +ve  di points to record with least key not larger than key.
4683  * \retval  0   di points to exact matched key
4684  * \retval -ve  failure
4685  */
4686
4687 static int osd_it_iam_load(const struct lu_env *env,
4688                            const struct dt_it *di, __u64 hash)
4689 {
4690         struct osd_it_iam *it = (struct osd_it_iam *)di;
4691
4692         return iam_it_load(&it->oi_it, hash);
4693 }
4694
4695 static const struct dt_index_operations osd_index_iam_ops = {
4696         .dio_lookup         = osd_index_iam_lookup,
4697         .dio_declare_insert = osd_index_declare_iam_insert,
4698         .dio_insert         = osd_index_iam_insert,
4699         .dio_declare_delete = osd_index_declare_iam_delete,
4700         .dio_delete         = osd_index_iam_delete,
4701         .dio_it     = {
4702                 .init     = osd_it_iam_init,
4703                 .fini     = osd_it_iam_fini,
4704                 .get      = osd_it_iam_get,
4705                 .put      = osd_it_iam_put,
4706                 .next     = osd_it_iam_next,
4707                 .key      = osd_it_iam_key,
4708                 .key_size = osd_it_iam_key_size,
4709                 .rec      = osd_it_iam_rec,
4710                 .store    = osd_it_iam_store,
4711                 .load     = osd_it_iam_load
4712         }
4713 };
4714
4715
4716 /**
4717  * Creates or initializes iterator context.
4718  *
4719  * \retval struct osd_it_ea, iterator structure on success
4720  *
4721  */
4722 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4723                                     struct dt_object *dt,
4724                                     __u32 attr,
4725                                     struct lustre_capa *capa)
4726 {
4727         struct osd_object       *obj  = osd_dt_obj(dt);
4728         struct osd_thread_info  *info = osd_oti_get(env);
4729         struct osd_it_ea        *it   = &info->oti_it_ea;
4730         struct file             *file = &it->oie_file;
4731         struct lu_object        *lo   = &dt->do_lu;
4732         struct dentry           *obj_dentry = &info->oti_it_dentry;
4733         ENTRY;
4734         LASSERT(lu_object_exists(lo));
4735
4736         obj_dentry->d_inode = obj->oo_inode;
4737         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4738         obj_dentry->d_name.hash = 0;
4739
4740         it->oie_rd_dirent       = 0;
4741         it->oie_it_dirent       = 0;
4742         it->oie_dirent          = NULL;
4743         it->oie_buf             = info->oti_it_ea_buf;
4744         it->oie_obj             = obj;
4745
4746         /* Reset the "file" totally to avoid to reuse any old value from
4747          * former readdir handling, the "file->f_pos" should be zero. */
4748         memset(file, 0, sizeof(*file));
4749         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4750         if (attr & LUDA_64BITHASH)
4751                 file->f_mode    = FMODE_64BITHASH;
4752         else
4753                 file->f_mode    = FMODE_32BITHASH;
4754         file->f_dentry          = obj_dentry;
4755         file->f_mapping         = obj->oo_inode->i_mapping;
4756         file->f_op              = obj->oo_inode->i_fop;
4757         set_file_inode(file, obj->oo_inode);
4758
4759         lu_object_get(lo);
4760         RETURN((struct dt_it *) it);
4761 }
4762
4763 /**
4764  * Destroy or finishes iterator context.
4765  *
4766  * \param di iterator structure to be destroyed
4767  */
4768 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4769 {
4770         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4771         struct osd_object    *obj  = it->oie_obj;
4772         struct inode       *inode  = obj->oo_inode;
4773
4774         ENTRY;
4775         it->oie_file.f_op->release(inode, &it->oie_file);
4776         lu_object_put(env, &obj->oo_dt.do_lu);
4777         EXIT;
4778 }
4779
4780 /**
4781  * It position the iterator at given key, so that next lookup continues from
4782  * that key Or it is similar to dio_it->load() but based on a key,
4783  * rather than file position.
4784  *
4785  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4786  * to the beginning.
4787  *
4788  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4789  */
4790 static int osd_it_ea_get(const struct lu_env *env,
4791                          struct dt_it *di, const struct dt_key *key)
4792 {
4793         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4794
4795         ENTRY;
4796         LASSERT(((const char *)key)[0] == '\0');
4797         it->oie_file.f_pos      = 0;
4798         it->oie_rd_dirent       = 0;
4799         it->oie_it_dirent       = 0;
4800         it->oie_dirent          = NULL;
4801
4802         RETURN(+1);
4803 }
4804
4805 /**
4806  * Does nothing
4807  */
4808 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4809 {
4810 }
4811
4812 struct osd_filldir_cbs {
4813 #ifdef HAVE_DIR_CONTEXT
4814         struct dir_context ctx;
4815 #endif
4816         struct osd_it_ea  *it;
4817 };
4818 /**
4819  * It is called internally by ->readdir(). It fills the
4820  * iterator's in-memory data structure with required
4821  * information i.e. name, namelen, rec_size etc.
4822  *
4823  * \param buf in which information to be filled in.
4824  * \param name name of the file in given dir
4825  *
4826  * \retval 0 on success
4827  * \retval 1 on buffer full
4828  */
4829 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4830                                loff_t offset, __u64 ino,
4831                                unsigned d_type)
4832 {
4833         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4834         struct osd_object       *obj  = it->oie_obj;
4835         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4836         struct lu_fid           *fid  = &ent->oied_fid;
4837         struct osd_fid_pack     *rec;
4838         ENTRY;
4839
4840         /* this should never happen */
4841         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4842                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4843                 RETURN(-EIO);
4844         }
4845
4846         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4847             OSD_IT_EA_BUFSIZE)
4848                 RETURN(1);
4849
4850         /* "." is just the object itself. */
4851         if (namelen == 1 && name[0] == '.') {
4852                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4853         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4854                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4855                 if (osd_fid_unpack(fid, rec) != 0)
4856                         fid_zero(fid);
4857         } else {
4858                 fid_zero(fid);
4859         }
4860         d_type &= ~LDISKFS_DIRENT_LUFID;
4861
4862         /* NOT export local root. */
4863         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4864                 ino = obj->oo_inode->i_ino;
4865                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4866         }
4867
4868         ent->oied_ino     = ino;
4869         ent->oied_off     = offset;
4870         ent->oied_namelen = namelen;
4871         ent->oied_type    = d_type;
4872
4873         memcpy(ent->oied_name, name, namelen);
4874
4875         it->oie_rd_dirent++;
4876         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4877         RETURN(0);
4878 }
4879
4880 /**
4881  * Calls ->readdir() to load a directory entry at a time
4882  * and stored it in iterator's in-memory data structure.
4883  *
4884  * \param di iterator's in memory structure
4885  *
4886  * \retval   0 on success
4887  * \retval -ve on error
4888  * \retval +1 reach the end of entry
4889  */
4890 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4891                                const struct dt_it *di)
4892 {
4893         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4894         struct osd_object  *obj   = it->oie_obj;
4895         struct inode       *inode = obj->oo_inode;
4896         struct htree_lock  *hlock = NULL;
4897         struct file        *filp  = &it->oie_file;
4898         int                 rc = 0;
4899         struct osd_filldir_cbs buf = {
4900 #ifdef HAVE_DIR_CONTEXT
4901                 .ctx.actor = osd_ldiskfs_filldir,
4902 #endif
4903                 .it = it
4904         };
4905
4906         ENTRY;
4907         it->oie_dirent = it->oie_buf;
4908         it->oie_rd_dirent = 0;
4909
4910         if (obj->oo_hl_head != NULL) {
4911                 hlock = osd_oti_get(env)->oti_hlock;
4912                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4913                                    inode, LDISKFS_HLOCK_READDIR);
4914         } else {
4915                 down_read(&obj->oo_ext_idx_sem);
4916         }
4917
4918 #ifdef HAVE_DIR_CONTEXT
4919         buf.ctx.pos = filp->f_pos;
4920         rc = inode->i_fop->iterate(filp, &buf.ctx);
4921         filp->f_pos = buf.ctx.pos;
4922 #else
4923         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4924 #endif
4925
4926         if (hlock != NULL)
4927                 ldiskfs_htree_unlock(hlock);
4928         else
4929                 up_read(&obj->oo_ext_idx_sem);
4930
4931         if (it->oie_rd_dirent == 0) {
4932                 /*If it does not get any dirent, it means it has been reached
4933                  *to the end of the dir */
4934                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4935                 if (rc == 0)
4936                         rc = 1;
4937         } else {
4938                 it->oie_dirent = it->oie_buf;
4939                 it->oie_it_dirent = 1;
4940         }
4941
4942         RETURN(rc);
4943 }
4944
4945 /**
4946  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4947  * to load a directory entry at a time and stored it in
4948  * iterator's in-memory data structure.
4949  *
4950  * \param di iterator's in memory structure
4951  *
4952  * \retval +ve iterator reached to end
4953  * \retval   0 iterator not reached to end
4954  * \retval -ve on error
4955  */
4956 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4957 {
4958         struct osd_it_ea *it = (struct osd_it_ea *)di;
4959         int rc;
4960
4961         ENTRY;
4962
4963         if (it->oie_it_dirent < it->oie_rd_dirent) {
4964                 it->oie_dirent =
4965                         (void *) it->oie_dirent +
4966                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4967                                        it->oie_dirent->oied_namelen);
4968                 it->oie_it_dirent++;
4969                 RETURN(0);
4970         } else {
4971                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4972                         rc = +1;
4973                 else
4974                         rc = osd_ldiskfs_it_fill(env, di);
4975         }
4976
4977         RETURN(rc);
4978 }
4979
4980 /**
4981  * Returns the key at current position from iterator's in memory structure.
4982  *
4983  * \param di iterator's in memory structure
4984  *
4985  * \retval key i.e. struct dt_key on success
4986  */
4987 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4988                                     const struct dt_it *di)
4989 {
4990         struct osd_it_ea *it = (struct osd_it_ea *)di;
4991
4992         return (struct dt_key *)it->oie_dirent->oied_name;
4993 }
4994
4995 /**
4996  * Returns the key's size at current position from iterator's in memory structure.
4997  *
4998  * \param di iterator's in memory structure
4999  *
5000  * \retval key_size i.e. struct dt_key on success
5001  */
5002 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5003 {
5004         struct osd_it_ea *it = (struct osd_it_ea *)di;
5005
5006         return it->oie_dirent->oied_namelen;
5007 }
5008
5009 static int
5010 osd_dirent_update(handle_t *jh, struct super_block *sb,
5011                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5012                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5013 {
5014         struct osd_fid_pack *rec;
5015         int                  rc;
5016         ENTRY;
5017
5018         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5019         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5020
5021         rc = ldiskfs_journal_get_write_access(jh, bh);
5022         if (rc != 0)
5023                 RETURN(rc);
5024
5025         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5026         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5027         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5028
5029         RETURN(rc);
5030 }
5031
5032 static inline int
5033 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5034 {
5035         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5036             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5037                 return 1;
5038         else
5039                 return 0;
5040 }
5041
5042 static inline int
5043 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5044 {
5045         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5046                  "dot_dotdot = %d\n", dot_dotdot);
5047
5048         if (LDISKFS_DIR_REC_LEN(de) >=
5049             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5050                 return 1;
5051         else
5052                 return 0;
5053 }
5054
5055 static int
5056 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5057                     struct inode *dir, struct inode *inode,
5058                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5059                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5060                     struct htree_lock *hlock)
5061 {
5062         struct dentry               *dentry;
5063         struct osd_fid_pack         *rec;
5064         struct ldiskfs_dentry_param *ldp;
5065         int                          rc;
5066         ENTRY;
5067
5068         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5069                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5070                 RETURN(0);
5071
5072         /* There is enough space to hold the FID-in-dirent. */
5073         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5074                                  dir->i_sb->s_blocksize)) {
5075                 rc = ldiskfs_journal_get_write_access(jh, bh);
5076                 if (rc != 0)
5077                         RETURN(rc);
5078
5079                 de->name[de->name_len] = 0;
5080                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5081                 rec->fp_len = sizeof(struct lu_fid) + 1;
5082                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5083                 de->file_type |= LDISKFS_DIRENT_LUFID;
5084
5085                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5086
5087                 RETURN(rc);
5088         }
5089
5090         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5091         if (rc != 0)
5092                 RETURN(rc);
5093
5094         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5095                                            ent->oied_namelen);
5096         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5097         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5098         dentry->d_fsdata = (void *)ldp;
5099         ll_vfs_dq_init(dir);
5100         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5101         /* It is too bad, we cannot reinsert the name entry back.
5102          * That means we lose it! */
5103         if (rc != 0)
5104                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5105                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5106                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5107                        dir->i_ino, dir->i_generation,
5108                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5109
5110         RETURN(rc);
5111 }
5112
5113 static int
5114 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5115                         struct osd_it_ea *it, struct lu_fid *fid,
5116                         struct osd_inode_id *id, __u32 *attr)
5117 {
5118         struct osd_thread_info     *info        = osd_oti_get(env);
5119         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5120         struct osd_device          *dev         = osd_obj2dev(obj);
5121         struct super_block         *sb          = osd_sb(dev);
5122         const char                 *devname     =
5123                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5124         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5125         struct inode               *dir         = obj->oo_inode;
5126         struct htree_lock          *hlock       = NULL;
5127         struct buffer_head         *bh          = NULL;
5128         handle_t                   *jh          = NULL;
5129         struct ldiskfs_dir_entry_2 *de;
5130         struct dentry              *dentry;
5131         struct inode               *inode;
5132         int                         credits;
5133         int                         rc;
5134         int                         dot_dotdot  = 0;
5135         bool                        dirty       = false;
5136         ENTRY;
5137
5138         if (ent->oied_name[0] == '.') {
5139                 if (ent->oied_namelen == 1)
5140                         dot_dotdot = 1;
5141                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5142                         dot_dotdot = 2;
5143         }
5144
5145         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5146                                       ent->oied_namelen);
5147
5148         /* We need to ensure that the name entry is still valid.
5149          * Because it may be removed or renamed by other already.
5150          *
5151          * The unlink or rename operation will start journal before PDO lock,
5152          * so to avoid deadlock, here we need to start journal handle before
5153          * related PDO lock also. But because we do not know whether there
5154          * will be something to be repaired before PDO lock, we just start
5155          * journal without conditions.
5156          *
5157          * We may need to remove the name entry firstly, then insert back.
5158          * One credit is for user quota file update.
5159          * One credit is for group quota file update.
5160          * Two credits are for dirty inode. */
5161         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5162                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5163
5164 again:
5165         if (dev->od_dirent_journal) {
5166                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5167                 if (IS_ERR(jh)) {
5168                         rc = PTR_ERR(jh);
5169                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5170                                "check_repair, dir = %lu/%u, credits = %d, "
5171                                "name = %.*s: rc = %d\n",
5172                                devname, dir->i_ino, dir->i_generation, credits,
5173                                ent->oied_namelen, ent->oied_name, rc);
5174                         RETURN(rc);
5175                 }
5176
5177                 if (obj->oo_hl_head != NULL) {
5178                         hlock = osd_oti_get(env)->oti_hlock;
5179                         /* "0" means exclusive lock for the whole directory.
5180                          * We need to prevent others access such name entry
5181                          * during the delete + insert. Neither HLOCK_ADD nor
5182                          * HLOCK_DEL cannot guarantee the atomicity. */
5183                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5184                 } else {
5185                         down_write(&obj->oo_ext_idx_sem);
5186                 }
5187         } else {
5188                 if (obj->oo_hl_head != NULL) {
5189                         hlock = osd_oti_get(env)->oti_hlock;
5190                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5191                                            LDISKFS_HLOCK_LOOKUP);
5192                 } else {
5193                         down_read(&obj->oo_ext_idx_sem);
5194                 }
5195         }
5196
5197         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5198         /* For dot/dotdot entry, if there is not enough space to hold the
5199          * FID-in-dirent, just keep them there. It only happens when the
5200          * device upgraded from 1.8 or restored from MDT file-level backup.
5201          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5202          * and needs to get FID from LMA when readdir, it will not affect the
5203          * performance much. */
5204         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5205             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5206                 *attr |= LUDA_IGNORE;
5207                 GOTO(out_journal, rc = 0);
5208         }
5209
5210         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5211         inode = osd_iget(info, dev, id);
5212         if (IS_ERR(inode)) {
5213                 rc = PTR_ERR(inode);
5214                 if (rc == -ENOENT || rc == -ESTALE) {
5215                         *attr |= LUDA_IGNORE;
5216                         rc = 0;
5217                 } else {
5218                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5219                                "check_repair, dir = %lu/%u, name = %.*s: "
5220                                "rc = %d\n",
5221                                devname, dir->i_ino, dir->i_generation,
5222                                ent->oied_namelen, ent->oied_name, rc);
5223                 }
5224
5225                 GOTO(out_journal, rc);
5226         }
5227
5228         /* skip the REMOTE_PARENT_DIR. */
5229         if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5230                 GOTO(out_inode, rc = 0);
5231
5232         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5233         if (rc == 0) {
5234                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5235
5236                 if (fid_is_sane(fid)) {
5237                         /* FID-in-dirent is valid. */
5238                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5239                                 GOTO(out_inode, rc = 0);
5240
5241                         /* Do not repair under dryrun mode. */
5242                         if (*attr & LUDA_VERIFY_DRYRUN) {
5243                                 *attr |= LUDA_REPAIR;
5244                                 GOTO(out_inode, rc = 0);
5245                         }
5246
5247                         if (!dev->od_dirent_journal) {
5248                                 iput(inode);
5249                                 brelse(bh);
5250                                 if (hlock != NULL)
5251                                         ldiskfs_htree_unlock(hlock);
5252                                 else
5253                                         up_read(&obj->oo_ext_idx_sem);
5254                                 dev->od_dirent_journal = 1;
5255                                 goto again;
5256                         }
5257
5258                         *fid = lma->lma_self_fid;
5259                         dirty = true;
5260                         /* Update the FID-in-dirent. */
5261                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5262                         if (rc == 0)
5263                                 *attr |= LUDA_REPAIR;
5264                         else
5265                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5266                                        "in the dirent, dir = %lu/%u, "
5267                                        "name = %.*s, "DFID": rc = %d\n",
5268                                        devname, dir->i_ino, dir->i_generation,
5269                                        ent->oied_namelen, ent->oied_name,
5270                                        PFID(fid), rc);
5271                 } else {
5272                         /* Do not repair under dryrun mode. */
5273                         if (*attr & LUDA_VERIFY_DRYRUN) {
5274                                 *fid = lma->lma_self_fid;
5275                                 *attr |= LUDA_REPAIR;
5276                                 GOTO(out_inode, rc = 0);
5277                         }
5278
5279                         if (!dev->od_dirent_journal) {
5280                                 iput(inode);
5281                                 brelse(bh);
5282                                 if (hlock != NULL)
5283                                         ldiskfs_htree_unlock(hlock);
5284                                 else
5285                                         up_read(&obj->oo_ext_idx_sem);
5286                                 dev->od_dirent_journal = 1;
5287                                 goto again;
5288                         }
5289
5290                         *fid = lma->lma_self_fid;
5291                         dirty = true;
5292                         /* Append the FID-in-dirent. */
5293                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5294                                                  fid, bh, de, hlock);
5295                         if (rc == 0)
5296                                 *attr |= LUDA_REPAIR;
5297                         else
5298                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5299                                        "after the dirent, dir = %lu/%u, "
5300                                        "name = %.*s, "DFID": rc = %d\n",
5301                                        devname, dir->i_ino, dir->i_generation,
5302                                        ent->oied_namelen, ent->oied_name,
5303                                        PFID(fid), rc);
5304                 }
5305         } else if (rc == -ENODATA) {
5306                 /* Do not repair under dryrun mode. */
5307                 if (*attr & LUDA_VERIFY_DRYRUN) {
5308                         if (fid_is_sane(fid)) {
5309                                 *attr |= LUDA_REPAIR;
5310                         } else {
5311                                 lu_igif_build(fid, inode->i_ino,
5312                                               inode->i_generation);
5313                                 *attr |= LUDA_UPGRADE;
5314                         }
5315                         GOTO(out_inode, rc = 0);
5316                 }
5317
5318                 if (!dev->od_dirent_journal) {
5319                         iput(inode);
5320                         brelse(bh);
5321                         if (hlock != NULL)
5322                                 ldiskfs_htree_unlock(hlock);
5323                         else
5324                                 up_read(&obj->oo_ext_idx_sem);
5325                         dev->od_dirent_journal = 1;
5326                         goto again;
5327                 }
5328
5329                 dirty = true;
5330                 if (unlikely(fid_is_sane(fid))) {
5331                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5332                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5333                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5334                         if (rc == 0)
5335                                 *attr |= LUDA_REPAIR;
5336                         else
5337                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5338                                        "update dirent, dir = %lu/%u, "
5339                                        "name = %.*s, "DFID": rc = %d\n",
5340                                        devname, dir->i_ino, dir->i_generation,
5341                                        ent->oied_namelen, ent->oied_name,
5342                                        PFID(fid), rc);
5343                 } else {
5344                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5345                         /* It is probably IGIF object. Only aappend the
5346                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5347                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5348                                                  fid, bh, de, hlock);
5349                         if (rc == 0)
5350                                 *attr |= LUDA_UPGRADE;
5351                         else
5352                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5353                                        "after the dirent, dir = %lu/%u, "
5354                                        "name = %.*s, "DFID": rc = %d\n",
5355                                        devname, dir->i_ino, dir->i_generation,
5356                                        ent->oied_namelen, ent->oied_name,
5357                                        PFID(fid), rc);
5358                 }
5359         }
5360
5361         GOTO(out_inode, rc);
5362
5363 out_inode:
5364         iput(inode);
5365
5366 out_journal:
5367         brelse(bh);
5368         if (hlock != NULL) {
5369                 ldiskfs_htree_unlock(hlock);
5370         } else {
5371                 if (dev->od_dirent_journal)
5372                         up_write(&obj->oo_ext_idx_sem);
5373                 else
5374                         up_read(&obj->oo_ext_idx_sem);
5375         }
5376         if (jh != NULL)
5377                 ldiskfs_journal_stop(jh);
5378         if (rc >= 0 && !dirty)
5379                 dev->od_dirent_journal = 0;
5380         return rc;
5381 }
5382
5383 /**
5384  * Returns the value at current position from iterator's in memory structure.
5385  *
5386  * \param di struct osd_it_ea, iterator's in memory structure
5387  * \param attr attr requested for dirent.
5388  * \param lde lustre dirent
5389  *
5390  * \retval   0 no error and \param lde has correct lustre dirent.
5391  * \retval -ve on error
5392  */
5393 static inline int osd_it_ea_rec(const struct lu_env *env,
5394                                 const struct dt_it *di,
5395                                 struct dt_rec *dtrec, __u32 attr)
5396 {
5397         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5398         struct osd_object      *obj   = it->oie_obj;
5399         struct osd_device      *dev   = osd_obj2dev(obj);
5400         struct osd_scrub       *scrub = &dev->od_scrub;
5401         struct scrub_file      *sf    = &scrub->os_file;
5402         struct osd_thread_info *oti   = osd_oti_get(env);
5403         struct osd_inode_id    *id    = &oti->oti_id;
5404         struct osd_idmap_cache *oic   = &oti->oti_cache;
5405         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5406         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5407         __u32                   ino   = it->oie_dirent->oied_ino;
5408         int                     rc    = 0;
5409         ENTRY;
5410
5411         if (attr & LUDA_VERIFY) {
5412                 attr |= LUDA_TYPE;
5413                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5414                         attr |= LUDA_IGNORE;
5415                         rc = 0;
5416                 } else {
5417                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5418                                                      &attr);
5419                 }
5420         } else {
5421                 attr &= ~LU_DIRENT_ATTRS_MASK;
5422                 if (!fid_is_sane(fid)) {
5423                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5424                             likely(it->oie_dirent->oied_namelen != 2 ||
5425                                    it->oie_dirent->oied_name[0] != '.' ||
5426                                    it->oie_dirent->oied_name[1] != '.'))
5427                                 RETURN(-ENOENT);
5428
5429                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
5430                 } else {
5431                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5432                 }
5433         }
5434
5435         /* Pack the entry anyway, at least the offset is right. */
5436         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5437                            it->oie_dirent->oied_name,
5438                            it->oie_dirent->oied_namelen,
5439                            it->oie_dirent->oied_type, attr);
5440
5441         if (rc < 0)
5442                 RETURN(rc);
5443
5444         if (osd_remote_fid(env, dev, fid))
5445                 RETURN(0);
5446
5447         if (likely(!(attr & LUDA_IGNORE)))
5448                 rc = osd_add_oi_cache(oti, dev, id, fid);
5449
5450         if (!(attr & LUDA_VERIFY) &&
5451             (scrub->os_pos_current <= ino) &&
5452             ((sf->sf_flags & SF_INCONSISTENT) ||
5453              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5454              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5455                 osd_consistency_check(oti, dev, oic);
5456
5457         RETURN(rc);
5458 }
5459
5460 /**
5461  * Returns the record size size at current position.
5462  *
5463  * This function will return record(lu_dirent) size in bytes.
5464  *
5465  * \param[in] env       execution environment
5466  * \param[in] di        iterator's in memory structure
5467  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5468  *                      calculate the lu_dirent size.
5469  *
5470  * \retval      record size(in bytes & in memory) of the current lu_dirent
5471  *              entry.
5472  */
5473 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5474                               __u32 attr)
5475 {
5476         struct osd_it_ea *it = (struct osd_it_ea *)di;
5477
5478         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5479 }
5480
5481 /**
5482  * Returns a cookie for current position of the iterator head, so that
5483  * user can use this cookie to load/start the iterator next time.
5484  *
5485  * \param di iterator's in memory structure
5486  *
5487  * \retval cookie for current position, on success
5488  */
5489 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5490 {
5491         struct osd_it_ea *it = (struct osd_it_ea *)di;
5492
5493         return it->oie_dirent->oied_off;
5494 }
5495
5496 /**
5497  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5498  * to load a directory entry at a time and stored it i inn,
5499  * in iterator's in-memory data structure.
5500  *
5501  * \param di struct osd_it_ea, iterator's in memory structure
5502  *
5503  * \retval +ve on success
5504  * \retval -ve on error
5505  */
5506 static int osd_it_ea_load(const struct lu_env *env,
5507                           const struct dt_it *di, __u64 hash)
5508 {
5509         struct osd_it_ea *it = (struct osd_it_ea *)di;
5510         int rc;
5511
5512         ENTRY;
5513         it->oie_file.f_pos = hash;
5514
5515         rc =  osd_ldiskfs_it_fill(env, di);
5516         if (rc > 0)
5517                 rc = -ENODATA;
5518
5519         if (rc == 0)
5520                 rc = +1;
5521
5522         RETURN(rc);
5523 }
5524
5525 /**
5526  * Index lookup function for interoperability mode (b11826).
5527  *
5528  * \param key,  key i.e. file name to be searched
5529  *
5530  * \retval +ve, on success
5531  * \retval -ve, on error
5532  */
5533 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5534                                struct dt_rec *rec, const struct dt_key *key,
5535                                struct lustre_capa *capa)
5536 {
5537         struct osd_object *obj = osd_dt_obj(dt);
5538         int rc = 0;
5539
5540         ENTRY;
5541
5542         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5543         LINVRNT(osd_invariant(obj));
5544
5545         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5546                 return -EACCES;
5547
5548         rc = osd_ea_lookup_rec(env, obj, rec, key);
5549         if (rc == 0)
5550                 rc = +1;
5551         RETURN(rc);
5552 }
5553
5554 /**
5555  * Index and Iterator operations for interoperability
5556  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5557  */
5558 static const struct dt_index_operations osd_index_ea_ops = {
5559         .dio_lookup         = osd_index_ea_lookup,
5560         .dio_declare_insert = osd_index_declare_ea_insert,
5561         .dio_insert         = osd_index_ea_insert,
5562         .dio_declare_delete = osd_index_declare_ea_delete,
5563         .dio_delete         = osd_index_ea_delete,
5564         .dio_it     = {
5565                 .init     = osd_it_ea_init,
5566                 .fini     = osd_it_ea_fini,
5567                 .get      = osd_it_ea_get,
5568                 .put      = osd_it_ea_put,
5569                 .next     = osd_it_ea_next,
5570                 .key      = osd_it_ea_key,
5571                 .key_size = osd_it_ea_key_size,
5572                 .rec      = osd_it_ea_rec,
5573                 .rec_size = osd_it_ea_rec_size,
5574                 .store    = osd_it_ea_store,
5575                 .load     = osd_it_ea_load
5576         }
5577 };
5578
5579 static void *osd_key_init(const struct lu_context *ctx,
5580                           struct lu_context_key *key)
5581 {
5582         struct osd_thread_info *info;
5583
5584         OBD_ALLOC_PTR(info);
5585         if (info == NULL)
5586                 return ERR_PTR(-ENOMEM);
5587
5588         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5589         if (info->oti_it_ea_buf == NULL)
5590                 goto out_free_info;
5591
5592         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5593
5594         info->oti_hlock = ldiskfs_htree_lock_alloc();
5595         if (info->oti_hlock == NULL)
5596                 goto out_free_ea;
5597
5598         return info;
5599
5600  out_free_ea:
5601         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5602  out_free_info:
5603         OBD_FREE_PTR(info);
5604         return ERR_PTR(-ENOMEM);
5605 }
5606
5607 static void osd_key_fini(const struct lu_context *ctx,
5608                          struct lu_context_key *key, void* data)
5609 {
5610         struct osd_thread_info *info = data;
5611
5612         if (info->oti_hlock != NULL)
5613                 ldiskfs_htree_lock_free(info->oti_hlock);
5614         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5615         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5616         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5617         OBD_FREE_PTR(info);
5618 }
5619
5620 static void osd_key_exit(const struct lu_context *ctx,
5621                          struct lu_context_key *key, void *data)
5622 {
5623         struct osd_thread_info *info = data;
5624
5625         LASSERT(info->oti_r_locks == 0);
5626         LASSERT(info->oti_w_locks == 0);
5627         LASSERT(info->oti_txns    == 0);
5628 }
5629
5630 /* type constructor/destructor: osd_type_init, osd_type_fini */
5631 LU_TYPE_INIT_FINI(osd, &osd_key);
5632
5633 struct lu_context_key osd_key = {
5634         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5635         .lct_init = osd_key_init,
5636         .lct_fini = osd_key_fini,
5637         .lct_exit = osd_key_exit
5638 };
5639
5640
5641 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5642                            const char *name, struct lu_device *next)
5643 {
5644         struct osd_device *osd = osd_dev(d);
5645
5646         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5647             >= sizeof(osd->od_svname))
5648                 return -E2BIG;
5649         return osd_procfs_init(osd, name);
5650 }
5651
5652 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5653 {
5654         struct seq_server_site  *ss = osd_seq_site(osd);
5655         int                     rc;
5656         ENTRY;
5657
5658         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5659                 RETURN(0);
5660
5661         if (unlikely(ss == NULL))
5662                 RETURN(-ENODEV);
5663
5664         OBD_ALLOC_PTR(osd->od_cl_seq);
5665         if (osd->od_cl_seq == NULL)
5666                 RETURN(-ENOMEM);
5667
5668         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5669                              osd->od_svname, ss->ss_server_seq);
5670
5671         if (rc != 0) {
5672                 OBD_FREE_PTR(osd->od_cl_seq);
5673                 osd->od_cl_seq = NULL;
5674         }
5675
5676         RETURN(rc);
5677 }
5678
5679 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5680 {
5681         if (osd->od_cl_seq == NULL)
5682                 return;
5683
5684         seq_client_fini(osd->od_cl_seq);
5685         OBD_FREE_PTR(osd->od_cl_seq);
5686         osd->od_cl_seq = NULL;
5687 }
5688
5689 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5690 {
5691         ENTRY;
5692
5693         /* shutdown quota slave instance associated with the device */
5694         if (o->od_quota_slave != NULL) {
5695                 qsd_fini(env, o->od_quota_slave);
5696                 o->od_quota_slave = NULL;
5697         }
5698
5699         osd_fid_fini(env, o);
5700
5701         RETURN(0);
5702 }
5703
5704 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5705 {
5706         ENTRY;
5707
5708         if (o->od_mnt != NULL) {
5709                 shrink_dcache_sb(osd_sb(o));
5710                 osd_sync(env, &o->od_dt_dev);
5711
5712                 mntput(o->od_mnt);
5713                 o->od_mnt = NULL;
5714         }
5715
5716         EXIT;
5717 }
5718
5719 static int osd_mount(const struct lu_env *env,
5720                      struct osd_device *o, struct lustre_cfg *cfg)
5721 {
5722         const char              *name  = lustre_cfg_string(cfg, 0);
5723         const char              *dev  = lustre_cfg_string(cfg, 1);
5724         const char              *opts;
5725         unsigned long            page, s_flags, lmd_flags = 0;
5726         struct page             *__page;
5727         struct file_system_type *type;
5728         char                    *options = NULL;
5729         char                    *str;
5730         struct osd_thread_info  *info = osd_oti_get(env);
5731         struct lu_fid           *fid = &info->oti_fid;
5732         struct inode            *inode;
5733         int                      rc = 0;
5734         ENTRY;
5735
5736         if (o->od_mnt != NULL)
5737                 RETURN(0);
5738
5739         if (strlen(dev) >= sizeof(o->od_mntdev))
5740                 RETURN(-E2BIG);
5741         strcpy(o->od_mntdev, dev);
5742
5743         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5744         if (__page == NULL)
5745                 GOTO(out, rc = -ENOMEM);
5746
5747         str = lustre_cfg_string(cfg, 2);
5748         s_flags = simple_strtoul(str, NULL, 0);
5749         str = strstr(str, ":");
5750         if (str)
5751                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5752         opts = lustre_cfg_string(cfg, 3);
5753         page = (unsigned long)page_address(__page);
5754         options = (char *)page;
5755         *options = '\0';
5756         if (opts == NULL)
5757                 strcat(options, "user_xattr,acl");
5758         else
5759                 strcat(options, opts);
5760
5761         /* Glom up mount options */
5762         if (*options != '\0')
5763                 strcat(options, ",");
5764         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5765
5766         type = get_fs_type("ldiskfs");
5767         if (!type) {
5768                 CERROR("%s: cannot find ldiskfs module\n", name);
5769                 GOTO(out, rc = -ENODEV);
5770         }
5771
5772         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5773         module_put(type->owner);
5774
5775         if (IS_ERR(o->od_mnt)) {
5776                 rc = PTR_ERR(o->od_mnt);
5777                 o->od_mnt = NULL;
5778                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5779                 GOTO(out, rc);
5780         }
5781
5782 #ifdef HAVE_DEV_SET_RDONLY
5783         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5784                 CERROR("%s: underlying device %s is marked as read-only. "
5785                        "Setup failed\n", name, dev);
5786                 GOTO(out_mnt, rc = -EROFS);
5787         }
5788 #endif
5789
5790         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5791                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5792                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5793                 GOTO(out_mnt, rc = -EINVAL);
5794         }
5795
5796 #ifdef LDISKFS_MOUNT_DIRDATA
5797         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5798                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5799                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5800         else if (!o->od_is_ost)
5801                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5802                       "enabling the dirdata feature. If you do not want to "
5803                       "downgrade to Lustre-1.x again, you can enable it via "
5804                       "'tune2fs -O dirdata device'\n", name, dev);
5805 #endif
5806         inode = osd_sb(o)->s_root->d_inode;
5807         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5808         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5809         if (rc != 0) {
5810                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5811                 GOTO(out_mnt, rc);
5812         }
5813
5814         if (lmd_flags & LMD_FLG_NOSCRUB)
5815                 o->od_noscrub = 1;
5816
5817         GOTO(out, rc = 0);
5818
5819 out_mnt:
5820         mntput(o->od_mnt);
5821         o->od_mnt = NULL;
5822
5823 out:
5824         if (__page)
5825                 OBD_PAGE_FREE(__page);
5826
5827         return rc;
5828 }
5829
5830 static struct lu_device *osd_device_fini(const struct lu_env *env,
5831                                          struct lu_device *d)
5832 {
5833         struct osd_device *o = osd_dev(d);
5834         ENTRY;
5835
5836         osd_shutdown(env, o);
5837         osd_procfs_fini(o);
5838         osd_scrub_cleanup(env, o);
5839         osd_obj_map_fini(o);
5840         osd_umount(env, o);
5841
5842         RETURN(NULL);
5843 }
5844
5845 static int osd_device_init0(const struct lu_env *env,
5846                             struct osd_device *o,
5847                             struct lustre_cfg *cfg)
5848 {
5849         struct lu_device        *l = osd2lu_dev(o);
5850         struct osd_thread_info *info;
5851         int                     rc;
5852         int                     cplen = 0;
5853
5854         /* if the module was re-loaded, env can loose its keys */
5855         rc = lu_env_refill((struct lu_env *) env);
5856         if (rc)
5857                 GOTO(out, rc);
5858         info = osd_oti_get(env);
5859         LASSERT(info);
5860
5861         l->ld_ops = &osd_lu_ops;
5862         o->od_dt_dev.dd_ops = &osd_dt_ops;
5863
5864         spin_lock_init(&o->od_osfs_lock);
5865         mutex_init(&o->od_otable_mutex);
5866
5867         o->od_capa_hash = init_capa_hash();
5868         if (o->od_capa_hash == NULL)
5869                 GOTO(out, rc = -ENOMEM);
5870
5871         o->od_read_cache = 1;
5872         o->od_writethrough_cache = 1;
5873         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5874
5875         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5876                         sizeof(o->od_svname));
5877         if (cplen >= sizeof(o->od_svname)) {
5878                 rc = -E2BIG;
5879                 GOTO(out_capa, rc);
5880         }
5881
5882         if (server_name_is_ost(o->od_svname))
5883                 o->od_is_ost = 1;
5884
5885         rc = osd_mount(env, o, cfg);
5886         if (rc != 0)
5887                 GOTO(out_capa, rc);
5888
5889         rc = osd_obj_map_init(env, o);
5890         if (rc != 0)
5891                 GOTO(out_mnt, rc);
5892
5893         rc = lu_site_init(&o->od_site, l);
5894         if (rc != 0)
5895                 GOTO(out_compat, rc);
5896         o->od_site.ls_bottom_dev = l;
5897
5898         rc = lu_site_init_finish(&o->od_site);
5899         if (rc != 0)
5900                 GOTO(out_site, rc);
5901
5902         /* self-repair LMA by default */
5903         o->od_lma_self_repair = 1;
5904
5905         INIT_LIST_HEAD(&o->od_ios_list);
5906         /* setup scrub, including OI files initialization */
5907         rc = osd_scrub_setup(env, o);
5908         if (rc < 0)
5909                 GOTO(out_site, rc);
5910
5911         rc = osd_procfs_init(o, o->od_svname);
5912         if (rc != 0) {
5913                 CERROR("%s: can't initialize procfs: rc = %d\n",
5914                        o->od_svname, rc);
5915                 GOTO(out_scrub, rc);
5916         }
5917
5918         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5919
5920         /* initialize quota slave instance */
5921         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5922                                      o->od_proc_entry);
5923         if (IS_ERR(o->od_quota_slave)) {
5924                 rc = PTR_ERR(o->od_quota_slave);
5925                 o->od_quota_slave = NULL;
5926                 GOTO(out_procfs, rc);
5927         }
5928
5929         RETURN(0);
5930
5931 out_procfs:
5932         osd_procfs_fini(o);
5933 out_scrub:
5934         osd_scrub_cleanup(env, o);
5935 out_site:
5936         lu_site_fini(&o->od_site);
5937 out_compat:
5938         osd_obj_map_fini(o);
5939 out_mnt:
5940         osd_umount(env, o);
5941 out_capa:
5942         cleanup_capa_hash(o->od_capa_hash);
5943 out:
5944         return rc;
5945 }
5946
5947 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5948                                           struct lu_device_type *t,
5949                                           struct lustre_cfg *cfg)
5950 {
5951         struct osd_device *o;
5952         int                rc;
5953
5954         OBD_ALLOC_PTR(o);
5955         if (o == NULL)
5956                 return ERR_PTR(-ENOMEM);
5957
5958         rc = dt_device_init(&o->od_dt_dev, t);
5959         if (rc == 0) {
5960                 /* Because the ctx might be revived in dt_device_init,
5961                  * refill the env here */
5962                 lu_env_refill((struct lu_env *)env);
5963                 rc = osd_device_init0(env, o, cfg);
5964                 if (rc)
5965                         dt_device_fini(&o->od_dt_dev);
5966         }
5967
5968         if (unlikely(rc != 0))
5969                 OBD_FREE_PTR(o);
5970
5971         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5972 }
5973
5974 static struct lu_device *osd_device_free(const struct lu_env *env,
5975                                          struct lu_device *d)
5976 {
5977         struct osd_device *o = osd_dev(d);
5978         ENTRY;
5979
5980         cleanup_capa_hash(o->od_capa_hash);
5981         /* XXX: make osd top device in order to release reference */
5982         d->ld_site->ls_top_dev = d;
5983         lu_site_purge(env, d->ld_site, -1);
5984         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5985                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5986                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5987         }
5988         lu_site_fini(&o->od_site);
5989         dt_device_fini(&o->od_dt_dev);
5990         OBD_FREE_PTR(o);
5991         RETURN(NULL);
5992 }
5993
5994 static int osd_process_config(const struct lu_env *env,
5995                               struct lu_device *d, struct lustre_cfg *cfg)
5996 {
5997         struct osd_device               *o = osd_dev(d);
5998         int                             rc;
5999         ENTRY;
6000
6001         switch (cfg->lcfg_command) {
6002         case LCFG_SETUP:
6003                 rc = osd_mount(env, o, cfg);
6004                 break;
6005         case LCFG_CLEANUP:
6006                 lu_dev_del_linkage(d->ld_site, d);
6007                 rc = osd_shutdown(env, o);
6008                 break;
6009         case LCFG_PARAM:
6010                 LASSERT(&o->od_dt_dev);
6011                 rc = class_process_proc_seq_param(PARAM_OSD,
6012                                                   lprocfs_osd_obd_vars,
6013                                                   cfg, &o->od_dt_dev);
6014                 if (rc > 0 || rc == -ENOSYS)
6015                         rc = class_process_proc_seq_param(PARAM_OST,
6016                                                           lprocfs_osd_obd_vars,
6017                                                           cfg, &o->od_dt_dev);
6018                 break;
6019         default:
6020                 rc = -ENOSYS;
6021         }
6022
6023         RETURN(rc);
6024 }
6025
6026 static int osd_recovery_complete(const struct lu_env *env,
6027                                  struct lu_device *d)
6028 {
6029         struct osd_device       *osd = osd_dev(d);
6030         int                      rc = 0;
6031         ENTRY;
6032
6033         if (osd->od_quota_slave == NULL)
6034                 RETURN(0);
6035
6036         /* start qsd instance on recovery completion, this notifies the quota
6037          * slave code that we are about to process new requests now */
6038         rc = qsd_start(env, osd->od_quota_slave);
6039         RETURN(rc);
6040 }
6041
6042 /*
6043  * we use exports to track all osd users
6044  */
6045 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6046                            struct obd_device *obd, struct obd_uuid *cluuid,
6047                            struct obd_connect_data *data, void *localdata)
6048 {
6049         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6050         struct lustre_handle  conn;
6051         int                   rc;
6052         ENTRY;
6053
6054         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6055
6056         rc = class_connect(&conn, obd, cluuid);
6057         if (rc)
6058                 RETURN(rc);
6059
6060         *exp = class_conn2export(&conn);
6061
6062         spin_lock(&osd->od_osfs_lock);
6063         osd->od_connects++;
6064         spin_unlock(&osd->od_osfs_lock);
6065
6066         RETURN(0);
6067 }
6068
6069 /*
6070  * once last export (we don't count self-export) disappeared
6071  * osd can be released
6072  */
6073 static int osd_obd_disconnect(struct obd_export *exp)
6074 {
6075         struct obd_device *obd = exp->exp_obd;
6076         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6077         int                rc, release = 0;
6078         ENTRY;
6079
6080         /* Only disconnect the underlying layers on the final disconnect. */
6081         spin_lock(&osd->od_osfs_lock);
6082         osd->od_connects--;
6083         if (osd->od_connects == 0)
6084                 release = 1;
6085         spin_unlock(&osd->od_osfs_lock);
6086
6087         rc = class_disconnect(exp); /* bz 9811 */
6088
6089         if (rc == 0 && release)
6090                 class_manual_cleanup(obd);
6091         RETURN(rc);
6092 }
6093
6094 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6095                        struct lu_device *dev)
6096 {
6097         struct osd_device *osd = osd_dev(dev);
6098         int                result = 0;
6099         ENTRY;
6100
6101         if (osd->od_quota_slave != NULL) {
6102                 /* set up quota slave objects */
6103                 result = qsd_prepare(env, osd->od_quota_slave);
6104                 if (result != 0)
6105                         RETURN(result);
6106         }
6107
6108         result = osd_fid_init(env, osd);
6109
6110         RETURN(result);
6111 }
6112
6113 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6114                   struct lu_fid *fid, struct md_op_data *op_data)
6115 {
6116         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6117
6118         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6119 }
6120
6121 static const struct lu_object_operations osd_lu_obj_ops = {
6122         .loo_object_init      = osd_object_init,
6123         .loo_object_delete    = osd_object_delete,
6124         .loo_object_release   = osd_object_release,
6125         .loo_object_free      = osd_object_free,
6126         .loo_object_print     = osd_object_print,
6127         .loo_object_invariant = osd_object_invariant
6128 };
6129
6130 const struct lu_device_operations osd_lu_ops = {
6131         .ldo_object_alloc      = osd_object_alloc,
6132         .ldo_process_config    = osd_process_config,
6133         .ldo_recovery_complete = osd_recovery_complete,
6134         .ldo_prepare           = osd_prepare,
6135 };
6136
6137 static const struct lu_device_type_operations osd_device_type_ops = {
6138         .ldto_init = osd_type_init,
6139         .ldto_fini = osd_type_fini,
6140
6141         .ldto_start = osd_type_start,
6142         .ldto_stop  = osd_type_stop,
6143
6144         .ldto_device_alloc = osd_device_alloc,
6145         .ldto_device_free  = osd_device_free,
6146
6147         .ldto_device_init    = osd_device_init,
6148         .ldto_device_fini    = osd_device_fini
6149 };
6150
6151 struct lu_device_type osd_device_type = {
6152         .ldt_tags     = LU_DEVICE_DT,
6153         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6154         .ldt_ops      = &osd_device_type_ops,
6155         .ldt_ctx_tags = LCT_LOCAL,
6156 };
6157
6158 /*
6159  * lprocfs legacy support.
6160  */
6161 static struct obd_ops osd_obd_device_ops = {
6162         .o_owner = THIS_MODULE,
6163         .o_connect      = osd_obd_connect,
6164         .o_disconnect   = osd_obd_disconnect,
6165         .o_fid_alloc    = osd_fid_alloc,
6166 };
6167
6168 static int __init osd_mod_init(void)
6169 {
6170         int rc;
6171
6172         osd_oi_mod_init();
6173
6174         rc = lu_kmem_init(ldiskfs_caches);
6175         if (rc)
6176                 return rc;
6177
6178         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6179                                  lprocfs_osd_module_vars,
6180 #ifndef HAVE_ONLY_PROCFS_SEQ
6181                                  NULL,
6182 #endif
6183                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6184         if (rc)
6185                 lu_kmem_fini(ldiskfs_caches);
6186         return rc;
6187 }
6188
6189 static void __exit osd_mod_exit(void)
6190 {
6191         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6192         lu_kmem_fini(ldiskfs_caches);
6193 }
6194
6195 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6196 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6197 MODULE_LICENSE("GPL");
6198
6199 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);