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_MDS
  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 (!cfs_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         cfs_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                 cfs_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                 CFS_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         if (ss == NULL)
1079                 RETURN(1);
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         if (!rc)
1155                 rc = iobuf->dr_error;
1156
1157         RETURN(rc);
1158 }
1159
1160 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1161 {
1162         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1163                                                ot_super);
1164
1165         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1166         LASSERT(&dcb->dcb_func != NULL);
1167         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1168
1169         return 0;
1170 }
1171
1172 /*
1173  * Called just before object is freed. Releases all resources except for
1174  * object itself (that is released by osd_object_free()).
1175  *
1176  * Concurrency: no concurrent access is possible that late in object
1177  * life-cycle.
1178  */
1179 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1180 {
1181         struct osd_object *obj   = osd_obj(l);
1182         struct inode      *inode = obj->oo_inode;
1183
1184         LINVRNT(osd_invariant(obj));
1185
1186         /*
1187          * If object is unlinked remove fid->ino mapping from object index.
1188          */
1189
1190         osd_index_fini(obj);
1191         if (inode != NULL) {
1192                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1193                 qid_t                    uid = inode->i_uid;
1194                 qid_t                    gid = inode->i_gid;
1195
1196                 iput(inode);
1197                 obj->oo_inode = NULL;
1198
1199                 if (qsd != NULL) {
1200                         struct osd_thread_info  *info = osd_oti_get(env);
1201                         struct lquota_id_info   *qi = &info->oti_qi;
1202
1203                         /* Release granted quota to master if necessary */
1204                         qi->lqi_id.qid_uid = uid;
1205                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1206
1207                         qi->lqi_id.qid_uid = gid;
1208                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1209                 }
1210         }
1211 }
1212
1213 /*
1214  * Concurrency: ->loo_object_release() is called under site spin-lock.
1215  */
1216 static void osd_object_release(const struct lu_env *env,
1217                                struct lu_object *l)
1218 {
1219 }
1220
1221 /*
1222  * Concurrency: shouldn't matter.
1223  */
1224 static int osd_object_print(const struct lu_env *env, void *cookie,
1225                             lu_printer_t p, const struct lu_object *l)
1226 {
1227         struct osd_object *o = osd_obj(l);
1228         struct iam_descr  *d;
1229
1230         if (o->oo_dir != NULL)
1231                 d = o->oo_dir->od_container.ic_descr;
1232         else
1233                 d = NULL;
1234         return (*p)(env, cookie,
1235                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1236                     o, o->oo_inode,
1237                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1238                     o->oo_inode ? o->oo_inode->i_generation : 0,
1239                     d ? d->id_ops->id_name : "plain");
1240 }
1241
1242 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1243
1244 /*
1245  * Concurrency: shouldn't matter.
1246  */
1247 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1248                struct obd_statfs *sfs)
1249 {
1250         struct osd_device  *osd = osd_dt_dev(d);
1251         struct super_block *sb = osd_sb(osd);
1252         struct kstatfs     *ksfs;
1253         int result = 0;
1254
1255         if (unlikely(osd->od_mnt == NULL))
1256                 return -EINPROGRESS;
1257
1258         /* osd_lproc.c call this without env, allocate ksfs for that case */
1259         if (unlikely(env == NULL)) {
1260                 OBD_ALLOC_PTR(ksfs);
1261                 if (ksfs == NULL)
1262                         return -ENOMEM;
1263         } else {
1264                 ksfs = &osd_oti_get(env)->oti_ksfs;
1265         }
1266
1267         spin_lock(&osd->od_osfs_lock);
1268         result = sb->s_op->statfs(sb->s_root, ksfs);
1269         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1270                 statfs_pack(sfs, ksfs);
1271                 if (sb->s_flags & MS_RDONLY)
1272                         sfs->os_state = OS_STATE_READONLY;
1273         }
1274
1275         spin_unlock(&osd->od_osfs_lock);
1276
1277         if (unlikely(env == NULL))
1278                 OBD_FREE_PTR(ksfs);
1279
1280         /* Reserve a small amount of space for local objects like last_rcvd,
1281          * llog, quota files, ... */
1282         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1283                 sfs->os_bavail = 0;
1284         } else {
1285                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1286                 /** Take out metadata overhead for indirect blocks */
1287                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1288         }
1289
1290         return result;
1291 }
1292
1293 /**
1294  * Estimate space needed for file creations. We assume the largest filename
1295  * which is 2^64 - 1, hence a filename of 20 chars.
1296  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1297  */
1298 #ifdef __LDISKFS_DIR_REC_LEN
1299 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1300 #else
1301 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1302 #endif
1303
1304 /*
1305  * Concurrency: doesn't access mutable data.
1306  */
1307 static void osd_conf_get(const struct lu_env *env,
1308                          const struct dt_device *dev,
1309                          struct dt_device_param *param)
1310 {
1311         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1312         int                ea_overhead;
1313
1314         /*
1315          * XXX should be taken from not-yet-existing fs abstraction layer.
1316          */
1317         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1318         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1319         param->ddp_block_shift  = sb->s_blocksize_bits;
1320         param->ddp_mount_type     = LDD_MT_LDISKFS;
1321         param->ddp_maxbytes       = sb->s_maxbytes;
1322         /* Overhead estimate should be fairly accurate, so we really take a tiny
1323          * error margin which also avoids fragmenting the filesystem too much */
1324         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1325         /* inode are statically allocated, so per-inode space consumption
1326          * is the space consumed by the directory entry */
1327         param->ddp_inodespace     = PER_OBJ_USAGE;
1328         /* per-fragment overhead to be used by the client code */
1329         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1330         param->ddp_mntopts      = 0;
1331         if (test_opt(sb, XATTR_USER))
1332                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1333         if (test_opt(sb, POSIX_ACL))
1334                 param->ddp_mntopts |= MNTOPT_ACL;
1335
1336         /* LOD might calculate the max stripe count based on max_ea_size,
1337          * so we need take account in the overhead as well,
1338          * xattr_header + magic + xattr_entry_head */
1339         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1340                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1341
1342 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1343         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1344                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1345                                                                 ea_overhead;
1346         else
1347 #endif
1348                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1349 }
1350
1351 /*
1352  * Concurrency: shouldn't matter.
1353  */
1354 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1355 {
1356         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1357         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1358 }
1359
1360 /**
1361  * Start commit for OSD device.
1362  *
1363  * An implementation of dt_commit_async method for OSD device.
1364  * Asychronously starts underlayng fs sync and thereby a transaction
1365  * commit.
1366  *
1367  * \param env environment
1368  * \param d dt device
1369  *
1370  * \see dt_device_operations
1371  */
1372 static int osd_commit_async(const struct lu_env *env,
1373                             struct dt_device *d)
1374 {
1375         struct super_block *s = osd_sb(osd_dt_dev(d));
1376         ENTRY;
1377
1378         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1379         RETURN(s->s_op->sync_fs(s, 0));
1380 }
1381
1382 /*
1383  * Concurrency: shouldn't matter.
1384  */
1385
1386 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1387 {
1388         struct super_block *sb = osd_sb(osd_dt_dev(d));
1389         struct block_device *dev = sb->s_bdev;
1390 #ifdef HAVE_DEV_SET_RDONLY
1391         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1392         int rc = 0;
1393 #else
1394         int rc = -EOPNOTSUPP;
1395 #endif
1396         ENTRY;
1397
1398 #ifdef HAVE_DEV_SET_RDONLY
1399         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1400
1401         if (jdev && (jdev != dev)) {
1402                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1403                        (long)jdev);
1404                 dev_set_rdonly(jdev);
1405         }
1406         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1407         dev_set_rdonly(dev);
1408 #else
1409         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1410                osd_dt_dev(d)->od_svname, (long)dev, rc);
1411 #endif
1412         RETURN(rc);
1413 }
1414
1415 /*
1416  * Concurrency: serialization provided by callers.
1417  */
1418 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1419                               int mode, unsigned long timeout, __u32 alg,
1420                               struct lustre_capa_key *keys)
1421 {
1422         struct osd_device *dev = osd_dt_dev(d);
1423         ENTRY;
1424
1425         dev->od_fl_capa = mode;
1426         dev->od_capa_timeout = timeout;
1427         dev->od_capa_alg = alg;
1428         dev->od_capa_keys = keys;
1429         RETURN(0);
1430 }
1431
1432 /**
1433  * Note: we do not count into QUOTA here.
1434  * If we mount with --data_journal we may need more.
1435  */
1436 const int osd_dto_credits_noquota[DTO_NR] = {
1437         /**
1438          * Insert.
1439          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1440          * SINGLEDATA_TRANS_BLOCKS(8)
1441          * XXX Note: maybe iam need more, since iam have more level than
1442          *           EXT3 htree.
1443          */
1444         [DTO_INDEX_INSERT]  = 16,
1445         /**
1446          * Delete
1447          * just modify a single entry, probably merge few within a block
1448          */
1449         [DTO_INDEX_DELETE]  = 1,
1450         /**
1451          * Used for OI scrub
1452          */
1453         [DTO_INDEX_UPDATE]  = 16,
1454         /**
1455          * 4(inode, inode bits, groups, GDT)
1456          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1457          */
1458         [DTO_OBJECT_CREATE] = 4,
1459         /**
1460          * 4(inode, inode bits, groups, GDT)
1461          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1462          */
1463         [DTO_OBJECT_DELETE] = 4,
1464         /**
1465          * Attr set credits (inode)
1466          */
1467         [DTO_ATTR_SET_BASE] = 1,
1468         /**
1469          * Xattr set. The same as xattr of EXT3.
1470          * DATA_TRANS_BLOCKS(14)
1471          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1472          * are also counted in. Do not know why?
1473          */
1474         [DTO_XATTR_SET]     = 14,
1475         /**
1476          * credits for inode change during write.
1477          */
1478         [DTO_WRITE_BASE]    = 3,
1479         /**
1480          * credits for single block write.
1481          */
1482         [DTO_WRITE_BLOCK]   = 14,
1483         /**
1484          * Attr set credits for chown.
1485          * This is extra credits for setattr, and it is null without quota
1486          */
1487         [DTO_ATTR_SET_CHOWN] = 0
1488 };
1489
1490 static const struct dt_device_operations osd_dt_ops = {
1491         .dt_root_get       = osd_root_get,
1492         .dt_statfs         = osd_statfs,
1493         .dt_trans_create   = osd_trans_create,
1494         .dt_trans_start    = osd_trans_start,
1495         .dt_trans_stop     = osd_trans_stop,
1496         .dt_trans_cb_add   = osd_trans_cb_add,
1497         .dt_conf_get       = osd_conf_get,
1498         .dt_sync           = osd_sync,
1499         .dt_ro             = osd_ro,
1500         .dt_commit_async   = osd_commit_async,
1501         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1502 };
1503
1504 static void osd_object_read_lock(const struct lu_env *env,
1505                                  struct dt_object *dt, unsigned role)
1506 {
1507         struct osd_object *obj = osd_dt_obj(dt);
1508         struct osd_thread_info *oti = osd_oti_get(env);
1509
1510         LINVRNT(osd_invariant(obj));
1511
1512         LASSERT(obj->oo_owner != env);
1513         down_read_nested(&obj->oo_sem, role);
1514
1515         LASSERT(obj->oo_owner == NULL);
1516         oti->oti_r_locks++;
1517 }
1518
1519 static void osd_object_write_lock(const struct lu_env *env,
1520                                   struct dt_object *dt, unsigned role)
1521 {
1522         struct osd_object *obj = osd_dt_obj(dt);
1523         struct osd_thread_info *oti = osd_oti_get(env);
1524
1525         LINVRNT(osd_invariant(obj));
1526
1527         LASSERT(obj->oo_owner != env);
1528         down_write_nested(&obj->oo_sem, role);
1529
1530         LASSERT(obj->oo_owner == NULL);
1531         obj->oo_owner = env;
1532         oti->oti_w_locks++;
1533 }
1534
1535 static void osd_object_read_unlock(const struct lu_env *env,
1536                                    struct dt_object *dt)
1537 {
1538         struct osd_object *obj = osd_dt_obj(dt);
1539         struct osd_thread_info *oti = osd_oti_get(env);
1540
1541         LINVRNT(osd_invariant(obj));
1542
1543         LASSERT(oti->oti_r_locks > 0);
1544         oti->oti_r_locks--;
1545         up_read(&obj->oo_sem);
1546 }
1547
1548 static void osd_object_write_unlock(const struct lu_env *env,
1549                                     struct dt_object *dt)
1550 {
1551         struct osd_object *obj = osd_dt_obj(dt);
1552         struct osd_thread_info *oti = osd_oti_get(env);
1553
1554         LINVRNT(osd_invariant(obj));
1555
1556         LASSERT(obj->oo_owner == env);
1557         LASSERT(oti->oti_w_locks > 0);
1558         oti->oti_w_locks--;
1559         obj->oo_owner = NULL;
1560         up_write(&obj->oo_sem);
1561 }
1562
1563 static int osd_object_write_locked(const struct lu_env *env,
1564                                    struct dt_object *dt)
1565 {
1566         struct osd_object *obj = osd_dt_obj(dt);
1567
1568         LINVRNT(osd_invariant(obj));
1569
1570         return obj->oo_owner == env;
1571 }
1572
1573 static int capa_is_sane(const struct lu_env *env,
1574                         struct osd_device *dev,
1575                         struct lustre_capa *capa,
1576                         struct lustre_capa_key *keys)
1577 {
1578         struct osd_thread_info *oti = osd_oti_get(env);
1579         struct lustre_capa *tcapa = &oti->oti_capa;
1580         struct obd_capa *oc;
1581         int i, rc = 0;
1582         ENTRY;
1583
1584         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1585         if (oc) {
1586                 if (capa_is_expired(oc)) {
1587                         DEBUG_CAPA(D_ERROR, capa, "expired");
1588                         rc = -ESTALE;
1589                 }
1590                 capa_put(oc);
1591                 RETURN(rc);
1592         }
1593
1594         if (capa_is_expired_sec(capa)) {
1595                 DEBUG_CAPA(D_ERROR, capa, "expired");
1596                 RETURN(-ESTALE);
1597         }
1598
1599         spin_lock(&capa_lock);
1600         for (i = 0; i < 2; i++) {
1601                 if (keys[i].lk_keyid == capa->lc_keyid) {
1602                         oti->oti_capa_key = keys[i];
1603                         break;
1604                 }
1605         }
1606         spin_unlock(&capa_lock);
1607
1608         if (i == 2) {
1609                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1610                 RETURN(-ESTALE);
1611         }
1612
1613         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1614         if (rc)
1615                 RETURN(rc);
1616
1617         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1618                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1619                 RETURN(-EACCES);
1620         }
1621
1622         oc = capa_add(dev->od_capa_hash, capa);
1623         capa_put(oc);
1624
1625         RETURN(0);
1626 }
1627
1628 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1629                     struct lustre_capa *capa, __u64 opc)
1630 {
1631         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1632         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1633         struct lu_capainfo *lci;
1634         int rc;
1635
1636         if (!osd->od_fl_capa)
1637                 return 0;
1638
1639         if (capa == BYPASS_CAPA)
1640                 return 0;
1641
1642         lci = lu_capainfo_get(env);
1643         if (unlikely(lci == NULL))
1644                 return 0;
1645
1646         if (lci->lci_auth == LC_ID_NONE)
1647                 return 0;
1648
1649         if (capa == NULL) {
1650                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1651                        osd_name(osd), PFID(fid), -EACCES);
1652                 return -EACCES;
1653         }
1654
1655         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1656                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1657                            PFID(fid));
1658                 return -EACCES;
1659         }
1660
1661         if (!capa_opc_supported(capa, opc)) {
1662                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1663                 return -EACCES;
1664         }
1665
1666         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1667         if (rc != 0) {
1668                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1669                 return -EACCES;
1670         }
1671
1672         return 0;
1673 }
1674
1675 static struct timespec *osd_inode_time(const struct lu_env *env,
1676                                        struct inode *inode, __u64 seconds)
1677 {
1678         struct osd_thread_info  *oti = osd_oti_get(env);
1679         struct timespec         *t   = &oti->oti_time;
1680
1681         t->tv_sec = seconds;
1682         t->tv_nsec = 0;
1683         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1684         return t;
1685 }
1686
1687
1688 static void osd_inode_getattr(const struct lu_env *env,
1689                               struct inode *inode, struct lu_attr *attr)
1690 {
1691         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1692                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1693                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1694                                LA_TYPE;
1695
1696         attr->la_atime      = LTIME_S(inode->i_atime);
1697         attr->la_mtime      = LTIME_S(inode->i_mtime);
1698         attr->la_ctime      = LTIME_S(inode->i_ctime);
1699         attr->la_mode       = inode->i_mode;
1700         attr->la_size       = i_size_read(inode);
1701         attr->la_blocks     = inode->i_blocks;
1702         attr->la_uid        = inode->i_uid;
1703         attr->la_gid        = inode->i_gid;
1704         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1705         attr->la_nlink      = inode->i_nlink;
1706         attr->la_rdev       = inode->i_rdev;
1707         attr->la_blksize    = 1 << inode->i_blkbits;
1708         attr->la_blkbits    = inode->i_blkbits;
1709 }
1710
1711 static int osd_attr_get(const struct lu_env *env,
1712                         struct dt_object *dt,
1713                         struct lu_attr *attr,
1714                         struct lustre_capa *capa)
1715 {
1716         struct osd_object *obj = osd_dt_obj(dt);
1717
1718         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1719         LINVRNT(osd_invariant(obj));
1720
1721         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1722                 return -EACCES;
1723
1724         spin_lock(&obj->oo_guard);
1725         osd_inode_getattr(env, obj->oo_inode, attr);
1726         spin_unlock(&obj->oo_guard);
1727         return 0;
1728 }
1729
1730 static int osd_declare_attr_set(const struct lu_env *env,
1731                                 struct dt_object *dt,
1732                                 const struct lu_attr *attr,
1733                                 struct thandle *handle)
1734 {
1735         struct osd_thandle     *oh;
1736         struct osd_object      *obj;
1737         struct osd_thread_info *info = osd_oti_get(env);
1738         struct lquota_id_info  *qi = &info->oti_qi;
1739         long long               bspace;
1740         int                     rc = 0;
1741         bool                    allocated;
1742         ENTRY;
1743
1744         LASSERT(dt != NULL);
1745         LASSERT(handle != NULL);
1746
1747         obj = osd_dt_obj(dt);
1748         LASSERT(osd_invariant(obj));
1749
1750         oh = container_of0(handle, struct osd_thandle, ot_super);
1751         LASSERT(oh->ot_handle == NULL);
1752
1753         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1754                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1755
1756         if (attr == NULL || obj->oo_inode == NULL)
1757                 RETURN(rc);
1758
1759         bspace   = obj->oo_inode->i_blocks;
1760         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1761         bspace   = toqb(bspace);
1762
1763         /* Changing ownership is always preformed by super user, it should not
1764          * fail with EDQUOT.
1765          *
1766          * We still need to call the osd_declare_qid() to calculate the journal
1767          * credits for updating quota accounting files and to trigger quota
1768          * space adjustment once the operation is completed.*/
1769         if ((attr->la_valid & LA_UID) != 0 &&
1770              attr->la_uid != obj->oo_inode->i_uid) {
1771                 qi->lqi_type = USRQUOTA;
1772
1773                 /* inode accounting */
1774                 qi->lqi_is_blk = false;
1775
1776                 /* one more inode for the new owner ... */
1777                 qi->lqi_id.qid_uid = attr->la_uid;
1778                 qi->lqi_space      = 1;
1779                 allocated = (attr->la_uid == 0) ? true : false;
1780                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1781                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1782                         rc = 0;
1783                 if (rc)
1784                         RETURN(rc);
1785
1786                 /* and one less inode for the current uid */
1787                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1788                 qi->lqi_space      = -1;
1789                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1790                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1791                         rc = 0;
1792                 if (rc)
1793                         RETURN(rc);
1794
1795                 /* block accounting */
1796                 qi->lqi_is_blk = true;
1797
1798                 /* more blocks for the new owner ... */
1799                 qi->lqi_id.qid_uid = attr->la_uid;
1800                 qi->lqi_space      = bspace;
1801                 allocated = (attr->la_uid == 0) ? true : false;
1802                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1803                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1804                         rc = 0;
1805                 if (rc)
1806                         RETURN(rc);
1807
1808                 /* and finally less blocks for the current owner */
1809                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1810                 qi->lqi_space      = -bspace;
1811                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1812                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1813                         rc = 0;
1814                 if (rc)
1815                         RETURN(rc);
1816         }
1817
1818         if (attr->la_valid & LA_GID &&
1819             attr->la_gid != obj->oo_inode->i_gid) {
1820                 qi->lqi_type = GRPQUOTA;
1821
1822                 /* inode accounting */
1823                 qi->lqi_is_blk = false;
1824
1825                 /* one more inode for the new group owner ... */
1826                 qi->lqi_id.qid_gid = attr->la_gid;
1827                 qi->lqi_space      = 1;
1828                 allocated = (attr->la_gid == 0) ? true : false;
1829                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1830                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1831                         rc = 0;
1832                 if (rc)
1833                         RETURN(rc);
1834
1835                 /* and one less inode for the current gid */
1836                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1837                 qi->lqi_space      = -1;
1838                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1839                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1840                         rc = 0;
1841                 if (rc)
1842                         RETURN(rc);
1843
1844                 /* block accounting */
1845                 qi->lqi_is_blk = true;
1846
1847                 /* more blocks for the new owner ... */
1848                 qi->lqi_id.qid_gid = attr->la_gid;
1849                 qi->lqi_space      = bspace;
1850                 allocated = (attr->la_gid == 0) ? true : false;
1851                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1852                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1853                         rc = 0;
1854                 if (rc)
1855                         RETURN(rc);
1856
1857                 /* and finally less blocks for the current owner */
1858                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1859                 qi->lqi_space      = -bspace;
1860                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1861                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1862                         rc = 0;
1863                 if (rc)
1864                         RETURN(rc);
1865         }
1866
1867         RETURN(rc);
1868 }
1869
1870 static int osd_inode_setattr(const struct lu_env *env,
1871                              struct inode *inode, const struct lu_attr *attr)
1872 {
1873         __u64 bits;
1874
1875         bits = attr->la_valid;
1876
1877         if (bits & LA_ATIME)
1878                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1879         if (bits & LA_CTIME)
1880                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1881         if (bits & LA_MTIME)
1882                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1883         if (bits & LA_SIZE) {
1884                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1885                 i_size_write(inode, attr->la_size);
1886         }
1887
1888 #if 0
1889         /* OSD should not change "i_blocks" which is used by quota.
1890          * "i_blocks" should be changed by ldiskfs only. */
1891         if (bits & LA_BLOCKS)
1892                 inode->i_blocks = attr->la_blocks;
1893 #endif
1894         if (bits & LA_MODE)
1895                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1896                         (attr->la_mode & ~S_IFMT);
1897         if (bits & LA_UID)
1898                 inode->i_uid    = attr->la_uid;
1899         if (bits & LA_GID)
1900                 inode->i_gid    = attr->la_gid;
1901         if (bits & LA_NLINK)
1902                 set_nlink(inode, attr->la_nlink);
1903         if (bits & LA_RDEV)
1904                 inode->i_rdev   = attr->la_rdev;
1905
1906         if (bits & LA_FLAGS) {
1907                 /* always keep S_NOCMTIME */
1908                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1909                                  S_NOCMTIME;
1910         }
1911         return 0;
1912 }
1913
1914 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1915 {
1916         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1917             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1918                 struct iattr    iattr;
1919                 int             rc;
1920
1921                 iattr.ia_valid = 0;
1922                 if (attr->la_valid & LA_UID)
1923                         iattr.ia_valid |= ATTR_UID;
1924                 if (attr->la_valid & LA_GID)
1925                         iattr.ia_valid |= ATTR_GID;
1926                 iattr.ia_uid = attr->la_uid;
1927                 iattr.ia_gid = attr->la_gid;
1928
1929                 rc = ll_vfs_dq_transfer(inode, &iattr);
1930                 if (rc) {
1931                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1932                                "enforcement enabled on the ldiskfs filesystem?",
1933                                inode->i_sb->s_id, rc);
1934                         return rc;
1935                 }
1936         }
1937         return 0;
1938 }
1939
1940 static int osd_attr_set(const struct lu_env *env,
1941                         struct dt_object *dt,
1942                         const struct lu_attr *attr,
1943                         struct thandle *handle,
1944                         struct lustre_capa *capa)
1945 {
1946         struct osd_object *obj = osd_dt_obj(dt);
1947         struct inode      *inode;
1948         int rc;
1949
1950         LASSERT(handle != NULL);
1951         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1952         LASSERT(osd_invariant(obj));
1953
1954         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1955                 return -EACCES;
1956
1957         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1958
1959         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1960                 struct osd_thread_info  *oti  = osd_oti_get(env);
1961                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1962                 struct lu_fid           *fid1 = &oti->oti_fid;
1963                 struct osd_inode_id     *id   = &oti->oti_id;
1964                 struct iam_path_descr   *ipd;
1965                 struct iam_container    *bag;
1966                 struct osd_thandle      *oh;
1967                 int                      rc;
1968
1969                 fid_cpu_to_be(fid1, fid0);
1970                 memset(id, 1, sizeof(*id));
1971                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1972                                   fid0)->oi_dir.od_container;
1973                 ipd = osd_idx_ipd_get(env, bag);
1974                 if (unlikely(ipd == NULL))
1975                         RETURN(-ENOMEM);
1976
1977                 oh = container_of0(handle, struct osd_thandle, ot_super);
1978                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1979                                 (const struct iam_rec *)id, ipd);
1980                 osd_ipd_put(env, bag, ipd);
1981                 return(rc > 0 ? 0 : rc);
1982         }
1983
1984         inode = obj->oo_inode;
1985         ll_vfs_dq_init(inode);
1986
1987         rc = osd_quota_transfer(inode, attr);
1988         if (rc)
1989                 return rc;
1990
1991         spin_lock(&obj->oo_guard);
1992         rc = osd_inode_setattr(env, inode, attr);
1993         spin_unlock(&obj->oo_guard);
1994
1995         if (!rc)
1996                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1997         return rc;
1998 }
1999
2000 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2001                                     struct osd_object *obj,
2002                                     const char *name, const int namelen)
2003 {
2004         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2005 }
2006
2007 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2008                       umode_t mode, struct dt_allocation_hint *hint,
2009                       struct thandle *th)
2010 {
2011         int result;
2012         struct osd_device  *osd = osd_obj2dev(obj);
2013         struct osd_thandle *oth;
2014         struct dt_object   *parent = NULL;
2015         struct inode       *inode;
2016
2017         LINVRNT(osd_invariant(obj));
2018         LASSERT(obj->oo_inode == NULL);
2019         LASSERT(obj->oo_hl_head == NULL);
2020
2021         if (S_ISDIR(mode) && ldiskfs_pdo) {
2022                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2023                 if (obj->oo_hl_head == NULL)
2024                         return -ENOMEM;
2025         }
2026
2027         oth = container_of(th, struct osd_thandle, ot_super);
2028         LASSERT(oth->ot_handle->h_transaction != NULL);
2029
2030         if (hint && hint->dah_parent)
2031                 parent = hint->dah_parent;
2032
2033         inode = ldiskfs_create_inode(oth->ot_handle,
2034                                      parent ? osd_dt_obj(parent)->oo_inode :
2035                                               osd_sb(osd)->s_root->d_inode,
2036                                      mode);
2037         if (!IS_ERR(inode)) {
2038                 /* Do not update file c/mtime in ldiskfs.
2039                  * NB: don't need any lock because no contention at this
2040                  * early stage */
2041                 inode->i_flags |= S_NOCMTIME;
2042
2043                 /* For new created object, it must be consistent,
2044                  * and it is unnecessary to scrub against it. */
2045                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2046                 obj->oo_inode = inode;
2047                 result = 0;
2048         } else {
2049                 if (obj->oo_hl_head != NULL) {
2050                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2051                         obj->oo_hl_head = NULL;
2052                 }
2053                 result = PTR_ERR(inode);
2054         }
2055         LINVRNT(osd_invariant(obj));
2056         return result;
2057 }
2058
2059 enum {
2060         OSD_NAME_LEN = 255
2061 };
2062
2063 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2064                      struct lu_attr *attr,
2065                      struct dt_allocation_hint *hint,
2066                      struct dt_object_format *dof,
2067                      struct thandle *th)
2068 {
2069         int result;
2070         struct osd_thandle *oth;
2071         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2072
2073         LASSERT(S_ISDIR(attr->la_mode));
2074
2075         oth = container_of(th, struct osd_thandle, ot_super);
2076         LASSERT(oth->ot_handle->h_transaction != NULL);
2077         result = osd_mkfile(info, obj, mode, hint, th);
2078
2079         return result;
2080 }
2081
2082 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2083                         struct lu_attr *attr,
2084                         struct dt_allocation_hint *hint,
2085                         struct dt_object_format *dof,
2086                         struct thandle *th)
2087 {
2088         int result;
2089         struct osd_thandle *oth;
2090         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2091
2092         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2093
2094         LASSERT(S_ISREG(attr->la_mode));
2095
2096         oth = container_of(th, struct osd_thandle, ot_super);
2097         LASSERT(oth->ot_handle->h_transaction != NULL);
2098
2099         result = osd_mkfile(info, obj, mode, hint, th);
2100         if (result == 0) {
2101                 LASSERT(obj->oo_inode != NULL);
2102                 if (feat->dif_flags & DT_IND_VARKEY)
2103                         result = iam_lvar_create(obj->oo_inode,
2104                                                  feat->dif_keysize_max,
2105                                                  feat->dif_ptrsize,
2106                                                  feat->dif_recsize_max,
2107                                                  oth->ot_handle);
2108                 else
2109                         result = iam_lfix_create(obj->oo_inode,
2110                                                  feat->dif_keysize_max,
2111                                                  feat->dif_ptrsize,
2112                                                  feat->dif_recsize_max,
2113                                                  oth->ot_handle);
2114
2115         }
2116         return result;
2117 }
2118
2119 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2120                      struct lu_attr *attr,
2121                      struct dt_allocation_hint *hint,
2122                      struct dt_object_format *dof,
2123                      struct thandle *th)
2124 {
2125         LASSERT(S_ISREG(attr->la_mode));
2126         return osd_mkfile(info, obj, (attr->la_mode &
2127                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2128 }
2129
2130 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2131                      struct lu_attr *attr,
2132                      struct dt_allocation_hint *hint,
2133                      struct dt_object_format *dof,
2134                      struct thandle *th)
2135 {
2136         LASSERT(S_ISLNK(attr->la_mode));
2137         return osd_mkfile(info, obj, (attr->la_mode &
2138                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2139 }
2140
2141 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2142                      struct lu_attr *attr,
2143                      struct dt_allocation_hint *hint,
2144                      struct dt_object_format *dof,
2145                      struct thandle *th)
2146 {
2147         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2148         int result;
2149
2150         LINVRNT(osd_invariant(obj));
2151         LASSERT(obj->oo_inode == NULL);
2152         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2153                 S_ISFIFO(mode) || S_ISSOCK(mode));
2154
2155         result = osd_mkfile(info, obj, mode, hint, th);
2156         if (result == 0) {
2157                 LASSERT(obj->oo_inode != NULL);
2158                 /*
2159                  * This inode should be marked dirty for i_rdev.  Currently
2160                  * that is done in the osd_attr_init().
2161                  */
2162                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2163                                    attr->la_rdev);
2164         }
2165         LINVRNT(osd_invariant(obj));
2166         return result;
2167 }
2168
2169 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2170                               struct lu_attr *,
2171                               struct dt_allocation_hint *hint,
2172                               struct dt_object_format *dof,
2173                               struct thandle *);
2174
2175 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2176 {
2177         osd_obj_type_f result;
2178
2179         switch (type) {
2180         case DFT_DIR:
2181                 result = osd_mkdir;
2182                 break;
2183         case DFT_REGULAR:
2184                 result = osd_mkreg;
2185                 break;
2186         case DFT_SYM:
2187                 result = osd_mksym;
2188                 break;
2189         case DFT_NODE:
2190                 result = osd_mknod;
2191                 break;
2192         case DFT_INDEX:
2193                 result = osd_mk_index;
2194                 break;
2195
2196         default:
2197                 LBUG();
2198                 break;
2199         }
2200         return result;
2201 }
2202
2203
2204 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2205                         struct dt_object *parent, struct dt_object *child,
2206                         umode_t child_mode)
2207 {
2208         LASSERT(ah);
2209
2210         ah->dah_parent = parent;
2211         ah->dah_mode = child_mode;
2212 }
2213
2214 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2215                           struct lu_attr *attr, struct dt_object_format *dof)
2216 {
2217         struct inode   *inode = obj->oo_inode;
2218         __u64           valid = attr->la_valid;
2219         int             result;
2220
2221         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2222
2223         if (dof->dof_type != DFT_NODE)
2224                 attr->la_valid &= ~LA_RDEV;
2225         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2226                 attr->la_valid &= ~LA_ATIME;
2227         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2228                 attr->la_valid &= ~LA_CTIME;
2229         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2230                 attr->la_valid &= ~LA_MTIME;
2231
2232         result = osd_quota_transfer(inode, attr);
2233         if (result)
2234                 return;
2235
2236         if (attr->la_valid != 0) {
2237                 result = osd_inode_setattr(info->oti_env, inode, attr);
2238                 /*
2239                  * The osd_inode_setattr() should always succeed here.  The
2240                  * only error that could be returned is EDQUOT when we are
2241                  * trying to change the UID or GID of the inode. However, this
2242                  * should not happen since quota enforcement is no longer
2243                  * enabled on ldiskfs (lquota takes care of it).
2244                  */
2245                 LASSERTF(result == 0, "%d", result);
2246                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2247         }
2248
2249         attr->la_valid = valid;
2250 }
2251
2252 /**
2253  * Helper function for osd_object_create()
2254  *
2255  * \retval 0, on success
2256  */
2257 static int __osd_object_create(struct osd_thread_info *info,
2258                                struct osd_object *obj, struct lu_attr *attr,
2259                                struct dt_allocation_hint *hint,
2260                                struct dt_object_format *dof,
2261                                struct thandle *th)
2262 {
2263         int     result;
2264         __u32   umask;
2265
2266         /* we drop umask so that permissions we pass are not affected */
2267         umask = current->fs->umask;
2268         current->fs->umask = 0;
2269
2270         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2271                                                   th);
2272         if (result == 0) {
2273                 osd_attr_init(info, obj, attr, dof);
2274                 osd_object_init0(obj);
2275                 /* bz 24037 */
2276                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2277                         unlock_new_inode(obj->oo_inode);
2278         }
2279
2280         /* restore previous umask value */
2281         current->fs->umask = umask;
2282
2283         return result;
2284 }
2285
2286 /**
2287  * Helper function for osd_object_create()
2288  *
2289  * \retval 0, on success
2290  */
2291 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2292                            const struct lu_fid *fid, struct thandle *th)
2293 {
2294         struct osd_thread_info *info = osd_oti_get(env);
2295         struct osd_inode_id    *id   = &info->oti_id;
2296         struct osd_device      *osd  = osd_obj2dev(obj);
2297         struct osd_thandle     *oh;
2298
2299         LASSERT(obj->oo_inode != NULL);
2300
2301         oh = container_of0(th, struct osd_thandle, ot_super);
2302         LASSERT(oh->ot_handle);
2303
2304         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2305         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2306 }
2307
2308 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2309                    obd_seq seq, struct lu_seq_range *range)
2310 {
2311         struct seq_server_site  *ss = osd_seq_site(osd);
2312
2313         if (fid_seq_is_idif(seq)) {
2314                 fld_range_set_ost(range);
2315                 range->lsr_index = idif_ost_idx(seq);
2316                 return 0;
2317         }
2318
2319         if (!fid_seq_in_fldb(seq)) {
2320                 fld_range_set_mdt(range);
2321                 if (ss != NULL)
2322                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2323                          * at all */
2324                         range->lsr_index = ss->ss_node_id;
2325                 return 0;
2326         }
2327
2328         LASSERT(ss != NULL);
2329         fld_range_set_any(range);
2330         /* OSD will only do local fld lookup */
2331         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2332 }
2333
2334 /*
2335  * Concurrency: no external locking is necessary.
2336  */
2337 static int osd_declare_object_create(const struct lu_env *env,
2338                                      struct dt_object *dt,
2339                                      struct lu_attr *attr,
2340                                      struct dt_allocation_hint *hint,
2341                                      struct dt_object_format *dof,
2342                                      struct thandle *handle)
2343 {
2344         struct osd_thandle      *oh;
2345         int                      rc;
2346         ENTRY;
2347
2348         LASSERT(handle != NULL);
2349
2350         oh = container_of0(handle, struct osd_thandle, ot_super);
2351         LASSERT(oh->ot_handle == NULL);
2352
2353         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2354                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2355         /* Reuse idle OI block may cause additional one OI block
2356          * to be changed. */
2357         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2358                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2359
2360         /* If this is directory, then we expect . and .. to be inserted as
2361          * well. The one directory block always needs to be created for the
2362          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2363          * block), there is no danger of needing a tree for the first block.
2364          */
2365         if (attr && S_ISDIR(attr->la_mode)) {
2366                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2367                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2368                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2369         }
2370
2371         if (!attr)
2372                 RETURN(0);
2373
2374         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2375                                    false, false, NULL, false);
2376         if (rc != 0)
2377                 RETURN(rc);
2378
2379         RETURN(rc);
2380 }
2381
2382 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2383                              struct lu_attr *attr,
2384                              struct dt_allocation_hint *hint,
2385                              struct dt_object_format *dof,
2386                              struct thandle *th)
2387 {
2388         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2389         struct osd_object      *obj    = osd_dt_obj(dt);
2390         struct osd_thread_info *info   = osd_oti_get(env);
2391         int result;
2392
2393         ENTRY;
2394
2395         LINVRNT(osd_invariant(obj));
2396         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2397         LASSERT(osd_write_locked(env, obj));
2398         LASSERT(th != NULL);
2399
2400         if (unlikely(fid_is_acct(fid)))
2401                 /* Quota files can't be created from the kernel any more,
2402                  * 'tune2fs -O quota' will take care of creating them */
2403                 RETURN(-EPERM);
2404
2405         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2406         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2407
2408         result = __osd_object_create(info, obj, attr, hint, dof, th);
2409         if (result == 0)
2410                 result = __osd_oi_insert(env, obj, fid, th);
2411
2412         LASSERT(ergo(result == 0,
2413                      dt_object_exists(dt) && !dt_object_remote(dt)));
2414
2415         LASSERT(osd_invariant(obj));
2416         RETURN(result);
2417 }
2418
2419 /**
2420  * Called to destroy on-disk representation of the object
2421  *
2422  * Concurrency: must be locked
2423  */
2424 static int osd_declare_object_destroy(const struct lu_env *env,
2425                                       struct dt_object *dt,
2426                                       struct thandle *th)
2427 {
2428         struct osd_object  *obj = osd_dt_obj(dt);
2429         struct inode       *inode = obj->oo_inode;
2430         struct osd_thandle *oh;
2431         int                 rc;
2432         ENTRY;
2433
2434         oh = container_of0(th, struct osd_thandle, ot_super);
2435         LASSERT(oh->ot_handle == NULL);
2436         LASSERT(inode);
2437
2438         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2439                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2440         /* Recycle idle OI leaf may cause additional three OI blocks
2441          * to be changed. */
2442         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2443                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2444         /* one less inode */
2445         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2446                                    false, true, NULL, false);
2447         if (rc)
2448                 RETURN(rc);
2449         /* data to be truncated */
2450         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2451                                    true, true, NULL, false);
2452         RETURN(rc);
2453 }
2454
2455 static int osd_object_destroy(const struct lu_env *env,
2456                               struct dt_object *dt,
2457                               struct thandle *th)
2458 {
2459         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2460         struct osd_object      *obj = osd_dt_obj(dt);
2461         struct inode           *inode = obj->oo_inode;
2462         struct osd_device      *osd = osd_obj2dev(obj);
2463         struct osd_thandle     *oh;
2464         int                     result;
2465         ENTRY;
2466
2467         oh = container_of0(th, struct osd_thandle, ot_super);
2468         LASSERT(oh->ot_handle);
2469         LASSERT(inode);
2470         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2471
2472         if (unlikely(fid_is_acct(fid)))
2473                 RETURN(-EPERM);
2474
2475         if (S_ISDIR(inode->i_mode)) {
2476                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2477                         inode->i_nlink == 2);
2478                 /* it will check/delete the inode from remote parent,
2479                  * how to optimize it? unlink performance impaction XXX */
2480                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2481                 if (result != 0 && result != -ENOENT) {
2482                         CERROR("%s: delete inode "DFID": rc = %d\n",
2483                                osd_name(osd), PFID(fid), result);
2484                 }
2485                 spin_lock(&obj->oo_guard);
2486                 clear_nlink(inode);
2487                 spin_unlock(&obj->oo_guard);
2488                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2489         }
2490
2491         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2492
2493         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2494                                OI_CHECK_FLD);
2495
2496         /* XXX: add to ext3 orphan list */
2497         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2498
2499         /* not needed in the cache anymore */
2500         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2501
2502         RETURN(0);
2503 }
2504
2505 /**
2506  * Put the fid into lustre_mdt_attrs, and then place the structure
2507  * inode's ea. This fid should not be altered during the life time
2508  * of the inode.
2509  *
2510  * \retval +ve, on success
2511  * \retval -ve, on error
2512  *
2513  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2514  */
2515 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2516                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2517 {
2518         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2519         int                      rc;
2520         ENTRY;
2521
2522         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2523                 RETURN(0);
2524
2525         lustre_lma_init(lma, fid, compat, incompat);
2526         lustre_lma_swab(lma);
2527
2528         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2529                              XATTR_CREATE);
2530         /* LMA may already exist, but we need to check that all the
2531          * desired compat/incompat flags have been added. */
2532         if (unlikely(rc == -EEXIST)) {
2533                 if (compat == 0 && incompat == 0)
2534                         RETURN(0);
2535
2536                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2537                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2538                                      LMA_OLD_SIZE);
2539                 if (rc <= 0)
2540                         RETURN(-EINVAL);
2541
2542                 lustre_lma_swab(lma);
2543                 if (!(~lma->lma_compat & compat) &&
2544                     !(~lma->lma_incompat & incompat))
2545                         RETURN(0);
2546
2547                 lma->lma_compat |= compat;
2548                 lma->lma_incompat |= incompat;
2549                 lustre_lma_swab(lma);
2550                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2551                                      sizeof(*lma), XATTR_REPLACE);
2552         }
2553
2554         RETURN(rc);
2555 }
2556
2557 /**
2558  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2559  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2560  * To have compatilibility with 1.8 ldiskfs driver we need to have
2561  * magic number at start of fid data.
2562  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2563  * its inmemory API.
2564  */
2565 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2566                                   const struct dt_rec *fid)
2567 {
2568         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2569             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2570                 param->edp_magic = 0;
2571                 return;
2572         }
2573
2574         param->edp_magic = LDISKFS_LUFID_MAGIC;
2575         param->edp_len =  sizeof(struct lu_fid) + 1;
2576         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2577 }
2578
2579 /**
2580  * Try to read the fid from inode ea into dt_rec.
2581  *
2582  * \param fid object fid.
2583  *
2584  * \retval 0 on success
2585  */
2586 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2587                           __u32 ino, struct lu_fid *fid,
2588                           struct osd_inode_id *id)
2589 {
2590         struct osd_thread_info *info  = osd_oti_get(env);
2591         struct inode           *inode;
2592         ENTRY;
2593
2594         osd_id_gen(id, ino, OSD_OII_NOGEN);
2595         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2596         if (IS_ERR(inode))
2597                 RETURN(PTR_ERR(inode));
2598
2599         iput(inode);
2600         RETURN(0);
2601 }
2602
2603 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2604                                         struct inode *dir,
2605                                         struct inode  *parent_dir,
2606                                         const struct dt_rec *dot_fid,
2607                                         const struct dt_rec *dot_dot_fid,
2608                                         struct osd_thandle *oth)
2609 {
2610         struct ldiskfs_dentry_param *dot_ldp;
2611         struct ldiskfs_dentry_param *dot_dot_ldp;
2612
2613         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2614         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2615
2616         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2617         dot_ldp->edp_magic = 0;
2618         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2619                                         dir, dot_ldp, dot_dot_ldp);
2620 }
2621
2622 /**
2623  * Create an local agent inode for remote entry
2624  */
2625 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2626                                                   struct osd_device *osd,
2627                                                   struct osd_object *pobj,
2628                                                   const struct lu_fid *fid,
2629                                                   struct thandle *th)
2630 {
2631         struct osd_thread_info  *info = osd_oti_get(env);
2632         struct inode            *local;
2633         struct osd_thandle      *oh;
2634         int                     rc;
2635         ENTRY;
2636
2637         LASSERT(th);
2638         oh = container_of(th, struct osd_thandle, ot_super);
2639         LASSERT(oh->ot_handle->h_transaction != NULL);
2640
2641         /* FIXME: Insert index api needs to know the mode of
2642          * the remote object. Just use S_IFDIR for now */
2643         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2644         if (IS_ERR(local)) {
2645                 CERROR("%s: create local error %d\n", osd_name(osd),
2646                        (int)PTR_ERR(local));
2647                 RETURN(local);
2648         }
2649
2650         /* Set special LMA flag for local agent inode */
2651         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2652         if (rc != 0) {
2653                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2654                        osd_name(osd), PFID(fid), rc);
2655                 RETURN(ERR_PTR(rc));
2656         }
2657
2658         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2659                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2660                 (const struct dt_rec *)fid, oh);
2661         if (rc != 0) {
2662                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2663                         osd_name(osd), PFID(fid), rc);
2664                 RETURN(ERR_PTR(rc));
2665         }
2666
2667         RETURN(local);
2668 }
2669
2670 /**
2671  * Delete local agent inode for remote entry
2672  */
2673 static int osd_delete_local_agent_inode(const struct lu_env *env,
2674                                         struct osd_device *osd,
2675                                         const struct lu_fid *fid,
2676                                         __u32 ino, struct osd_thandle *oh)
2677 {
2678         struct osd_thread_info  *oti = osd_oti_get(env);
2679         struct osd_inode_id     *id = &oti->oti_id;
2680         struct inode            *inode;
2681         ENTRY;
2682
2683         id->oii_ino = le32_to_cpu(ino);
2684         id->oii_gen = OSD_OII_NOGEN;
2685         inode = osd_iget(oti, osd, id);
2686         if (IS_ERR(inode)) {
2687                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2688                        PFID(fid), id->oii_ino, id->oii_gen);
2689                 RETURN(PTR_ERR(inode));
2690         }
2691
2692         clear_nlink(inode);
2693         mark_inode_dirty(inode);
2694         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2695                 osd_name(osd), PFID(fid), inode->i_ino);
2696         iput(inode);
2697         RETURN(0);
2698 }
2699
2700 /**
2701  * OSD layer object create function for interoperability mode (b11826).
2702  * This is mostly similar to osd_object_create(). Only difference being, fid is
2703  * inserted into inode ea here.
2704  *
2705  * \retval   0, on success
2706  * \retval -ve, on error
2707  */
2708 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2709                                 struct lu_attr *attr,
2710                                 struct dt_allocation_hint *hint,
2711                                 struct dt_object_format *dof,
2712                                 struct thandle *th)
2713 {
2714         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2715         struct osd_object      *obj    = osd_dt_obj(dt);
2716         struct osd_thread_info *info   = osd_oti_get(env);
2717         int                     result;
2718
2719         ENTRY;
2720
2721         LASSERT(osd_invariant(obj));
2722         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2723         LASSERT(osd_write_locked(env, obj));
2724         LASSERT(th != NULL);
2725
2726         if (unlikely(fid_is_acct(fid)))
2727                 /* Quota files can't be created from the kernel any more,
2728                  * 'tune2fs -O quota' will take care of creating them */
2729                 RETURN(-EPERM);
2730
2731         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2732         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2733
2734         result = __osd_object_create(info, obj, attr, hint, dof, th);
2735         if (result == 0)
2736                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2737                                 fid_is_on_ost(info, osd_obj2dev(obj),
2738                                               fid, OI_CHECK_FLD) ?
2739                                 LMAC_FID_ON_OST : 0, 0);
2740
2741         if (result == 0)
2742                 result = __osd_oi_insert(env, obj, fid, th);
2743
2744         LASSERT(ergo(result == 0,
2745                      dt_object_exists(dt) && !dt_object_remote(dt)));
2746         LINVRNT(osd_invariant(obj));
2747         RETURN(result);
2748 }
2749
2750 static int osd_declare_object_ref_add(const struct lu_env *env,
2751                                       struct dt_object *dt,
2752                                       struct thandle *handle)
2753 {
2754         struct osd_thandle       *oh;
2755
2756         /* it's possible that object doesn't exist yet */
2757         LASSERT(handle != NULL);
2758
2759         oh = container_of0(handle, struct osd_thandle, ot_super);
2760         LASSERT(oh->ot_handle == NULL);
2761
2762         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2763                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2764
2765         return 0;
2766 }
2767
2768 /*
2769  * Concurrency: @dt is write locked.
2770  */
2771 static int osd_object_ref_add(const struct lu_env *env,
2772                               struct dt_object *dt, struct thandle *th)
2773 {
2774         struct osd_object  *obj = osd_dt_obj(dt);
2775         struct inode       *inode = obj->oo_inode;
2776         struct osd_thandle *oh;
2777         int                 rc = 0;
2778
2779         LINVRNT(osd_invariant(obj));
2780         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2781         LASSERT(osd_write_locked(env, obj));
2782         LASSERT(th != NULL);
2783
2784         oh = container_of0(th, struct osd_thandle, ot_super);
2785         LASSERT(oh->ot_handle != NULL);
2786
2787         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2788
2789         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2790                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2791         /*
2792          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2793          * (65000) subdirectories by storing "1" in i_nlink if the link count
2794          * would otherwise overflow. Directory tranversal tools understand
2795          * that (st_nlink == 1) indicates that the filesystem dose not track
2796          * hard links count on the directory, and will not abort subdirectory
2797          * scanning early once (st_nlink - 2) subdirs have been found.
2798          *
2799          * This also has to properly handle the case of inodes with nlink == 0
2800          * in case they are being linked into the PENDING directory
2801          */
2802         spin_lock(&obj->oo_guard);
2803         ldiskfs_inc_count(oh->ot_handle, inode);
2804         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2805         spin_unlock(&obj->oo_guard);
2806
2807         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2808         LINVRNT(osd_invariant(obj));
2809
2810         return rc;
2811 }
2812
2813 static int osd_declare_object_ref_del(const struct lu_env *env,
2814                                       struct dt_object *dt,
2815                                       struct thandle *handle)
2816 {
2817         struct osd_thandle *oh;
2818
2819         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2820         LASSERT(handle != NULL);
2821
2822         oh = container_of0(handle, struct osd_thandle, ot_super);
2823         LASSERT(oh->ot_handle == NULL);
2824
2825         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2826                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2827
2828         return 0;
2829 }
2830
2831 /*
2832  * Concurrency: @dt is write locked.
2833  */
2834 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2835                               struct thandle *th)
2836 {
2837         struct osd_object       *obj = osd_dt_obj(dt);
2838         struct inode            *inode = obj->oo_inode;
2839         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2840         struct osd_thandle      *oh;
2841
2842         LINVRNT(osd_invariant(obj));
2843         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2844         LASSERT(osd_write_locked(env, obj));
2845         LASSERT(th != NULL);
2846
2847         oh = container_of0(th, struct osd_thandle, ot_super);
2848         LASSERT(oh->ot_handle != NULL);
2849
2850         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2851
2852         spin_lock(&obj->oo_guard);
2853         /* That can be result of upgrade from old Lustre version and
2854          * applied only to local files.  Just skip this ref_del call.
2855          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2856         if (inode->i_nlink == 0) {
2857                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2858                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2859                              ", maybe an upgraded file? (LU-3915)\n",
2860                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2861                 spin_unlock(&obj->oo_guard);
2862                 return 0;
2863         }
2864
2865         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2866                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2867
2868         ldiskfs_dec_count(oh->ot_handle, inode);
2869         spin_unlock(&obj->oo_guard);
2870
2871         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2872         LINVRNT(osd_invariant(obj));
2873
2874         return 0;
2875 }
2876
2877 /*
2878  * Get the 64-bit version for an inode.
2879  */
2880 static int osd_object_version_get(const struct lu_env *env,
2881                                   struct dt_object *dt, dt_obj_version_t *ver)
2882 {
2883         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2884
2885         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2886                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2887         *ver = LDISKFS_I(inode)->i_fs_version;
2888         return 0;
2889 }
2890
2891 /*
2892  * Concurrency: @dt is read locked.
2893  */
2894 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2895                          struct lu_buf *buf, const char *name,
2896                          struct lustre_capa *capa)
2897 {
2898         struct osd_object      *obj    = osd_dt_obj(dt);
2899         struct inode           *inode  = obj->oo_inode;
2900         struct osd_thread_info *info   = osd_oti_get(env);
2901         struct dentry          *dentry = &info->oti_obj_dentry;
2902
2903         /* version get is not real XATTR but uses xattr API */
2904         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2905                 /* for version we are just using xattr API but change inode
2906                  * field instead */
2907                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2908                 osd_object_version_get(env, dt, buf->lb_buf);
2909                 return sizeof(dt_obj_version_t);
2910         }
2911
2912         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2913         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2914
2915         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2916                 return -EACCES;
2917
2918         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2919 }
2920
2921
2922 static int osd_declare_xattr_set(const struct lu_env *env,
2923                                  struct dt_object *dt,
2924                                  const struct lu_buf *buf, const char *name,
2925                                  int fl, struct thandle *handle)
2926 {
2927         struct osd_thandle *oh;
2928         int credits;
2929
2930         LASSERT(handle != NULL);
2931
2932         oh = container_of0(handle, struct osd_thandle, ot_super);
2933         LASSERT(oh->ot_handle == NULL);
2934
2935         /* optimistic optimization: LMA is set first and usually fit inode */
2936         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2937                 if (dt_object_exists(dt))
2938                         credits = 0;
2939                 else
2940                         credits = 1;
2941         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2942                 credits = 1;
2943         } else {
2944                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2945                 struct super_block *sb = osd_sb(osd);
2946                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2947                 if (buf && buf->lb_len > sb->s_blocksize) {
2948                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2949                                         sb->s_blocksize_bits;
2950                 }
2951         }
2952
2953         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2954
2955         return 0;
2956 }
2957
2958 /*
2959  * Set the 64-bit version for object
2960  */
2961 static void osd_object_version_set(const struct lu_env *env,
2962                                    struct dt_object *dt,
2963                                    dt_obj_version_t *new_version)
2964 {
2965         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2966
2967         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2968                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2969
2970         LDISKFS_I(inode)->i_fs_version = *new_version;
2971         /** Version is set after all inode operations are finished,
2972          *  so we should mark it dirty here */
2973         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2974 }
2975
2976 /*
2977  * Concurrency: @dt is write locked.
2978  */
2979 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2980                          const struct lu_buf *buf, const char *name, int fl,
2981                          struct thandle *handle, struct lustre_capa *capa)
2982 {
2983         struct osd_object      *obj      = osd_dt_obj(dt);
2984         struct inode           *inode    = obj->oo_inode;
2985         struct osd_thread_info *info     = osd_oti_get(env);
2986         int                     fs_flags = 0;
2987         ENTRY;
2988
2989         LASSERT(handle != NULL);
2990
2991         /* version set is not real XATTR */
2992         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2993                 /* for version we are just using xattr API but change inode
2994                  * field instead */
2995                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2996                 osd_object_version_set(env, dt, buf->lb_buf);
2997                 return sizeof(dt_obj_version_t);
2998         }
2999
3000         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3001                 return -EACCES;
3002
3003         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3004                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3005
3006         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3007         if (fl & LU_XATTR_REPLACE)
3008                 fs_flags |= XATTR_REPLACE;
3009
3010         if (fl & LU_XATTR_CREATE)
3011                 fs_flags |= XATTR_CREATE;
3012
3013         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3014                                fs_flags);
3015 }
3016
3017 /*
3018  * Concurrency: @dt is read locked.
3019  */
3020 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3021                           struct lu_buf *buf, struct lustre_capa *capa)
3022 {
3023         struct osd_object      *obj    = osd_dt_obj(dt);
3024         struct inode           *inode  = obj->oo_inode;
3025         struct osd_thread_info *info   = osd_oti_get(env);
3026         struct dentry          *dentry = &info->oti_obj_dentry;
3027
3028         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3029         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3030
3031         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3032                 return -EACCES;
3033
3034         dentry->d_inode = inode;
3035         dentry->d_sb = inode->i_sb;
3036         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3037 }
3038
3039 static int osd_declare_xattr_del(const struct lu_env *env,
3040                                  struct dt_object *dt, const char *name,
3041                                  struct thandle *handle)
3042 {
3043         struct osd_thandle *oh;
3044
3045         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3046         LASSERT(handle != NULL);
3047
3048         oh = container_of0(handle, struct osd_thandle, ot_super);
3049         LASSERT(oh->ot_handle == NULL);
3050
3051         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3052                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3053
3054         return 0;
3055 }
3056
3057 /*
3058  * Concurrency: @dt is write locked.
3059  */
3060 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3061                          const char *name, struct thandle *handle,
3062                          struct lustre_capa *capa)
3063 {
3064         struct osd_object      *obj    = osd_dt_obj(dt);
3065         struct inode           *inode  = obj->oo_inode;
3066         struct osd_thread_info *info   = osd_oti_get(env);
3067         struct dentry          *dentry = &info->oti_obj_dentry;
3068         int                     rc;
3069
3070         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3071         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3072         LASSERT(handle != NULL);
3073
3074         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3075                 return -EACCES;
3076
3077         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3078
3079         ll_vfs_dq_init(inode);
3080         dentry->d_inode = inode;
3081         dentry->d_sb = inode->i_sb;
3082         rc = inode->i_op->removexattr(dentry, name);
3083         return rc;
3084 }
3085
3086 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3087                                      struct dt_object *dt,
3088                                      struct lustre_capa *old, __u64 opc)
3089 {
3090         struct osd_thread_info *info = osd_oti_get(env);
3091         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3092         struct osd_object *obj = osd_dt_obj(dt);
3093         struct osd_device *osd = osd_obj2dev(obj);
3094         struct lustre_capa_key *key = &info->oti_capa_key;
3095         struct lustre_capa *capa = &info->oti_capa;
3096         struct obd_capa *oc;
3097         struct lu_capainfo *lci;
3098         int rc;
3099         ENTRY;
3100
3101         if (!osd->od_fl_capa)
3102                 RETURN(ERR_PTR(-ENOENT));
3103
3104         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3105         LINVRNT(osd_invariant(obj));
3106
3107         /* renewal sanity check */
3108         if (old && osd_object_auth(env, dt, old, opc))
3109                 RETURN(ERR_PTR(-EACCES));
3110
3111         lci = lu_capainfo_get(env);
3112         if (unlikely(lci == NULL))
3113                 RETURN(ERR_PTR(-ENOENT));
3114
3115         switch (lci->lci_auth) {
3116         case LC_ID_NONE:
3117                 RETURN(NULL);
3118         case LC_ID_PLAIN:
3119                 capa->lc_uid = obj->oo_inode->i_uid;
3120                 capa->lc_gid = obj->oo_inode->i_gid;
3121                 capa->lc_flags = LC_ID_PLAIN;
3122                 break;
3123         case LC_ID_CONVERT: {
3124                 __u32 d[4], s[4];
3125
3126                 s[0] = obj->oo_inode->i_uid;
3127                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3128                 s[2] = obj->oo_inode->i_gid;
3129                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3130                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3131                 if (unlikely(rc))
3132                         RETURN(ERR_PTR(rc));
3133
3134                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3135                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3136                 capa->lc_flags = LC_ID_CONVERT;
3137                 break;
3138         }
3139         default:
3140                 RETURN(ERR_PTR(-EINVAL));
3141         }
3142
3143         capa->lc_fid = *fid;
3144         capa->lc_opc = opc;
3145         capa->lc_flags |= osd->od_capa_alg << 24;
3146         capa->lc_timeout = osd->od_capa_timeout;
3147         capa->lc_expiry = 0;
3148
3149         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3150         if (oc) {
3151                 LASSERT(!capa_is_expired(oc));
3152                 RETURN(oc);
3153         }
3154
3155         spin_lock(&capa_lock);
3156         *key = osd->od_capa_keys[1];
3157         spin_unlock(&capa_lock);
3158
3159         capa->lc_keyid = key->lk_keyid;
3160         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3161
3162         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3163         if (rc) {
3164                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3165                 RETURN(ERR_PTR(rc));
3166         }
3167
3168         oc = capa_add(osd->od_capa_hash, capa);
3169         RETURN(oc);
3170 }
3171
3172 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3173                            __u64 start, __u64 end)
3174 {
3175         struct osd_object       *obj    = osd_dt_obj(dt);
3176         struct inode            *inode  = obj->oo_inode;
3177     &