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 (!list_empty(&scrub->os_inconsistent_items)) {
 590                 /* Search order: 2. OI scrub pending list. */
 591                 result = osd_oii_lookup(dev, fid, id);
 592                 if (result == 0)
 593                         goto iget;
 594         }
 595
 596         /* Search order: 3. OI files. */
 597         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 598         if (result == -ENOENT) {
 599                 if (!fid_is_norm(fid) ||
 600                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 601                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 602                                       sf->sf_oi_bitmap))
 603                         GOTO(out, result = 0);
 604
 605                 goto trigger;
 606         }
 607
 608         if (result != 0)
 609                 GOTO(out, result);
 610
 611         in_oi = true;
 612
 613 iget:
 614         inode = osd_iget_check(info, dev, fid, id, in_oi);
 615         if (IS_ERR(inode)) {
 616                 result = PTR_ERR(inode);
 617                 if (result == -ENOENT || result == -ESTALE) {
 618                         if (!in_oi)
 619                                 fid_zero(&oic->oic_fid);
 620
 621                         GOTO(out, result = -ENOENT);
 622                 } else if (result == -EREMCHG) {
 623
 624 trigger:
 625                         if (!in_oi)
 626                                 fid_zero(&oic->oic_fid);
 627
 628                         if (unlikely(triggered))
 629                                 GOTO(out, result = saved);
 630
 631                         triggered = true;
 632                         if (thread_is_running(&scrub->os_thread)) {
 633                                 result = -EINPROGRESS;
 634                         } else if (!dev->od_noscrub) {
 635                                 result = osd_scrub_start(dev);
 636                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 637                                               "for "DFID", rc = %d [1]\n",
 638                                               osd_name(dev), PFID(fid),result);
 639                                 if (result == 0 || result == -EALREADY)
 640                                         result = -EINPROGRESS;
 641                                 else
 642                                         result = -EREMCHG;
 643                         }
 644
 645                         /* We still have chance to get the valid inode: for the
 646                          * object which is referenced by remote name entry, the
 647                          * object on the local MDT will be linked under the dir
 648                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 649                          *
 650                          * We do not know whether the object for the given FID
 651                          * is referenced by some remote name entry or not, and
 652                          * especially for DNE II, a multiple-linked object may
 653                          * have many name entries reside on many MDTs.
 654                          *
 655                          * To simplify the operation, OSD will not distinguish
 656                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 657                          * only happened for the RPC from other MDT during the
 658                          * OI scrub, or for the client side RPC with FID only,
 659                          * such as FID to path, or from old connected client. */
 660                         saved = result;
 661                         result = osd_lookup_in_remote_parent(info, dev,
 662                                                              fid, id);
 663                         if (result == 0) {
 664                                 in_oi = false;
 665                                 goto iget;
 666                         }
 667
 668                         result = saved;
 669                 }
 670
 671                 GOTO(out, result);
 672         }
 673
 674         obj->oo_inode = inode;
 675         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 676
 677         result = osd_check_lma(env, obj);
 678         if (result != 0) {
 679                 iput(inode);
 680                 obj->oo_inode = NULL;
 681                 if (result == -EREMCHG) {
 682                         if (!in_oi) {
 683                                 result = osd_oi_lookup(info, dev, fid, id,
 684                                                        OI_CHECK_FLD);
 685                                 if (result != 0) {
 686                                         fid_zero(&oic->oic_fid);
 687                                         GOTO(out, result);
 688                                 }
 689                         }
 690
 691                         goto trigger;
 692                 }
 693
 694                 GOTO(out, result);
 695         }
 696
 697         obj->oo_compat_dot_created = 1;
 698         obj->oo_compat_dotdot_created = 1;
 699
 700         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 701                 GOTO(out, result = 0);
 702
 703         LASSERT(obj->oo_hl_head == NULL);
 704         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 705         if (obj->oo_hl_head == NULL) {
 706                 obj->oo_inode = NULL;
 707                 iput(inode);
 708                 GOTO(out, result = -ENOMEM);
 709         }
 710         GOTO(out, result = 0);
 711
 712 out:
 713         LINVRNT(osd_invariant(obj));
 714         return result;
 715 }
 716
 717 /*
 718  * Concurrency: shouldn't matter.
 719  */
 720 static void osd_object_init0(struct osd_object *obj)
 721 {
 722         LASSERT(obj->oo_inode != NULL);
 723         obj->oo_dt.do_body_ops = &osd_body_ops;
 724         obj->oo_dt.do_lu.lo_header->loh_attr |=
 725                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 726 }
 727
 728 /*
 729  * Concurrency: no concurrent access is possible that early in object
 730  * life-cycle.
 731  */
 732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 733                            const struct lu_object_conf *conf)
 734 {
 735         struct osd_object *obj = osd_obj(l);
 736         int result;
 737
 738         LINVRNT(osd_invariant(obj));
 739
 740         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 741                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 742                 l->lo_header->loh_attr |= LOHA_EXISTS;
 743                 return 0;
 744         }
 745
 746         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 747         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 748         if (result == 0 && obj->oo_inode != NULL)
 749                 osd_object_init0(obj);
 750
 751         LINVRNT(osd_invariant(obj));
 752         return result;
 753 }
 754
 755 /*
 756  * Concurrency: no concurrent access is possible that late in object
 757  * life-cycle.
 758  */
 759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 760 {
 761         struct osd_object *obj = osd_obj(l);
 762
 763         LINVRNT(osd_invariant(obj));
 764
 765         dt_object_fini(&obj->oo_dt);
 766         if (obj->oo_hl_head != NULL)
 767                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 768         OBD_FREE_PTR(obj);
 769 }
 770
 771 /*
 772  * Concurrency: no concurrent access is possible that late in object
 773  * life-cycle.
 774  */
 775 static void osd_index_fini(struct osd_object *o)
 776 {
 777         struct iam_container *bag;
 778
 779         if (o->oo_dir != NULL) {
 780                 bag = &o->oo_dir->od_container;
 781                 if (o->oo_inode != NULL) {
 782                         if (bag->ic_object == o->oo_inode)
 783                                 iam_container_fini(bag);
 784                 }
 785                 OBD_FREE_PTR(o->oo_dir);
 786                 o->oo_dir = NULL;
 787         }
 788 }
 789
 790 /*
 791  * Concurrency: no concurrent access is possible that late in object
 792  * life-cycle (for all existing callers, that is. New callers have to provide
 793  * their own locking.)
 794  */
 795 static int osd_inode_unlinked(const struct inode *inode)
 796 {
 797         return inode->i_nlink == 0;
 798 }
 799
 800 enum {
 801         OSD_TXN_OI_DELETE_CREDITS    = 20,
 802         OSD_TXN_INODE_DELETE_CREDITS = 20
 803 };
 804
 805 /*
 806  * Journal
 807  */
 808
 809 #if OSD_THANDLE_STATS
 810 /**
 811  * Set time when the handle is allocated
 812  */
 813 static void osd_th_alloced(struct osd_thandle *oth)
 814 {
 815         oth->oth_alloced = cfs_time_current();
 816 }
 817
 818 /**
 819  * Set time when the handle started
 820  */
 821 static void osd_th_started(struct osd_thandle *oth)
 822 {
 823         oth->oth_started = cfs_time_current();
 824 }
 825
 826 /**
 827  * Helper function to convert time interval to microseconds packed in
 828  * long int.
 829  */
 830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 831 {
 832         struct timeval val;
 833
 834         cfs_duration_usec(cfs_time_sub(end, start), &val);
 835         return val.tv_sec * 1000000 + val.tv_usec;
 836 }
 837
 838 /**
 839  * Check whether the we deal with this handle for too long.
 840  */
 841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 842                                 cfs_time_t alloced, cfs_time_t started,
 843                                 cfs_time_t closed)
 844 {
 845         cfs_time_t now = cfs_time_current();
 846
 847         LASSERT(dev != NULL);
 848
 849         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 850                             interval_to_usec(alloced, started));
 851         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 852                             interval_to_usec(started, closed));
 853         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 854                             interval_to_usec(closed, now));
 855
 856         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 857                 CWARN("transaction handle %p was open for too long: "
 858                       "now "CFS_TIME_T" ,"
 859                       "alloced "CFS_TIME_T" ,"
 860                       "started "CFS_TIME_T" ,"
 861                       "closed "CFS_TIME_T"\n",
 862                       oth, now, alloced, started, closed);
 863                 libcfs_debug_dumpstack(NULL);
 864         }
 865 }
 866
 867 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 868 {                                                                       \
 869         cfs_time_t __closed = cfs_time_current();                       \
 870         cfs_time_t __alloced = oth->oth_alloced;                        \
 871         cfs_time_t __started = oth->oth_started;                        \
 872                                                                         \
 873         expr;                                                           \
 874         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 875 }
 876
 877 #else /* OSD_THANDLE_STATS */
 878
 879 #define osd_th_alloced(h)                  do {} while(0)
 880 #define osd_th_started(h)                  do {} while(0)
 881 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 882
 883 #endif /* OSD_THANDLE_STATS */
 884
 885 /*
 886  * Concurrency: doesn't access mutable data.
 887  */
 888 static int osd_param_is_not_sane(const struct osd_device *dev,
 889                                  const struct thandle *th)
 890 {
 891         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 892
 893         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 894 }
 895
 896 /*
 897  * Concurrency: shouldn't matter.
 898  */
 899 static void osd_trans_commit_cb(struct super_block *sb,
 900                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 901 {
 902         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 903         struct thandle     *th  = &oh->ot_super;
 904         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 905         struct dt_txn_commit_cb *dcb, *tmp;
 906
 907         LASSERT(oh->ot_handle == NULL);
 908
 909         if (error)
 910                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 911
 912         dt_txn_hook_commit(th);
 913
 914         /* call per-transaction callbacks if any */
 915         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 916                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 917                          "commit callback entry: magic=%x name='%s'\n",
 918                          dcb->dcb_magic, dcb->dcb_name);
 919                 list_del_init(&dcb->dcb_linkage);
 920                 dcb->dcb_func(NULL, th, dcb, error);
 921         }
 922
 923         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 924         lu_device_put(lud);
 925         th->th_dev = NULL;
 926
 927         lu_context_exit(&th->th_ctx);
 928         lu_context_fini(&th->th_ctx);
 929         thandle_put(th);
 930 }
 931
 932 static struct thandle *osd_trans_create(const struct lu_env *env,
 933                                         struct dt_device *d)
 934 {
 935         struct osd_thread_info  *oti = osd_oti_get(env);
 936         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 937         struct osd_thandle      *oh;
 938         struct thandle          *th;
 939         ENTRY;
 940
 941         /* on pending IO in this thread should left from prev. request */
 942         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 943
 944         th = ERR_PTR(-ENOMEM);
 945         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 946         if (oh != NULL) {
 947                 oh->ot_quota_trans = &oti->oti_quota_trans;
 948                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 949                 th = &oh->ot_super;
 950                 th->th_dev = d;
 951                 th->th_result = 0;
 952                 th->th_tags = LCT_TX_HANDLE;
 953                 oh->ot_credits = 0;
 954                 atomic_set(&th->th_refc, 1);
 955                 th->th_alloc_size = sizeof(*oh);
 956                 oti->oti_dev = osd_dt_dev(d);
 957                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 958                 osd_th_alloced(oh);
 959
 960                 memset(oti->oti_declare_ops, 0,
 961                                         sizeof(oti->oti_declare_ops));
 962                 memset(oti->oti_declare_ops_rb, 0,
 963                                         sizeof(oti->oti_declare_ops_rb));
 964                 memset(oti->oti_declare_ops_cred, 0,
 965                                         sizeof(oti->oti_declare_ops_cred));
 966                 oti->oti_rollback = false;
 967         }
 968         RETURN(th);
 969 }
 970
 971 /*
 972  * Concurrency: shouldn't matter.
 973  */
 974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 975                     struct thandle *th)
 976 {
 977         struct osd_thread_info *oti = osd_oti_get(env);
 978         struct osd_device  *dev = osd_dt_dev(d);
 979         handle_t           *jh;
 980         struct osd_thandle *oh;
 981         int rc;
 982
 983         ENTRY;
 984
 985         LASSERT(current->journal_info == NULL);
 986
 987         oh = container_of0(th, struct osd_thandle, ot_super);
 988         LASSERT(oh != NULL);
 989         LASSERT(oh->ot_handle == NULL);
 990
 991         rc = dt_txn_hook_start(env, d, th);
 992         if (rc != 0)
 993                 GOTO(out, rc);
 994
 995         if (unlikely(osd_param_is_not_sane(dev, th))) {
 996                 static unsigned long last_printed;
 997                 static int last_credits;
 998
 999                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1001                       oh->ot_credits,
1002                       osd_journal(dev)->j_max_transaction_buffers);
1003                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1004                       oti->oti_declare_ops[OSD_OT_CREATE],
1005                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006                       oti->oti_declare_ops[OSD_OT_DESTROY],
1007                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_WRITE],
1015                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016                       oti->oti_declare_ops[OSD_OT_PUNCH],
1017                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops[OSD_OT_QUOTA],
1019                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1021                       oti->oti_declare_ops[OSD_OT_INSERT],
1022                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023                       oti->oti_declare_ops[OSD_OT_DELETE],
1024                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1027                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1029                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1030
1031                 if (last_credits != oh->ot_credits &&
1032                     time_after(jiffies, last_printed + 60 * HZ)) {
1033                         libcfs_debug_dumpstack(NULL);
1034                         last_credits = oh->ot_credits;
1035                         last_printed = jiffies;
1036                 }
1037                 /* XXX Limit the credits to 'max_transaction_buffers', and
1038                  *     let the underlying filesystem to catch the error if
1039                  *     we really need so many credits.
1040                  *
1041                  *     This should be removed when we can calculate the
1042                  *     credits precisely. */
1043                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1044         }
1045
1046         /*
1047          * XXX temporary stuff. Some abstraction layer should
1048          * be used.
1049          */
1050         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1051         osd_th_started(oh);
1052         if (!IS_ERR(jh)) {
1053                 oh->ot_handle = jh;
1054                 LASSERT(oti->oti_txns == 0);
1055                 lu_context_init(&th->th_ctx, th->th_tags);
1056                 lu_context_enter(&th->th_ctx);
1057
1058                 lu_device_get(&d->dd_lu_dev);
1059                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1060                               "osd-tx", th);
1061                 oti->oti_txns++;
1062                 rc = 0;
1063         } else {
1064                 rc = PTR_ERR(jh);
1065         }
1066 out:
1067         RETURN(rc);
1068 }
1069
1070 static int osd_seq_exists(const struct lu_env *env,
1071                               struct osd_device *osd, obd_seq seq)
1072 {
1073         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1074         struct seq_server_site  *ss = osd_seq_site(osd);
1075         int                     rc;
1076         ENTRY;
1077
1078         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         osd_fini_iobuf(oti->oti_dev, iobuf);
1155         if (!rc)
1156                 rc = iobuf->dr_error;
1157
1158         RETURN(rc);
1159 }
1160
1161 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 {
1163         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1164                                                ot_super);
1165
1166         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1167         LASSERT(&dcb->dcb_func != NULL);
1168         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1169
1170         return 0;
1171 }
1172
1173 /*
1174  * Called just before object is freed. Releases all resources except for
1175  * object itself (that is released by osd_object_free()).
1176  *
1177  * Concurrency: no concurrent access is possible that late in object
1178  * life-cycle.
1179  */
1180 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 {
1182         struct osd_object *obj   = osd_obj(l);
1183         struct inode      *inode = obj->oo_inode;
1184
1185         LINVRNT(osd_invariant(obj));
1186
1187         /*
1188          * If object is unlinked remove fid->ino mapping from object index.
1189          */
1190
1191         osd_index_fini(obj);
1192         if (inode != NULL) {
1193                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1194                 qid_t                    uid = i_uid_read(inode);
1195                 qid_t                    gid = i_gid_read(inode);
1196
1197                 iput(inode);
1198                 obj->oo_inode = NULL;
1199
1200                 if (qsd != NULL) {
1201                         struct osd_thread_info  *info = osd_oti_get(env);
1202                         struct lquota_id_info   *qi = &info->oti_qi;
1203
1204                         /* Release granted quota to master if necessary */
1205                         qi->lqi_id.qid_uid = uid;
1206                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207
1208                         qi->lqi_id.qid_uid = gid;
1209                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1210                 }
1211         }
1212 }
1213
1214 /*
1215  * Concurrency: ->loo_object_release() is called under site spin-lock.
1216  */
1217 static void osd_object_release(const struct lu_env *env,
1218                                struct lu_object *l)
1219 {
1220 }
1221
1222 /*
1223  * Concurrency: shouldn't matter.
1224  */
1225 static int osd_object_print(const struct lu_env *env, void *cookie,
1226                             lu_printer_t p, const struct lu_object *l)
1227 {
1228         struct osd_object *o = osd_obj(l);
1229         struct iam_descr  *d;
1230
1231         if (o->oo_dir != NULL)
1232                 d = o->oo_dir->od_container.ic_descr;
1233         else
1234                 d = NULL;
1235         return (*p)(env, cookie,
1236                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237                     o, o->oo_inode,
1238                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1239                     o->oo_inode ? o->oo_inode->i_generation : 0,
1240                     d ? d->id_ops->id_name : "plain");
1241 }
1242
1243 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1244
1245 /*
1246  * Concurrency: shouldn't matter.
1247  */
1248 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1249                struct obd_statfs *sfs)
1250 {
1251         struct osd_device  *osd = osd_dt_dev(d);
1252         struct super_block *sb = osd_sb(osd);
1253         struct kstatfs     *ksfs;
1254         int result = 0;
1255
1256         if (unlikely(osd->od_mnt == NULL))
1257                 return -EINPROGRESS;
1258
1259         /* osd_lproc.c call this without env, allocate ksfs for that case */
1260         if (unlikely(env == NULL)) {
1261                 OBD_ALLOC_PTR(ksfs);
1262                 if (ksfs == NULL)
1263                         return -ENOMEM;
1264         } else {
1265                 ksfs = &osd_oti_get(env)->oti_ksfs;
1266         }
1267
1268         spin_lock(&osd->od_osfs_lock);
1269         result = sb->s_op->statfs(sb->s_root, ksfs);
1270         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1271                 statfs_pack(sfs, ksfs);
1272                 if (sb->s_flags & MS_RDONLY)
1273                         sfs->os_state = OS_STATE_READONLY;
1274         }
1275
1276         spin_unlock(&osd->od_osfs_lock);
1277
1278         if (unlikely(env == NULL))
1279                 OBD_FREE_PTR(ksfs);
1280
1281         /* Reserve a small amount of space for local objects like last_rcvd,
1282          * llog, quota files, ... */
1283         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1284                 sfs->os_bavail = 0;
1285         } else {
1286                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1287                 /** Take out metadata overhead for indirect blocks */
1288                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1289         }
1290
1291         return result;
1292 }
1293
1294 /**
1295  * Estimate space needed for file creations. We assume the largest filename
1296  * which is 2^64 - 1, hence a filename of 20 chars.
1297  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298  */
1299 #ifdef __LDISKFS_DIR_REC_LEN
1300 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #else
1302 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1303 #endif
1304
1305 /*
1306  * Concurrency: doesn't access mutable data.
1307  */
1308 static void osd_conf_get(const struct lu_env *env,
1309                          const struct dt_device *dev,
1310                          struct dt_device_param *param)
1311 {
1312         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1313         int                ea_overhead;
1314
1315         /*
1316          * XXX should be taken from not-yet-existing fs abstraction layer.
1317          */
1318         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1319         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1320         param->ddp_block_shift  = sb->s_blocksize_bits;
1321         param->ddp_mount_type     = LDD_MT_LDISKFS;
1322         param->ddp_maxbytes       = sb->s_maxbytes;
1323         /* Overhead estimate should be fairly accurate, so we really take a tiny
1324          * error margin which also avoids fragmenting the filesystem too much */
1325         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1326         /* inode are statically allocated, so per-inode space consumption
1327          * is the space consumed by the directory entry */
1328         param->ddp_inodespace     = PER_OBJ_USAGE;
1329         /* per-fragment overhead to be used by the client code */
1330         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1331         param->ddp_mntopts      = 0;
1332         if (test_opt(sb, XATTR_USER))
1333                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1334         if (test_opt(sb, POSIX_ACL))
1335                 param->ddp_mntopts |= MNTOPT_ACL;
1336
1337         /* LOD might calculate the max stripe count based on max_ea_size,
1338          * so we need take account in the overhead as well,
1339          * xattr_header + magic + xattr_entry_head */
1340         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1341                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342
1343 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1344         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1345                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1346                                                                 ea_overhead;
1347         else
1348 #endif
1349                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1350 }
1351
1352 /*
1353  * Concurrency: shouldn't matter.
1354  */
1355 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 {
1357         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1358         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1359 }
1360
1361 /**
1362  * Start commit for OSD device.
1363  *
1364  * An implementation of dt_commit_async method for OSD device.
1365  * Asychronously starts underlayng fs sync and thereby a transaction
1366  * commit.
1367  *
1368  * \param env environment
1369  * \param d dt device
1370  *
1371  * \see dt_device_operations
1372  */
1373 static int osd_commit_async(const struct lu_env *env,
1374                             struct dt_device *d)
1375 {
1376         struct super_block *s = osd_sb(osd_dt_dev(d));
1377         ENTRY;
1378
1379         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1380         RETURN(s->s_op->sync_fs(s, 0));
1381 }
1382
1383 /*
1384  * Concurrency: shouldn't matter.
1385  */
1386
1387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 {
1389         struct super_block *sb = osd_sb(osd_dt_dev(d));
1390         struct block_device *dev = sb->s_bdev;
1391 #ifdef HAVE_DEV_SET_RDONLY
1392         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1393         int rc = 0;
1394 #else
1395         int rc = -EOPNOTSUPP;
1396 #endif
1397         ENTRY;
1398
1399 #ifdef HAVE_DEV_SET_RDONLY
1400         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401
1402         if (jdev && (jdev != dev)) {
1403                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404                        (long)jdev);
1405                 dev_set_rdonly(jdev);
1406         }
1407         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1408         dev_set_rdonly(dev);
1409 #else
1410         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1411                osd_dt_dev(d)->od_svname, (long)dev, rc);
1412 #endif
1413         RETURN(rc);
1414 }
1415
1416 /*
1417  * Concurrency: serialization provided by callers.
1418  */
1419 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1420                               int mode, unsigned long timeout, __u32 alg,
1421                               struct lustre_capa_key *keys)
1422 {
1423         struct osd_device *dev = osd_dt_dev(d);
1424         ENTRY;
1425
1426         dev->od_fl_capa = mode;
1427         dev->od_capa_timeout = timeout;
1428         dev->od_capa_alg = alg;
1429         dev->od_capa_keys = keys;
1430         RETURN(0);
1431 }
1432
1433 /**
1434  * Note: we do not count into QUOTA here.
1435  * If we mount with --data_journal we may need more.
1436  */
1437 const int osd_dto_credits_noquota[DTO_NR] = {
1438         /**
1439          * Insert.
1440          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1441          * SINGLEDATA_TRANS_BLOCKS(8)
1442          * XXX Note: maybe iam need more, since iam have more level than
1443          *           EXT3 htree.
1444          */
1445         [DTO_INDEX_INSERT]  = 16,
1446         /**
1447          * Delete
1448          * just modify a single entry, probably merge few within a block
1449          */
1450         [DTO_INDEX_DELETE]  = 1,
1451         /**
1452          * Used for OI scrub
1453          */
1454         [DTO_INDEX_UPDATE]  = 16,
1455         /**
1456          * 4(inode, inode bits, groups, GDT)
1457          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1458          */
1459         [DTO_OBJECT_CREATE] = 4,
1460         /**
1461          * 4(inode, inode bits, groups, GDT)
1462          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1463          */
1464         [DTO_OBJECT_DELETE] = 4,
1465         /**
1466          * Attr set credits (inode)
1467          */
1468         [DTO_ATTR_SET_BASE] = 1,
1469         /**
1470          * Xattr set. The same as xattr of EXT3.
1471          * DATA_TRANS_BLOCKS(14)
1472          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1473          * are also counted in. Do not know why?
1474          */
1475         [DTO_XATTR_SET]     = 14,
1476         /**
1477          * credits for inode change during write.
1478          */
1479         [DTO_WRITE_BASE]    = 3,
1480         /**
1481          * credits for single block write.
1482          */
1483         [DTO_WRITE_BLOCK]   = 14,
1484         /**
1485          * Attr set credits for chown.
1486          * This is extra credits for setattr, and it is null without quota
1487          */
1488         [DTO_ATTR_SET_CHOWN] = 0
1489 };
1490
1491 static const struct dt_device_operations osd_dt_ops = {
1492         .dt_root_get       = osd_root_get,
1493         .dt_statfs         = osd_statfs,
1494         .dt_trans_create   = osd_trans_create,
1495         .dt_trans_start    = osd_trans_start,
1496         .dt_trans_stop     = osd_trans_stop,
1497         .dt_trans_cb_add   = osd_trans_cb_add,
1498         .dt_conf_get       = osd_conf_get,
1499         .dt_sync           = osd_sync,
1500         .dt_ro             = osd_ro,
1501         .dt_commit_async   = osd_commit_async,
1502         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1503 };
1504
1505 static void osd_object_read_lock(const struct lu_env *env,
1506                                  struct dt_object *dt, unsigned role)
1507 {
1508         struct osd_object *obj = osd_dt_obj(dt);
1509         struct osd_thread_info *oti = osd_oti_get(env);
1510
1511         LINVRNT(osd_invariant(obj));
1512
1513         LASSERT(obj->oo_owner != env);
1514         down_read_nested(&obj->oo_sem, role);
1515
1516         LASSERT(obj->oo_owner == NULL);
1517         oti->oti_r_locks++;
1518 }
1519
1520 static void osd_object_write_lock(const struct lu_env *env,
1521                                   struct dt_object *dt, unsigned role)
1522 {
1523         struct osd_object *obj = osd_dt_obj(dt);
1524         struct osd_thread_info *oti = osd_oti_get(env);
1525
1526         LINVRNT(osd_invariant(obj));
1527
1528         LASSERT(obj->oo_owner != env);
1529         down_write_nested(&obj->oo_sem, role);
1530
1531         LASSERT(obj->oo_owner == NULL);
1532         obj->oo_owner = env;
1533         oti->oti_w_locks++;
1534 }
1535
1536 static void osd_object_read_unlock(const struct lu_env *env,
1537                                    struct dt_object *dt)
1538 {
1539         struct osd_object *obj = osd_dt_obj(dt);
1540         struct osd_thread_info *oti = osd_oti_get(env);
1541
1542         LINVRNT(osd_invariant(obj));
1543
1544         LASSERT(oti->oti_r_locks > 0);
1545         oti->oti_r_locks--;
1546         up_read(&obj->oo_sem);
1547 }
1548
1549 static void osd_object_write_unlock(const struct lu_env *env,
1550                                     struct dt_object *dt)
1551 {
1552         struct osd_object *obj = osd_dt_obj(dt);
1553         struct osd_thread_info *oti = osd_oti_get(env);
1554
1555         LINVRNT(osd_invariant(obj));
1556
1557         LASSERT(obj->oo_owner == env);
1558         LASSERT(oti->oti_w_locks > 0);
1559         oti->oti_w_locks--;
1560         obj->oo_owner = NULL;
1561         up_write(&obj->oo_sem);
1562 }
1563
1564 static int osd_object_write_locked(const struct lu_env *env,
1565                                    struct dt_object *dt)
1566 {
1567         struct osd_object *obj = osd_dt_obj(dt);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         return obj->oo_owner == env;
1572 }
1573
1574 static int capa_is_sane(const struct lu_env *env,
1575                         struct osd_device *dev,
1576                         struct lustre_capa *capa,
1577                         struct lustre_capa_key *keys)
1578 {
1579         struct osd_thread_info *oti = osd_oti_get(env);
1580         struct lustre_capa *tcapa = &oti->oti_capa;
1581         struct obd_capa *oc;
1582         int i, rc = 0;
1583         ENTRY;
1584
1585         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586         if (oc) {
1587                 if (capa_is_expired(oc)) {
1588                         DEBUG_CAPA(D_ERROR, capa, "expired");
1589                         rc = -ESTALE;
1590                 }
1591                 capa_put(oc);
1592                 RETURN(rc);
1593         }
1594
1595         if (capa_is_expired_sec(capa)) {
1596                 DEBUG_CAPA(D_ERROR, capa, "expired");
1597                 RETURN(-ESTALE);
1598         }
1599
1600         spin_lock(&capa_lock);
1601         for (i = 0; i < 2; i++) {
1602                 if (keys[i].lk_keyid == capa->lc_keyid) {
1603                         oti->oti_capa_key = keys[i];
1604                         break;
1605                 }
1606         }
1607         spin_unlock(&capa_lock);
1608
1609         if (i == 2) {
1610                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1619                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1620                 RETURN(-EACCES);
1621         }
1622
1623         oc = capa_add(dev->od_capa_hash, capa);
1624         capa_put(oc);
1625
1626         RETURN(0);
1627 }
1628
1629 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1630                     struct lustre_capa *capa, __u64 opc)
1631 {
1632         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1633         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1634         struct lu_capainfo *lci;
1635         int rc;
1636
1637         if (!osd->od_fl_capa)
1638                 return 0;
1639
1640         if (capa == BYPASS_CAPA)
1641                 return 0;
1642
1643         lci = lu_capainfo_get(env);
1644         if (unlikely(lci == NULL))
1645                 return 0;
1646
1647         if (lci->lci_auth == LC_ID_NONE)
1648                 return 0;
1649
1650         if (capa == NULL) {
1651                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1652                        osd_name(osd), PFID(fid), -EACCES);
1653                 return -EACCES;
1654         }
1655
1656         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1657                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1658                            PFID(fid));
1659                 return -EACCES;
1660         }
1661
1662         if (!capa_opc_supported(capa, opc)) {
1663                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1664                 return -EACCES;
1665         }
1666
1667         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668         if (rc != 0) {
1669                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1670                 return -EACCES;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static struct timespec *osd_inode_time(const struct lu_env *env,
1677                                        struct inode *inode, __u64 seconds)
1678 {
1679         struct osd_thread_info  *oti = osd_oti_get(env);
1680         struct timespec         *t   = &oti->oti_time;
1681
1682         t->tv_sec = seconds;
1683         t->tv_nsec = 0;
1684         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1685         return t;
1686 }
1687
1688
1689 static void osd_inode_getattr(const struct lu_env *env,
1690                               struct inode *inode, struct lu_attr *attr)
1691 {
1692         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1693                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1694                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1695                            LA_TYPE;
1696
1697         attr->la_atime   = LTIME_S(inode->i_atime);
1698         attr->la_mtime   = LTIME_S(inode->i_mtime);
1699         attr->la_ctime   = LTIME_S(inode->i_ctime);
1700         attr->la_mode    = inode->i_mode;
1701         attr->la_size    = i_size_read(inode);
1702         attr->la_blocks  = inode->i_blocks;
1703         attr->la_uid     = i_uid_read(inode);
1704         attr->la_gid     = i_gid_read(inode);
1705         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1706         attr->la_nlink   = inode->i_nlink;
1707         attr->la_rdev    = inode->i_rdev;
1708         attr->la_blksize = 1 << inode->i_blkbits;
1709         attr->la_blkbits = inode->i_blkbits;
1710 }
1711
1712 static int osd_attr_get(const struct lu_env *env,
1713                         struct dt_object *dt,
1714                         struct lu_attr *attr,
1715                         struct lustre_capa *capa)
1716 {
1717         struct osd_object *obj = osd_dt_obj(dt);
1718
1719         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1720         LINVRNT(osd_invariant(obj));
1721
1722         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1723                 return -EACCES;
1724
1725         spin_lock(&obj->oo_guard);
1726         osd_inode_getattr(env, obj->oo_inode, attr);
1727         spin_unlock(&obj->oo_guard);
1728         return 0;
1729 }
1730
1731 static int osd_declare_attr_set(const struct lu_env *env,
1732                                 struct dt_object *dt,
1733                                 const struct lu_attr *attr,
1734                                 struct thandle *handle)
1735 {
1736         struct osd_thandle     *oh;
1737         struct osd_object      *obj;
1738         struct osd_thread_info *info = osd_oti_get(env);
1739         struct lquota_id_info  *qi = &info->oti_qi;
1740         qid_t                   uid;
1741         qid_t                   gid;
1742         long long               bspace;
1743         int                     rc = 0;
1744         bool                    allocated;
1745         ENTRY;
1746
1747         LASSERT(dt != NULL);
1748         LASSERT(handle != NULL);
1749
1750         obj = osd_dt_obj(dt);
1751         LASSERT(osd_invariant(obj));
1752
1753         oh = container_of0(handle, struct osd_thandle, ot_super);
1754         LASSERT(oh->ot_handle == NULL);
1755
1756         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1757                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1758
1759         if (attr == NULL || obj->oo_inode == NULL)
1760                 RETURN(rc);
1761
1762         bspace   = obj->oo_inode->i_blocks;
1763         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1764         bspace   = toqb(bspace);
1765
1766         /* Changing ownership is always preformed by super user, it should not
1767          * fail with EDQUOT.
1768          *
1769          * We still need to call the osd_declare_qid() to calculate the journal
1770          * credits for updating quota accounting files and to trigger quota
1771          * space adjustment once the operation is completed.*/
1772         if ((attr->la_valid & LA_UID) != 0 &&
1773              attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
1774                 qi->lqi_type = USRQUOTA;
1775
1776                 /* inode accounting */
1777                 qi->lqi_is_blk = false;
1778
1779                 /* one more inode for the new owner ... */
1780                 qi->lqi_id.qid_uid = attr->la_uid;
1781                 qi->lqi_space      = 1;
1782                 allocated = (attr->la_uid == 0) ? true : false;
1783                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1784                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1785                         rc = 0;
1786                 if (rc)
1787                         RETURN(rc);
1788
1789                 /* and one less inode for the current uid */
1790                 qi->lqi_id.qid_uid = uid;
1791                 qi->lqi_space      = -1;
1792                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1793                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1794                         rc = 0;
1795                 if (rc)
1796                         RETURN(rc);
1797
1798                 /* block accounting */
1799                 qi->lqi_is_blk = true;
1800
1801                 /* more blocks for the new owner ... */
1802                 qi->lqi_id.qid_uid = attr->la_uid;
1803                 qi->lqi_space      = bspace;
1804                 allocated = (attr->la_uid == 0) ? true : false;
1805                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1806                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1807                         rc = 0;
1808                 if (rc)
1809                         RETURN(rc);
1810
1811                 /* and finally less blocks for the current owner */
1812                 qi->lqi_id.qid_uid = uid;
1813                 qi->lqi_space      = -bspace;
1814                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1815                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1816                         rc = 0;
1817                 if (rc)
1818                         RETURN(rc);
1819         }
1820
1821         if (attr->la_valid & LA_GID &&
1822             attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
1823                 qi->lqi_type = GRPQUOTA;
1824
1825                 /* inode accounting */
1826                 qi->lqi_is_blk = false;
1827
1828                 /* one more inode for the new group owner ... */
1829                 qi->lqi_id.qid_gid = attr->la_gid;
1830                 qi->lqi_space      = 1;
1831                 allocated = (attr->la_gid == 0) ? true : false;
1832                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1833                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1834                         rc = 0;
1835                 if (rc)
1836                         RETURN(rc);
1837
1838                 /* and one less inode for the current gid */
1839                 qi->lqi_id.qid_gid = gid;
1840                 qi->lqi_space      = -1;
1841                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1842                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1843                         rc = 0;
1844                 if (rc)
1845                         RETURN(rc);
1846
1847                 /* block accounting */
1848                 qi->lqi_is_blk = true;
1849
1850                 /* more blocks for the new owner ... */
1851                 qi->lqi_id.qid_gid = attr->la_gid;
1852                 qi->lqi_space      = bspace;
1853                 allocated = (attr->la_gid == 0) ? true : false;
1854                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1855                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1856                         rc = 0;
1857                 if (rc)
1858                         RETURN(rc);
1859
1860                 /* and finally less blocks for the current owner */
1861                 qi->lqi_id.qid_gid = gid;
1862                 qi->lqi_space      = -bspace;
1863                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1864                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1865                         rc = 0;
1866                 if (rc)
1867                         RETURN(rc);
1868         }
1869
1870         RETURN(rc);
1871 }
1872
1873 static int osd_inode_setattr(const struct lu_env *env,
1874                              struct inode *inode, const struct lu_attr *attr)
1875 {
1876         __u64 bits;
1877
1878         bits = attr->la_valid;
1879
1880         if (bits & LA_ATIME)
1881                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1882         if (bits & LA_CTIME)
1883                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1884         if (bits & LA_MTIME)
1885                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1886         if (bits & LA_SIZE) {
1887                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1888                 i_size_write(inode, attr->la_size);
1889         }
1890
1891 #if 0
1892         /* OSD should not change "i_blocks" which is used by quota.
1893          * "i_blocks" should be changed by ldiskfs only. */
1894         if (bits & LA_BLOCKS)
1895                 inode->i_blocks = attr->la_blocks;
1896 #endif
1897         if (bits & LA_MODE)
1898                 inode->i_mode = (inode->i_mode & S_IFMT) |
1899                                 (attr->la_mode & ~S_IFMT);
1900         if (bits & LA_UID)
1901                 i_uid_write(inode, attr->la_uid);
1902         if (bits & LA_GID)
1903                 i_gid_write(inode, attr->la_gid);
1904         if (bits & LA_NLINK)
1905                 set_nlink(inode, attr->la_nlink);
1906         if (bits & LA_RDEV)
1907                 inode->i_rdev = attr->la_rdev;
1908
1909         if (bits & LA_FLAGS) {
1910                 /* always keep S_NOCMTIME */
1911                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1912                                  S_NOCMTIME;
1913         }
1914         return 0;
1915 }
1916
1917 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1918 {
1919         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1920             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1921                 struct iattr    iattr;
1922                 int             rc;
1923
1924                 iattr.ia_valid = 0;
1925                 if (attr->la_valid & LA_UID)
1926                         iattr.ia_valid |= ATTR_UID;
1927                 if (attr->la_valid & LA_GID)
1928                         iattr.ia_valid |= ATTR_GID;
1929                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1930                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1931
1932                 rc = ll_vfs_dq_transfer(inode, &iattr);
1933                 if (rc) {
1934                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1935                                "enforcement enabled on the ldiskfs filesystem?",
1936                                inode->i_sb->s_id, rc);
1937                         return rc;
1938                 }
1939         }
1940         return 0;
1941 }
1942
1943 static int osd_attr_set(const struct lu_env *env,
1944                         struct dt_object *dt,
1945                         const struct lu_attr *attr,
1946                         struct thandle *handle,
1947                         struct lustre_capa *capa)
1948 {
1949         struct osd_object *obj = osd_dt_obj(dt);
1950         struct inode      *inode;
1951         int rc;
1952
1953         LASSERT(handle != NULL);
1954         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1955         LASSERT(osd_invariant(obj));
1956
1957         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1958                 return -EACCES;
1959
1960         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1961
1962         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1963                 struct osd_thread_info  *oti  = osd_oti_get(env);
1964                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1965                 struct lu_fid           *fid1 = &oti->oti_fid;
1966                 struct osd_inode_id     *id   = &oti->oti_id;
1967                 struct iam_path_descr   *ipd;
1968                 struct iam_container    *bag;
1969                 struct osd_thandle      *oh;
1970                 int                      rc;
1971
1972                 fid_cpu_to_be(fid1, fid0);
1973                 memset(id, 1, sizeof(*id));
1974                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1975                                   fid0)->oi_dir.od_container;
1976                 ipd = osd_idx_ipd_get(env, bag);
1977                 if (unlikely(ipd == NULL))
1978                         RETURN(-ENOMEM);
1979
1980                 oh = container_of0(handle, struct osd_thandle, ot_super);
1981                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1982                                 (const struct iam_rec *)id, ipd);
1983                 osd_ipd_put(env, bag, ipd);
1984                 return(rc > 0 ? 0 : rc);
1985         }
1986
1987         inode = obj->oo_inode;
1988         ll_vfs_dq_init(inode);
1989
1990         rc = osd_quota_transfer(inode, attr);
1991         if (rc)
1992                 return rc;
1993
1994         spin_lock(&obj->oo_guard);
1995         rc = osd_inode_setattr(env, inode, attr);
1996         spin_unlock(&obj->oo_guard);
1997
1998         if (!rc)
1999                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2000         return rc;
2001 }
2002
2003 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2004                                     struct osd_object *obj,
2005                                     const char *name, const int namelen)
2006 {
2007         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2008 }
2009
2010 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2011                       umode_t mode, struct dt_allocation_hint *hint,
2012                       struct thandle *th)
2013 {
2014         int result;
2015         struct osd_device  *osd = osd_obj2dev(obj);
2016         struct osd_thandle *oth;
2017         struct dt_object   *parent = NULL;
2018         struct inode       *inode;
2019
2020         LINVRNT(osd_invariant(obj));
2021         LASSERT(obj->oo_inode == NULL);
2022         LASSERT(obj->oo_hl_head == NULL);
2023
2024         if (S_ISDIR(mode) && ldiskfs_pdo) {
2025                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2026                 if (obj->oo_hl_head == NULL)
2027                         return -ENOMEM;
2028         }
2029
2030         oth = container_of(th, struct osd_thandle, ot_super);
2031         LASSERT(oth->ot_handle->h_transaction != NULL);
2032
2033         if (hint && hint->dah_parent)
2034                 parent = hint->dah_parent;
2035
2036         inode = ldiskfs_create_inode(oth->ot_handle,
2037                                      parent ? osd_dt_obj(parent)->oo_inode :
2038                                               osd_sb(osd)->s_root->d_inode,
2039                                      mode);
2040         if (!IS_ERR(inode)) {
2041                 /* Do not update file c/mtime in ldiskfs.
2042                  * NB: don't need any lock because no contention at this
2043                  * early stage */
2044                 inode->i_flags |= S_NOCMTIME;
2045
2046                 /* For new created object, it must be consistent,
2047                  * and it is unnecessary to scrub against it. */
2048                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2049                 obj->oo_inode = inode;
2050                 result = 0;
2051         } else {
2052                 if (obj->oo_hl_head != NULL) {
2053                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2054                         obj->oo_hl_head = NULL;
2055                 }
2056                 result = PTR_ERR(inode);
2057         }
2058         LINVRNT(osd_invariant(obj));
2059         return result;
2060 }
2061
2062 enum {
2063         OSD_NAME_LEN = 255
2064 };
2065
2066 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2067                      struct lu_attr *attr,
2068                      struct dt_allocation_hint *hint,
2069                      struct dt_object_format *dof,
2070                      struct thandle *th)
2071 {
2072         int result;
2073         struct osd_thandle *oth;
2074         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2075
2076         LASSERT(S_ISDIR(attr->la_mode));
2077
2078         oth = container_of(th, struct osd_thandle, ot_super);
2079         LASSERT(oth->ot_handle->h_transaction != NULL);
2080         result = osd_mkfile(info, obj, mode, hint, th);
2081
2082         return result;
2083 }
2084
2085 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2086                         struct lu_attr *attr,
2087                         struct dt_allocation_hint *hint,
2088                         struct dt_object_format *dof,
2089                         struct thandle *th)
2090 {
2091         int result;
2092         struct osd_thandle *oth;
2093         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2094
2095         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2096
2097         LASSERT(S_ISREG(attr->la_mode));
2098
2099         oth = container_of(th, struct osd_thandle, ot_super);
2100         LASSERT(oth->ot_handle->h_transaction != NULL);
2101
2102         result = osd_mkfile(info, obj, mode, hint, th);
2103         if (result == 0) {
2104                 LASSERT(obj->oo_inode != NULL);
2105                 if (feat->dif_flags & DT_IND_VARKEY)
2106                         result = iam_lvar_create(obj->oo_inode,
2107                                                  feat->dif_keysize_max,
2108                                                  feat->dif_ptrsize,
2109                                                  feat->dif_recsize_max,
2110                                                  oth->ot_handle);
2111                 else
2112                         result = iam_lfix_create(obj->oo_inode,
2113                                                  feat->dif_keysize_max,
2114                                                  feat->dif_ptrsize,
2115                                                  feat->dif_recsize_max,
2116                                                  oth->ot_handle);
2117
2118         }
2119         return result;
2120 }
2121
2122 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2123                      struct lu_attr *attr,
2124                      struct dt_allocation_hint *hint,
2125                      struct dt_object_format *dof,
2126                      struct thandle *th)
2127 {
2128         LASSERT(S_ISREG(attr->la_mode));
2129         return osd_mkfile(info, obj, (attr->la_mode &
2130                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2131 }
2132
2133 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2134                      struct lu_attr *attr,
2135                      struct dt_allocation_hint *hint,
2136                      struct dt_object_format *dof,
2137                      struct thandle *th)
2138 {
2139         LASSERT(S_ISLNK(attr->la_mode));
2140         return osd_mkfile(info, obj, (attr->la_mode &
2141                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2142 }
2143
2144 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2145                      struct lu_attr *attr,
2146                      struct dt_allocation_hint *hint,
2147                      struct dt_object_format *dof,
2148                      struct thandle *th)
2149 {
2150         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2151         int result;
2152
2153         LINVRNT(osd_invariant(obj));
2154         LASSERT(obj->oo_inode == NULL);
2155         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2156                 S_ISFIFO(mode) || S_ISSOCK(mode));
2157
2158         result = osd_mkfile(info, obj, mode, hint, th);
2159         if (result == 0) {
2160                 LASSERT(obj->oo_inode != NULL);
2161                 /*
2162                  * This inode should be marked dirty for i_rdev.  Currently
2163                  * that is done in the osd_attr_init().
2164                  */
2165                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2166                                    attr->la_rdev);
2167         }
2168         LINVRNT(osd_invariant(obj));
2169         return result;
2170 }
2171
2172 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2173                               struct lu_attr *,
2174                               struct dt_allocation_hint *hint,
2175                               struct dt_object_format *dof,
2176                               struct thandle *);
2177
2178 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2179 {
2180         osd_obj_type_f result;
2181
2182         switch (type) {
2183         case DFT_DIR:
2184                 result = osd_mkdir;
2185                 break;
2186         case DFT_REGULAR:
2187                 result = osd_mkreg;
2188                 break;
2189         case DFT_SYM:
2190                 result = osd_mksym;
2191                 break;
2192         case DFT_NODE:
2193                 result = osd_mknod;
2194                 break;
2195         case DFT_INDEX:
2196                 result = osd_mk_index;
2197                 break;
2198
2199         default:
2200                 LBUG();
2201                 break;
2202         }
2203         return result;
2204 }
2205
2206
2207 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2208                         struct dt_object *parent, struct dt_object *child,
2209                         umode_t child_mode)
2210 {
2211         LASSERT(ah);
2212
2213         ah->dah_parent = parent;
2214         ah->dah_mode = child_mode;
2215 }
2216
2217 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2218                           struct lu_attr *attr, struct dt_object_format *dof)
2219 {
2220         struct inode   *inode = obj->oo_inode;
2221         __u64           valid = attr->la_valid;
2222         int             result;
2223
2224         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2225
2226         if (dof->dof_type != DFT_NODE)
2227                 attr->la_valid &= ~LA_RDEV;
2228         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2229                 attr->la_valid &= ~LA_ATIME;
2230         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2231                 attr->la_valid &= ~LA_CTIME;
2232         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2233                 attr->la_valid &= ~LA_MTIME;
2234
2235         result = osd_quota_transfer(inode, attr);
2236         if (result)
2237                 return;
2238
2239         if (attr->la_valid != 0) {
2240                 result = osd_inode_setattr(info->oti_env, inode, attr);
2241                 /*
2242                  * The osd_inode_setattr() should always succeed here.  The
2243                  * only error that could be returned is EDQUOT when we are
2244                  * trying to change the UID or GID of the inode. However, this
2245                  * should not happen since quota enforcement is no longer
2246                  * enabled on ldiskfs (lquota takes care of it).
2247                  */
2248                 LASSERTF(result == 0, "%d", result);
2249                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2250         }
2251
2252         attr->la_valid = valid;
2253 }
2254
2255 /**
2256  * Helper function for osd_object_create()
2257  *
2258  * \retval 0, on success
2259  */
2260 static int __osd_object_create(struct osd_thread_info *info,
2261                                struct osd_object *obj, struct lu_attr *attr,
2262                                struct dt_allocation_hint *hint,
2263                                struct dt_object_format *dof,
2264                                struct thandle *th)
2265 {
2266         int     result;
2267         __u32   umask;
2268
2269         /* we drop umask so that permissions we pass are not affected */
2270         umask = current->fs->umask;
2271         current->fs->umask = 0;
2272
2273         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2274                                                   th);
2275         if (result == 0) {
2276                 osd_attr_init(info, obj, attr, dof);
2277                 osd_object_init0(obj);
2278                 /* bz 24037 */
2279                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2280                         unlock_new_inode(obj->oo_inode);
2281         }
2282
2283         /* restore previous umask value */
2284         current->fs->umask = umask;
2285
2286         return result;
2287 }
2288
2289 /**
2290  * Helper function for osd_object_create()
2291  *
2292  * \retval 0, on success
2293  */
2294 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2295                            const struct lu_fid *fid, struct thandle *th)
2296 {
2297         struct osd_thread_info *info = osd_oti_get(env);
2298         struct osd_inode_id    *id   = &info->oti_id;
2299         struct osd_device      *osd  = osd_obj2dev(obj);
2300         struct osd_thandle     *oh;
2301
2302         LASSERT(obj->oo_inode != NULL);
2303
2304         oh = container_of0(th, struct osd_thandle, ot_super);
2305         LASSERT(oh->ot_handle);
2306
2307         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2308         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2309 }
2310
2311 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2312                    obd_seq seq, struct lu_seq_range *range)
2313 {
2314         struct seq_server_site  *ss = osd_seq_site(osd);
2315
2316         if (fid_seq_is_idif(seq)) {
2317                 fld_range_set_ost(range);
2318                 range->lsr_index = idif_ost_idx(seq);
2319                 return 0;
2320         }
2321
2322         if (!fid_seq_in_fldb(seq)) {
2323                 fld_range_set_mdt(range);
2324                 if (ss != NULL)
2325                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2326                          * at all */
2327                         range->lsr_index = ss->ss_node_id;
2328                 return 0;
2329         }
2330
2331         LASSERT(ss != NULL);
2332         fld_range_set_any(range);
2333         /* OSD will only do local fld lookup */
2334         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2335 }
2336
2337 /*
2338  * Concurrency: no external locking is necessary.
2339  */
2340 static int osd_declare_object_create(const struct lu_env *env,
2341                                      struct dt_object *dt,
2342                                      struct lu_attr *attr,
2343                                      struct dt_allocation_hint *hint,
2344                                      struct dt_object_format *dof,
2345                                      struct thandle *handle)
2346 {
2347         struct osd_thandle      *oh;
2348         int                      rc;
2349         ENTRY;
2350
2351         LASSERT(handle != NULL);
2352
2353         oh = container_of0(handle, struct osd_thandle, ot_super);
2354         LASSERT(oh->ot_handle == NULL);
2355
2356         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2357                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2358         /* Reuse idle OI block may cause additional one OI block
2359          * to be changed. */
2360         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2361                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2362
2363         /* If this is directory, then we expect . and .. to be inserted as
2364          * well. The one directory block always needs to be created for the
2365          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2366          * block), there is no danger of needing a tree for the first block.
2367          */
2368         if (attr && S_ISDIR(attr->la_mode)) {
2369                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2370                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2371                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2372         }
2373
2374         if (!attr)
2375                 RETURN(0);
2376
2377         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2378                                    false, false, NULL, false);
2379         if (rc != 0)
2380                 RETURN(rc);
2381
2382         RETURN(rc);
2383 }
2384
2385 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2386                              struct lu_attr *attr,
2387                              struct dt_allocation_hint *hint,
2388                              struct dt_object_format *dof,
2389                              struct thandle *th)
2390 {
2391         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2392         struct osd_object      *obj    = osd_dt_obj(dt);
2393         struct osd_thread_info *info   = osd_oti_get(env);
2394         int result;
2395
2396         ENTRY;
2397
2398         LINVRNT(osd_invariant(obj));
2399         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2400         LASSERT(osd_write_locked(env, obj));
2401         LASSERT(th != NULL);
2402
2403         if (unlikely(fid_is_acct(fid)))
2404                 /* Quota files can't be created from the kernel any more,
2405                  * 'tune2fs -O quota' will take care of creating them */
2406                 RETURN(-EPERM);
2407
2408         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2409         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2410
2411         result = __osd_object_create(info, obj, attr, hint, dof, th);
2412         if (result == 0)
2413                 result = __osd_oi_insert(env, obj, fid, th);
2414
2415         LASSERT(ergo(result == 0,
2416                      dt_object_exists(dt) && !dt_object_remote(dt)));
2417
2418         LASSERT(osd_invariant(obj));
2419         RETURN(result);
2420 }
2421
2422 /**
2423  * Called to destroy on-disk representation of the object
2424  *
2425  * Concurrency: must be locked
2426  */
2427 static int osd_declare_object_destroy(const struct lu_env *env,
2428                                       struct dt_object *dt,
2429                                       struct thandle *th)
2430 {
2431         struct osd_object  *obj = osd_dt_obj(dt);
2432         struct inode       *inode = obj->oo_inode;
2433         struct osd_thandle *oh;
2434         int                 rc;
2435         ENTRY;
2436
2437         oh = container_of0(th, struct osd_thandle, ot_super);
2438         LASSERT(oh->ot_handle == NULL);
2439         LASSERT(inode);
2440
2441         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2442                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2443         /* Recycle idle OI leaf may cause additional three OI blocks
2444          * to be changed. */
2445         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2446                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2447         /* one less inode */
2448         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2449                                    -1, oh, false, true, NULL, false);
2450         if (rc)
2451                 RETURN(rc);
2452         /* data to be truncated */
2453         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2454                                    0, oh, true, true, NULL, false);
2455         RETURN(rc);
2456 }
2457
2458 static int osd_object_destroy(const struct lu_env *env,
2459                               struct dt_object *dt,
2460                               struct thandle *th)
2461 {
2462         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2463         struct osd_object      *obj = osd_dt_obj(dt);
2464         struct inode           *inode = obj->oo_inode;
2465         struct osd_device      *osd = osd_obj2dev(obj);
2466         struct osd_thandle     *oh;
2467         int                     result;
2468         ENTRY;
2469
2470         oh = container_of0(th, struct osd_thandle, ot_super);
2471         LASSERT(oh->ot_handle);
2472         LASSERT(inode);
2473         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2474
2475         if (unlikely(fid_is_acct(fid)))
2476                 RETURN(-EPERM);
2477
2478         if (S_ISDIR(inode->i_mode)) {
2479                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2480                         inode->i_nlink == 2);
2481                 /* it will check/delete the inode from remote parent,
2482                  * how to optimize it? unlink performance impaction XXX */
2483                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2484                 if (result != 0 && result != -ENOENT) {
2485                         CERROR("%s: delete inode "DFID": rc = %d\n",
2486                                osd_name(osd), PFID(fid), result);
2487                 }
2488                 spin_lock(&obj->oo_guard);
2489                 clear_nlink(inode);
2490                 spin_unlock(&obj->oo_guard);
2491                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2492         }
2493
2494         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2495
2496         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2497                                OI_CHECK_FLD);
2498
2499         /* XXX: add to ext3 orphan list */
2500         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2501
2502         /* not needed in the cache anymore */
2503         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2504
2505         RETURN(0);
2506 }
2507
2508 /**
2509  * Put the fid into lustre_mdt_attrs, and then place the structure
2510  * inode's ea. This fid should not be altered during the life time
2511  * of the inode.
2512  *
2513  * \retval +ve, on success
2514  * \retval -ve, on error
2515  *
2516  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2517  */
2518 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2519                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2520 {
2521         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2522         int                      rc;
2523         ENTRY;
2524
2525         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2526                 RETURN(0);
2527
2528         lustre_lma_init(lma, fid, compat, incompat);
2529         lustre_lma_swab(lma);
2530
2531         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2532                              XATTR_CREATE);
2533         /* LMA may already exist, but we need to check that all the
2534          * desired compat/incompat flags have been added. */
2535         if (unlikely(rc == -EEXIST)) {
2536                 if (compat == 0 && incompat == 0)
2537                         RETURN(0);
2538
2539                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2540                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2541                                      LMA_OLD_SIZE);
2542                 if (rc <= 0)
2543                         RETURN(-EINVAL);
2544
2545                 lustre_lma_swab(lma);
2546                 if (!(~lma->lma_compat & compat) &&
2547                     !(~lma->lma_incompat & incompat))
2548                         RETURN(0);
2549
2550                 lma->lma_compat |= compat;
2551                 lma->lma_incompat |= incompat;
2552                 lustre_lma_swab(lma);
2553                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2554                                      sizeof(*lma), XATTR_REPLACE);
2555         }
2556
2557         RETURN(rc);
2558 }
2559
2560 /**
2561  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2562  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2563  * To have compatilibility with 1.8 ldiskfs driver we need to have
2564  * magic number at start of fid data.
2565  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2566  * its inmemory API.
2567  */
2568 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2569                                   const struct dt_rec *fid)
2570 {
2571         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2572             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2573                 param->edp_magic = 0;
2574                 return;
2575         }
2576
2577         param->edp_magic = LDISKFS_LUFID_MAGIC;
2578         param->edp_len =  sizeof(struct lu_fid) + 1;
2579         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2580 }
2581
2582 /**
2583  * Try to read the fid from inode ea into dt_rec.
2584  *
2585  * \param fid object fid.
2586  *
2587  * \retval 0 on success
2588  */
2589 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2590                           __u32 ino, struct lu_fid *fid,
2591                           struct osd_inode_id *id)
2592 {
2593         struct osd_thread_info *info  = osd_oti_get(env);
2594         struct inode           *inode;
2595         ENTRY;
2596
2597         osd_id_gen(id, ino, OSD_OII_NOGEN);
2598         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2599         if (IS_ERR(inode))
2600                 RETURN(PTR_ERR(inode));
2601
2602         iput(inode);
2603         RETURN(0);
2604 }
2605
2606 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2607                                         struct inode *dir,
2608                                         struct inode  *parent_dir,
2609                                         const struct dt_rec *dot_fid,
2610                                         const struct dt_rec *dot_dot_fid,
2611                                         struct osd_thandle *oth)
2612 {
2613         struct ldiskfs_dentry_param *dot_ldp;
2614         struct ldiskfs_dentry_param *dot_dot_ldp;
2615
2616         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2617         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2618
2619         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2620         dot_ldp->edp_magic = 0;
2621         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2622                                         dir, dot_ldp, dot_dot_ldp);
2623 }
2624
2625 /**
2626  * Create an local agent inode for remote entry
2627  */
2628 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2629                                                   struct osd_device *osd,
2630                                                   struct osd_object *pobj,
2631                                                   const struct lu_fid *fid,
2632                                                   struct thandle *th)
2633 {
2634         struct osd_thread_info  *info = osd_oti_get(env);
2635         struct inode            *local;
2636         struct osd_thandle      *oh;
2637         int                     rc;
2638         ENTRY;
2639
2640         LASSERT(th);
2641         oh = container_of(th, struct osd_thandle, ot_super);
2642         LASSERT(oh->ot_handle->h_transaction != NULL);
2643
2644         /* FIXME: Insert index api needs to know the mode of
2645          * the remote object. Just use S_IFDIR for now */
2646         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2647         if (IS_ERR(local)) {
2648                 CERROR("%s: create local error %d\n", osd_name(osd),
2649                        (int)PTR_ERR(local));
2650                 RETURN(local);
2651         }
2652
2653         /* Set special LMA flag for local agent inode */
2654         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2655         if (rc != 0) {
2656                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2657                        osd_name(osd), PFID(fid), rc);
2658                 RETURN(ERR_PTR(rc));
2659         }
2660
2661         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2662                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2663                 (const struct dt_rec *)fid, oh);
2664         if (rc != 0) {
2665                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2666                         osd_name(osd), PFID(fid), rc);
2667                 RETURN(ERR_PTR(rc));
2668         }
2669
2670         RETURN(local);
2671 }
2672
2673 /**
2674  * Delete local agent inode for remote entry
2675  */
2676 static int osd_delete_local_agent_inode(const struct lu_env *env,
2677                                         struct osd_device *osd,
2678                                         const struct lu_fid *fid,
2679                                         __u32 ino, struct osd_thandle *oh)
2680 {
2681         struct osd_thread_info  *oti = osd_oti_get(env);
2682         struct osd_inode_id     *id = &oti->oti_id;
2683         struct inode            *inode;
2684         ENTRY;
2685
2686         id->oii_ino = le32_to_cpu(ino);
2687         id->oii_gen = OSD_OII_NOGEN;
2688         inode = osd_iget(oti, osd, id);
2689         if (IS_ERR(inode)) {
2690                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2691                        PFID(fid), id->oii_ino, id->oii_gen);
2692                 RETURN(PTR_ERR(inode));
2693         }
2694
2695         clear_nlink(inode);
2696         mark_inode_dirty(inode);
2697         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2698                 osd_name(osd), PFID(fid), inode->i_ino);
2699         iput(inode);
2700         RETURN(0);
2701 }
2702
2703 /**
2704  * OSD layer object create function for interoperability mode (b11826).
2705  * This is mostly similar to osd_object_create(). Only difference being, fid is
2706  * inserted into inode ea here.
2707  *
2708  * \retval   0, on success
2709  * \retval -ve, on error
2710  */
2711 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2712                                 struct lu_attr *attr,
2713                                 struct dt_allocation_hint *hint,
2714                                 struct dt_object_format *dof,
2715                                 struct thandle *th)
2716 {
2717         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2718         struct osd_object      *obj    = osd_dt_obj(dt);
2719         struct osd_thread_info *info   = osd_oti_get(env);
2720         int                     result;
2721
2722         ENTRY;
2723
2724         LASSERT(osd_invariant(obj));
2725         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2726         LASSERT(osd_write_locked(env, obj));
2727         LASSERT(th != NULL);
2728
2729         if (unlikely(fid_is_acct(fid)))
2730                 /* Quota files can't be created from the kernel any more,
2731                  * 'tune2fs -O quota' will take care of creating them */
2732                 RETURN(-EPERM);
2733
2734         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2735         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2736
2737         result = __osd_object_create(info, obj, attr, hint, dof, th);
2738         if (result == 0)
2739                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2740                                 fid_is_on_ost(info, osd_obj2dev(obj),
2741                                               fid, OI_CHECK_FLD) ?
2742                                 LMAC_FID_ON_OST : 0, 0);
2743
2744         if (result == 0)
2745                 result = __osd_oi_insert(env, obj, fid, th);
2746
2747         LASSERT(ergo(result == 0,
2748                      dt_object_exists(dt) && !dt_object_remote(dt)));
2749         LINVRNT(osd_invariant(obj));
2750         RETURN(result);
2751 }
2752
2753 static int osd_declare_object_ref_add(const struct lu_env *env,
2754                                       struct dt_object *dt,
2755                                       struct thandle *handle)
2756 {
2757         struct osd_thandle       *oh;
2758
2759         /* it's possible that object doesn't exist yet */
2760         LASSERT(handle != NULL);
2761
2762         oh = container_of0(handle, struct osd_thandle, ot_super);
2763         LASSERT(oh->ot_handle == NULL);
2764
2765         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2766                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2767
2768         return 0;
2769 }
2770
2771 /*
2772  * Concurrency: @dt is write locked.
2773  */
2774 static int osd_object_ref_add(const struct lu_env *env,
2775                               struct dt_object *dt, struct thandle *th)
2776 {
2777         struct osd_object  *obj = osd_dt_obj(dt);
2778         struct inode       *inode = obj->oo_inode;
2779         struct osd_thandle *oh;
2780         int                 rc = 0;
2781
2782         LINVRNT(osd_invariant(obj));
2783         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2784         LASSERT(osd_write_locked(env, obj));
2785         LASSERT(th != NULL);
2786
2787         oh = container_of0(th, struct osd_thandle, ot_super);
2788         LASSERT(oh->ot_handle != NULL);
2789
2790         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2791
2792         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2793                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2794         /*
2795          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2796          * (65000) subdirectories by storing "1" in i_nlink if the link count
2797          * would otherwise overflow. Directory tranversal tools understand
2798          * that (st_nlink == 1) indicates that the filesystem dose not track
2799          * hard links count on the directory, and will not abort subdirectory
2800          * scanning early once (st_nlink - 2) subdirs have been found.
2801          *
2802          * This also has to properly handle the case of inodes with nlink == 0
2803          * in case they are being linked into the PENDING directory
2804          */
2805         spin_lock(&obj->oo_guard);
2806         ldiskfs_inc_count(oh->ot_handle, inode);
2807         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2808         spin_unlock(&obj->oo_guard);
2809
2810         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2811         LINVRNT(osd_invariant(obj));
2812
2813         return rc;
2814 }
2815
2816 static int osd_declare_object_ref_del(const struct lu_env *env,
2817                                       struct dt_object *dt,
2818                                       struct thandle *handle)
2819 {
2820         struct osd_thandle *oh;
2821
2822         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2823         LASSERT(handle != NULL);
2824
2825         oh = container_of0(handle, struct osd_thandle, ot_super);
2826         LASSERT(oh->ot_handle == NULL);
2827
2828         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2829                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2830
2831         return 0;
2832 }
2833
2834 /*
2835  * Concurrency: @dt is write locked.
2836  */
2837 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2838                               struct thandle *th)
2839 {
2840         struct osd_object       *obj = osd_dt_obj(dt);
2841         struct inode            *inode = obj->oo_inode;
2842         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2843         struct osd_thandle      *oh;
2844
2845         LINVRNT(osd_invariant(obj));
2846         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2847         LASSERT(osd_write_locked(env, obj));
2848         LASSERT(th != NULL);
2849
2850         oh = container_of0(th, struct osd_thandle, ot_super);
2851         LASSERT(oh->ot_handle != NULL);
2852
2853         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2854
2855         spin_lock(&obj->oo_guard);
2856         /* That can be result of upgrade from old Lustre version and
2857          * applied only to local files.  Just skip this ref_del call.
2858          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2859         if (inode->i_nlink == 0) {
2860                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2861                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2862                              ", maybe an upgraded file? (LU-3915)\n",
2863                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2864                 spin_unlock(&obj->oo_guard);
2865                 return 0;
2866         }
2867
2868         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2869                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2870
2871         ldiskfs_dec_count(oh->ot_handle, inode);
2872         spin_unlock(&obj->oo_guard);
2873
2874         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2875         LINVRNT(osd_invariant(obj));
2876
2877         return 0;
2878 }
2879
2880 /*
2881  * Get the 64-bit version for an inode.
2882  */
2883 static int osd_object_version_get(const struct lu_env *env,
2884                                   struct dt_object *dt, dt_obj_version_t *ver)
2885 {
2886         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2887
2888         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2889                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2890         *ver = LDISKFS_I(inode)->i_fs_version;
2891         return 0;
2892 }
2893
2894 /*
2895  * Concurrency: @dt is read locked.
2896  */
2897 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2898                          struct lu_buf *buf, const char *name,
2899                          struct lustre_capa *capa)
2900 {
2901         struct osd_object      *obj    = osd_dt_obj(dt);
2902         struct inode           *inode  = obj->oo_inode;
2903         struct osd_thread_info *info   = osd_oti_get(env);
2904         struct dentry          *dentry = &info->oti_obj_dentry;
2905
2906         /* version get is not real XATTR but uses xattr API */
2907         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2908                 /* for version we are just using xattr API but change inode
2909                  * field instead */
2910                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2911                 osd_object_version_get(env, dt, buf->lb_buf);
2912                 return sizeof(dt_obj_version_t);
2913         }
2914
2915         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2916         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2917
2918         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2919                 return -EACCES;
2920
2921         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2922 }
2923
2924
2925 static int osd_declare_xattr_set(const struct lu_env *env,
2926                                  struct dt_object *dt,
2927                                  const struct lu_buf *buf, const char *name,
2928                                  int fl, struct thandle *handle)
2929 {
2930         struct osd_thandle *oh;
2931         int credits;
2932
2933         LASSERT(handle != NULL);
2934
2935         oh = container_of0(handle, struct osd_thandle, ot_super);
2936         LASSERT(oh->ot_handle == NULL);
2937
2938         /* optimistic optimization: LMA is set first and usually fit inode */
2939         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2940                 if (dt_object_exists(dt))
2941                         credits = 0;
2942                 else
2943                         credits = 1;
2944         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2945                 credits = 1;
2946         } else {
2947                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2948                 struct super_block *sb = osd_sb(osd);
2949                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2950                 if (buf && buf->lb_len > sb->s_blocksize) {
2951                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2952                                         sb->s_blocksize_bits;
2953                 }
2954         }
2955
2956         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2957
2958         return 0;
2959 }
2960
2961 /*
2962  * Set the 64-bit version for object
2963  */
2964 static void osd_object_version_set(const struct lu_env *env,
2965                                    struct dt_object *dt,
2966                                    dt_obj_version_t *new_version)
2967 {
2968         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2969
2970         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2971                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2972
2973         LDISKFS_I(inode)->i_fs_version = *new_version;
2974         /** Version is set after all inode operations are finished,
2975          *  so we should mark it dirty here */
2976         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2977 }
2978
2979 /*
2980  * Concurrency: @dt is write locked.
2981  */
2982 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2983                          const struct lu_buf *buf, const char *name, int fl,
2984                          struct thandle *handle, struct lustre_capa *capa)
2985 {
2986         struct osd_object      *obj      = osd_dt_obj(dt);
2987         struct inode           *inode    = obj->oo_inode;
2988         struct osd_thread_info *info     = osd_oti_get(env);
2989         int                     fs_flags = 0;
2990         ENTRY;
2991
2992         LASSERT(handle != NULL);
2993
2994         /* version set is not real XATTR */
2995         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2996                 /* for version we are just using xattr API but change inode
2997                  * field instead */
2998                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2999                 osd_object_version_set(env, dt, buf->lb_buf);
3000                 return sizeof(dt_obj_version_t);
3001         }
3002
3003         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3004                 return -EACCES;
3005
3006         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3007                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3008
3009         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3010         if (fl & LU_XATTR_REPLACE)
3011                 fs_flags |= XATTR_REPLACE;
3012
3013         if (fl & LU_XATTR_CREATE)
3014                 fs_flags |= XATTR_CREATE;
3015
3016         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3017                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3018                 int                      rc;
3019
3020                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3021                 if (rc != 0)
3022                         RETURN(rc);
3023
3024                 lma->lma_incompat |= LMAI_STRIPED;
3025                 lustre_lma_swab(lma);
3026                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3027                                      sizeof(*lma), XATTR_REPLACE);
3028                 if (rc != 0)
3029                         RETURN(rc);
3030         }
3031
3032         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3033                                fs_flags);
3034 }
3035
3036 /*
3037  * Concurrency: @dt is read locked.
3038  */
3039 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3040                           struct lu_buf *buf, struct lustre_capa *capa)
3041 {
3042         struct osd_object      *obj    = osd_dt_obj(dt);
3043         struct inode           *inode  = obj->oo_inode;
3044         struct osd_thread_info *info   = osd_oti_get(env);
3045         struct dentry          *dentry = &info->oti_obj_dentry;
3046
3047         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3048         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3049
3050         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3051                 return -EACCES;
3052
3053         dentry->d_inode = inode;
3054         dentry->d_sb = inode->i_sb;
3055         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3056 }
3057
3058 static int osd_declare_xattr_del(const struct lu_env *env,
3059                                  struct dt_object *dt, const char *name,
3060                                  struct thandle *handle)
3061 {
3062         struct osd_thandle *oh;
3063
3064         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3065         LASSERT(handle != NULL);
3066
3067         oh = container_of0(handle, struct osd_thandle, ot_super);
3068         LASSERT(oh->ot_handle == NULL);
3069
3070         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3071                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3072
3073         return 0;
3074 }
3075
3076 /*
3077  * Concurrency: @dt is write locked.
3078  */
3079 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3080                          const char *name, struct thandle *handle,
3081                          struct lustre_capa *capa)
3082 {
3083         struct osd_object      *obj    = osd_dt_obj(dt);
3084         struct inode           *inode  = obj->oo_inode;
3085         struct osd_thread_info *info   = osd_oti_get(env);
3086         struct dentry          *dentry = &info->oti_obj_dentry;
3087         int                     rc;
3088
3089         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3090         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3091         LASSERT(handle != NULL);
3092
3093         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3094                 return -EACCES;
3095
3096         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3097
3098         ll_vfs_dq_init(inode);
3099         dentry->d_inode = inode;
3100         dentry->d_sb = inode->i_sb;
3101         rc = inode->i_op->removexattr(dentry, name);
3102         return rc;
3103 }
3104
3105 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3106                                      struct dt_object *dt,
3107                                      struct lustre_capa *old, __u64 opc)
3108 {
3109         struct osd_thread_info *info = osd_oti_get(env);
3110         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3111         struct osd_object *obj = osd_dt_obj(dt);
3112         struct osd_device *osd = osd_obj2dev(obj);
3113         struct lustre_capa_key *key = &info->oti_capa_key;
3114         struct lustre_capa *capa = &info->oti_capa;
3115         struct obd_capa *oc;
3116         struct lu_capainfo *lci;
3117         int rc;
3118         ENTRY;
3119
3120         if (!osd->od_fl_capa)
3121                 RETURN(ERR_PTR(-ENOENT));
3122
3123         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3124         LINVRNT(osd_invariant(obj));
3125
3126         /* renewal sanity check */
3127         if (old && osd_object_auth(env, dt, old, opc))
3128                 RETURN(ERR_PTR(-EACCES));
3129
3130         lci = lu_capainfo_get(env);
3131         if (unlikely(lci == NULL))
3132                 RETURN(ERR_PTR(-ENOENT));
3133
3134         switch (lci->lci_auth) {
3135         case LC_ID_NONE:
3136                 RETURN(NULL);
3137         case LC_ID_PLAIN:
3138                 capa->lc_uid = i_uid_read(obj->oo_inode);
3139                 capa->lc_gid = i_gid_read(obj->oo_inode);
3140                 capa->lc_flags = LC_ID_PLAIN;
3141                 break;
3142         case LC_ID_CONVERT: {
3143                 __u32 d[4], s[4];
3144
3145                 s[0] = i_uid_read(obj->oo_inode);
3146                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3147                 s[2] = i_uid_read(obj->oo_inode);
3148                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3149                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3150                 if (unlikely(rc))
3151                         RETURN(ERR_PTR(rc));
3152
3153                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3154                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3155                 capa->lc_flags = LC_ID_CONVERT;
3156                 break;
3157         }
3158         default:
3159                 RETURN(ERR_PTR(-EINVAL));
3160         }
3161
3162         capa->lc_fid = *fid;
3163         capa->lc_opc = opc;
3164         capa->lc_flags |= osd->od_capa_alg << 24;
3165         capa->lc_timeout = osd->od_capa_timeout;
3166         capa->lc_expiry = 0;
3167
3168         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3169         if (oc) {
3170                 LASSERT(!capa_is_expired(oc));
3171                 RETURN(oc);
3172         }
3173
3174         spin_lock(&capa_lock);
3175         *key = osd->od_capa_keys[1];