lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51
  52 /* LUSTRE_VERSION_CODE */
  53 #include <lustre_ver.h>
  54 /* prerequisite for linux/xattr.h */
  55 #include <linux/types.h>
  56 /* prerequisite for linux/xattr.h */
  57 #include <linux/fs.h>
  58 /* XATTR_{REPLACE,CREATE} */
  59 #include <linux/xattr.h>
  60
  61 #include <ldiskfs/ldiskfs.h>
  62 #include <ldiskfs/xattr.h>
  63 #undef ENTRY
  64 /*
  65  * struct OBD_{ALLOC,FREE}*()
  66  * OBD_FAIL_CHECK
  67  */
  68 #include <obd_support.h>
  69 /* struct ptlrpc_thread */
  70 #include <lustre_net.h>
  71 #include <lustre_fid.h>
  72 /* process_config */
  73 #include <lustre_param.h>
  74
  75 #include "osd_internal.h"
  76 #include "osd_dynlocks.h"
  77
  78 /* llo_* api support */
  79 #include <md_object.h>
  80 #include <lustre_quota.h>
  81
  82 #include <lustre_linkea.h>
  83
  84 int ldiskfs_pdo = 1;
  85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  86                 "ldiskfs with parallel directory operations");
  87
  88 int ldiskfs_track_declares_assert;
  89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  90                 "LBUG during tracking of declares");
  91
  92 /* Slab to allocate dynlocks */
  93 struct kmem_cache *dynlock_cachep;
  94
  95 /* Slab to allocate osd_it_ea */
  96 struct kmem_cache *osd_itea_cachep;
  97
  98 static struct lu_kmem_descr ldiskfs_caches[] = {
  99         {
 100                 .ckd_cache = &dynlock_cachep,
 101                 .ckd_name  = "dynlock_cache",
 102                 .ckd_size  = sizeof(struct dynlock_handle)
 103         },
 104         {
 105                 .ckd_cache = &osd_itea_cachep,
 106                 .ckd_name  = "osd_itea_cache",
 107                 .ckd_size  = sizeof(struct osd_it_ea)
 108         },
 109         {
 110                 .ckd_cache = NULL
 111         }
 112 };
 113
 114 static const char dot[] = ".";
 115 static const char dotdot[] = "..";
 116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 117
 118 static const struct lu_object_operations      osd_lu_obj_ops;
 119 static const struct dt_object_operations      osd_obj_ops;
 120 static const struct dt_object_operations      osd_obj_ea_ops;
 121 static const struct dt_object_operations      osd_obj_otable_it_ops;
 122 static const struct dt_index_operations       osd_index_iam_ops;
 123 static const struct dt_index_operations       osd_index_ea_ops;
 124
 125 int osd_trans_declare_op2rb[] = {
 126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 137 };
 138
 139 static int osd_has_index(const struct osd_object *obj)
 140 {
 141         return obj->oo_dt.do_index_ops != NULL;
 142 }
 143
 144 static int osd_object_invariant(const struct lu_object *l)
 145 {
 146         return osd_invariant(osd_obj(l));
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152
 153 /*
 154  * Concurrency: doesn't matter
 155  */
 156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 157 {
 158         struct osd_thread_info *oti = osd_oti_get(env);
 159         return oti->oti_w_locks > 0 && o->oo_owner == env;
 160 }
 161
 162 /*
 163  * Concurrency: doesn't access mutable data
 164  */
 165 static int osd_root_get(const struct lu_env *env,
 166                         struct dt_device *dev, struct lu_fid *f)
 167 {
 168         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 169         return 0;
 170 }
 171
 172 /*
 173  * OSD object methods.
 174  */
 175
 176 /*
 177  * Concurrency: no concurrent access is possible that early in object
 178  * life-cycle.
 179  */
 180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 181                                           const struct lu_object_header *hdr,
 182                                           struct lu_device *d)
 183 {
 184         struct osd_object *mo;
 185
 186         OBD_ALLOC_PTR(mo);
 187         if (mo != NULL) {
 188                 struct lu_object *l;
 189
 190                 l = &mo->oo_dt.do_lu;
 191                 dt_object_init(&mo->oo_dt, NULL, d);
 192                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 193                 l->lo_ops = &osd_lu_obj_ops;
 194                 init_rwsem(&mo->oo_sem);
 195                 init_rwsem(&mo->oo_ext_idx_sem);
 196                 spin_lock_init(&mo->oo_guard);
 197                 return l;
 198         } else {
 199                 return NULL;
 200         }
 201 }
 202
 203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 204                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 205 {
 206         int rc;
 207
 208         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 209         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 210                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 211         if (rc > 0) {
 212                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 213                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 214                 rc = 0;
 215                 lustre_lma_swab(lma);
 216                 /* Check LMA compatibility */
 217                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 218                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 219                               "for fid = "DFID", ino = %lu\n",
 220                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 221                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 222                               PFID(&lma->lma_self_fid), inode->i_ino);
 223                         rc = -EOPNOTSUPP;
 224                 }
 225         } else if (rc == 0) {
 226                 rc = -ENODATA;
 227         }
 228
 229         return rc;
 230 }
 231
 232 /*
 233  * retrieve object from backend ext fs.
 234  **/
 235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 236                        struct osd_inode_id *id)
 237 {
 238         struct inode *inode = NULL;
 239
 240         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 241         if (IS_ERR(inode)) {
 242                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 243                        id->oii_ino, PTR_ERR(inode));
 244         } else if (id->oii_gen != OSD_OII_NOGEN &&
 245                    inode->i_generation != id->oii_gen) {
 246                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 247                        "i_generation = %u\n",
 248                        id->oii_ino, id->oii_gen, inode->i_generation);
 249                 iput(inode);
 250                 inode = ERR_PTR(-ESTALE);
 251         } else if (inode->i_nlink == 0) {
 252                 /* due to parallel readdir and unlink,
 253                 * we can have dead inode here. */
 254                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 255                 iput(inode);
 256                 inode = ERR_PTR(-ESTALE);
 257         } else if (is_bad_inode(inode)) {
 258                 CWARN("%.16s: bad inode: ino = %u\n",
 259                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 260                 iput(inode);
 261                 inode = ERR_PTR(-ENOENT);
 262         } else {
 263                 if (id->oii_gen == OSD_OII_NOGEN)
 264                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 265
 266                 /* Do not update file c/mtime in ldiskfs.
 267                  * NB: we don't have any lock to protect this because we don't
 268                  * have reference on osd_object now, but contention with
 269                  * another lookup + attr_set can't happen in the tiny window
 270                  * between if (...) and set S_NOCMTIME. */
 271                 if (!(inode->i_flags & S_NOCMTIME))
 272                         inode->i_flags |= S_NOCMTIME;
 273         }
 274         return inode;
 275 }
 276
 277 static struct inode *
 278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 279              struct osd_inode_id *id, struct lu_fid *fid)
 280 {
 281         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 282         struct inode            *inode;
 283         int                      rc;
 284
 285         inode = osd_iget(info, dev, id);
 286         if (IS_ERR(inode))
 287                 return inode;
 288
 289         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 290         if (rc == 0) {
 291                 *fid = lma->lma_self_fid;
 292         } else if (rc == -ENODATA) {
 293                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 294                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 295                 else
 296                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 297         } else {
 298                 iput(inode);
 299                 inode = ERR_PTR(rc);
 300         }
 301         return inode;
 302 }
 303
 304 static struct inode *osd_iget_check(struct osd_thread_info *info,
 305                                     struct osd_device *dev,
 306                                     const struct lu_fid *fid,
 307                                     struct osd_inode_id *id,
 308                                     bool in_oi)
 309 {
 310         struct inode    *inode;
 311         int              rc     = 0;
 312         ENTRY;
 313
 314         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 315         if (IS_ERR(inode)) {
 316                 rc = PTR_ERR(inode);
 317                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 318                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 319                                id->oii_ino, rc);
 320
 321                         GOTO(put, rc);
 322                 }
 323
 324                 goto check_oi;
 325         }
 326
 327         if (is_bad_inode(inode)) {
 328                 rc = -ENOENT;
 329                 if (!in_oi) {
 330                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 331
 332                         GOTO(put, rc);
 333                 }
 334
 335                 goto check_oi;
 336         }
 337
 338         if (id->oii_gen != OSD_OII_NOGEN &&
 339             inode->i_generation != id->oii_gen) {
 340                 rc = -ESTALE;
 341                 if (!in_oi) {
 342                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 343                                "oii_gen = %u, i_generation = %u\n",
 344                                id->oii_ino, id->oii_gen, inode->i_generation);
 345
 346                         GOTO(put, rc);
 347                 }
 348
 349                 goto check_oi;
 350         }
 351
 352         if (inode->i_nlink == 0) {
 353                 rc = -ENOENT;
 354                 if (!in_oi) {
 355                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 356
 357                         GOTO(put, rc);
 358                 }
 359
 360                 goto check_oi;
 361         }
 362
 363 check_oi:
 364         if (rc != 0) {
 365                 struct osd_inode_id saved_id = *id;
 366
 367                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 368
 369                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 370                 /* XXX: There are some possible cases:
 371                  *      1. rc = 0.
 372                  *         Backup/restore caused the OI invalid.
 373                  *      2. rc = 0.
 374                  *         Someone unlinked the object but NOT removed
 375                  *         the OI mapping, such as mount target device
 376                  *         as ldiskfs, and modify something directly.
 377                  *      3. rc = -ENOENT.
 378                  *         Someone just removed the object between the
 379                  *         former oi_lookup and the iget. It is normal.
 380                  *      4. Other failure cases.
 381                  *
 382                  *      Generally, when the device is mounted, it will
 383                  *      auto check whether the system is restored from
 384                  *      file-level backup or not. We trust such detect
 385                  *      to distinguish the 1st case from the 2nd case. */
 386                 if (rc == 0) {
 387                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 388                             inode->i_generation == id->oii_gen) {
 389                                 rc = -ENOENT;
 390                         } else {
 391                                 __u32 level = D_LFSCK;
 392
 393                                 rc = -EREMCHG;
 394                                 if (!thread_is_running(&dev->od_scrub.os_thread))
 395                                         level |= D_CONSOLE;
 396
 397                                 CDEBUG(level, "%s: the OI mapping for the FID "
 398                                        DFID" become inconsistent, the given ID "
 399                                        "%u/%u, the ID in OI mapping %u/%u\n",
 400                                        osd_name(dev), PFID(fid),
 401                                        saved_id.oii_ino, saved_id.oii_gen,
 402                                        id->oii_ino, id->oii_ino);
 403                         }
 404                 }
 405         } else {
 406                 if (id->oii_gen == OSD_OII_NOGEN)
 407                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 408
 409                 /* Do not update file c/mtime in ldiskfs.
 410                  * NB: we don't have any lock to protect this because we don't
 411                  * have reference on osd_object now, but contention with
 412                  * another lookup + attr_set can't happen in the tiny window
 413                  * between if (...) and set S_NOCMTIME. */
 414                 if (!(inode->i_flags & S_NOCMTIME))
 415                         inode->i_flags |= S_NOCMTIME;
 416         }
 417
 418         GOTO(put, rc);
 419
 420 put:
 421         if (rc != 0) {
 422                 if (!IS_ERR(inode))
 423                         iput(inode);
 424
 425                 inode = ERR_PTR(rc);
 426         }
 427
 428         return inode;
 429 }
 430
 431 /**
 432  * \retval +v: new filter_fid, does not contain self-fid
 433  * \retval 0:  filter_fid_old, contains self-fid
 434  * \retval -v: other failure cases
 435  */
 436 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 437                  struct dentry *dentry, struct lu_fid *fid)
 438 {
 439         struct filter_fid_old   *ff     = &info->oti_ff;
 440         struct ost_id           *ostid  = &info->oti_ostid;
 441         int                      rc;
 442
 443         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 444         if (rc == sizeof(*ff)) {
 445                 rc = 0;
 446                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 447                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 448                 /* XXX: should use real OST index in the future. LU-3569 */
 449                 ostid_to_fid(fid, ostid, 0);
 450         } else if (rc == sizeof(struct filter_fid)) {
 451                 rc = 1;
 452         } else if (rc >= 0) {
 453                 rc = -EINVAL;
 454         }
 455
 456         return rc;
 457 }
 458
 459 static int osd_lma_self_repair(struct osd_thread_info *info,
 460                                struct osd_device *osd, struct inode *inode,
 461                                const struct lu_fid *fid, __u32 compat)
 462 {
 463         handle_t *jh;
 464         int       rc;
 465
 466         LASSERT(current->journal_info == NULL);
 467
 468         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 469                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 470         if (IS_ERR(jh)) {
 471                 rc = PTR_ERR(jh);
 472                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 473                       osd_name(osd), rc);
 474                 return rc;
 475         }
 476
 477         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 478         if (rc != 0)
 479                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 480                       osd_name(osd), rc);
 481         ldiskfs_journal_stop(jh);
 482         return rc;
 483 }
 484
 485 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 486 {
 487         struct osd_thread_info  *info   = osd_oti_get(env);
 488         struct osd_device       *osd    = osd_obj2dev(obj);
 489         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 490         struct inode            *inode  = obj->oo_inode;
 491         struct dentry           *dentry = &info->oti_obj_dentry;
 492         struct lu_fid           *fid    = NULL;
 493         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 494         int                      rc;
 495         ENTRY;
 496
 497         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 498         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 499                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 500         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 501                 fid = &lma->lma_self_fid;
 502                 rc = osd_get_idif(info, inode, dentry, fid);
 503                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 504                         /* For the given OST-object, if it has neither LMA nor
 505                          * FID in XATTR_NAME_FID, then the given FID (which is
 506                          * contained in the @obj, from client RPC for locating
 507                          * the OST-object) is trusted. We use it to generate
 508                          * the LMA. */
 509                         osd_lma_self_repair(info, osd, inode, rfid,
 510                                             LMAC_FID_ON_OST);
 511                         RETURN(0);
 512                 }
 513         }
 514
 515         if (unlikely(rc == -ENODATA))
 516                 RETURN(0);
 517
 518         if (rc < 0)
 519                 RETURN(rc);
 520
 521         if (rc > 0) {
 522                 rc = 0;
 523                 lustre_lma_swab(lma);
 524                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 525                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 526                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 527                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 528                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 529                               PFID(rfid), inode->i_ino);
 530                         rc = -EOPNOTSUPP;
 531                 } else {
 532                         fid = &lma->lma_self_fid;
 533                 }
 534         }
 535
 536         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 537                 __u32 level = D_LFSCK;
 538
 539                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 540                         struct ost_id   *oi   = &info->oti_ostid;
 541                         struct lu_fid   *fid1 = &info->oti_fid3;
 542                         __u32            idx  = fid_idif_ost_idx(rfid);
 543
 544                         /* For old IDIF, the OST index is not part of the IDIF,
 545                          * Means that different OSTs may have the same IDIFs.
 546                          * Under such case, we need to make some compatible
 547                          * check to make sure to trigger OI scrub properly. */
 548                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 549                                 /* Given @rfid is new, LMA is old. */
 550                                 fid_to_ostid(fid, oi);
 551                                 ostid_to_fid(fid1, oi, idx);
 552                                 if (lu_fid_eq(fid1, rfid)) {
 553                                         if (osd->od_index_in_idif)
 554                                                 osd_lma_self_repair(info, osd,
 555                                                         inode, rfid,
 556                                                         LMAC_FID_ON_OST);
 557                                         RETURN(0);
 558                                 }
 559                         }
 560                 }
 561
 562
 563                 rc = -EREMCHG;
 564                 if (!thread_is_running(&osd->od_scrub.os_thread))
 565                         level |= D_CONSOLE;
 566
 567                 CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
 568                        osd_name(osd), PFID(rfid), PFID(fid));
 569         }
 570
 571         RETURN(rc);
 572 }
 573
 574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 575                           const struct lu_fid *fid,
 576                           const struct lu_object_conf *conf)
 577 {
 578         struct osd_thread_info *info;
 579         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 580         struct osd_device      *dev;
 581         struct osd_idmap_cache *oic;
 582         struct osd_inode_id    *id;
 583         struct inode           *inode;
 584         struct osd_scrub       *scrub;
 585         struct scrub_file      *sf;
 586         int                     result;
 587         int                     saved  = 0;
 588         bool                    in_oi  = false;
 589         bool                    triggered = false;
 590         ENTRY;
 591
 592         LINVRNT(osd_invariant(obj));
 593         LASSERT(obj->oo_inode == NULL);
 594         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 595
 596         dev = osd_dev(ldev);
 597         scrub = &dev->od_scrub;
 598         sf = &scrub->os_file;
 599         info = osd_oti_get(env);
 600         LASSERT(info);
 601         oic = &info->oti_cache;
 602
 603         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 604                 RETURN(-ENOENT);
 605
 606         /* For the object is created as locking anchor, or for the object to
 607          * be created on disk. No need to osd_oi_lookup() at here because FID
 608          * shouldn't never be re-used, if it's really a duplicate FID from
 609          * unexpected reason, we should be able to detect it later by calling
 610          * do_create->osd_oi_insert(). */
 611         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 612                 GOTO(out, result = 0);
 613
 614         /* Search order: 1. per-thread cache. */
 615         if (lu_fid_eq(fid, &oic->oic_fid) &&
 616             likely(oic->oic_dev == dev)) {
 617                 id = &oic->oic_lid;
 618                 goto iget;
 619         }
 620
 621         id = &info->oti_id;
 622         if (!list_empty(&scrub->os_inconsistent_items)) {
 623                 /* Search order: 2. OI scrub pending list. */
 624                 result = osd_oii_lookup(dev, fid, id);
 625                 if (result == 0)
 626                         goto iget;
 627         }
 628
 629         /* Search order: 3. OI files. */
 630         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 631         if (result == -ENOENT) {
 632                 if (!fid_is_norm(fid) ||
 633                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 634                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 635                                       sf->sf_oi_bitmap))
 636                         GOTO(out, result = 0);
 637
 638                 goto trigger;
 639         }
 640
 641         if (result != 0)
 642                 GOTO(out, result);
 643
 644         in_oi = true;
 645
 646 iget:
 647         inode = osd_iget_check(info, dev, fid, id, in_oi);
 648         if (IS_ERR(inode)) {
 649                 result = PTR_ERR(inode);
 650                 if (result == -ENOENT || result == -ESTALE) {
 651                         if (!in_oi)
 652                                 fid_zero(&oic->oic_fid);
 653
 654                         GOTO(out, result = -ENOENT);
 655                 } else if (result == -EREMCHG) {
 656
 657 trigger:
 658                         if (!in_oi)
 659                                 fid_zero(&oic->oic_fid);
 660
 661                         if (unlikely(triggered))
 662                                 GOTO(out, result = saved);
 663
 664                         triggered = true;
 665                         if (thread_is_running(&scrub->os_thread)) {
 666                                 result = -EINPROGRESS;
 667                         } else if (!dev->od_noscrub) {
 668                                 /* Since we do not know the right OI mapping,
 669                                  * we have to trigger OI scrub to scan the
 670                                  * whole device. */
 671                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
 672                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 673                                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
 674                                        "scrub by RPC for "DFID", rc = %d [1]\n",
 675                                        osd_name(dev), PFID(fid),result);
 676                                 if (result == 0 || result == -EALREADY)
 677                                         result = -EINPROGRESS;
 678                                 else
 679                                         result = -EREMCHG;
 680                         }
 681
 682                         /* We still have chance to get the valid inode: for the
 683                          * object which is referenced by remote name entry, the
 684                          * object on the local MDT will be linked under the dir
 685                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 686                          *
 687                          * We do not know whether the object for the given FID
 688                          * is referenced by some remote name entry or not, and
 689                          * especially for DNE II, a multiple-linked object may
 690                          * have many name entries reside on many MDTs.
 691                          *
 692                          * To simplify the operation, OSD will not distinguish
 693                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 694                          * only happened for the RPC from other MDT during the
 695                          * OI scrub, or for the client side RPC with FID only,
 696                          * such as FID to path, or from old connected client. */
 697                         saved = result;
 698                         result = osd_lookup_in_remote_parent(info, dev,
 699                                                              fid, id);
 700                         if (result == 0) {
 701                                 in_oi = false;
 702                                 goto iget;
 703                         }
 704
 705                         result = saved;
 706                 }
 707
 708                 GOTO(out, result);
 709         }
 710
 711         obj->oo_inode = inode;
 712         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 713
 714         result = osd_check_lma(env, obj);
 715         if (result != 0) {
 716                 iput(inode);
 717                 obj->oo_inode = NULL;
 718                 if (result == -EREMCHG) {
 719                         if (!in_oi) {
 720                                 result = osd_oi_lookup(info, dev, fid, id,
 721                                                        OI_CHECK_FLD);
 722                                 if (result != 0) {
 723                                         fid_zero(&oic->oic_fid);
 724                                         GOTO(out, result);
 725                                 }
 726                         }
 727
 728                         goto trigger;
 729                 }
 730
 731                 GOTO(out, result);
 732         }
 733
 734         obj->oo_compat_dot_created = 1;
 735         obj->oo_compat_dotdot_created = 1;
 736
 737         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 738                 GOTO(out, result = 0);
 739
 740         LASSERT(obj->oo_hl_head == NULL);
 741         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 742         if (obj->oo_hl_head == NULL) {
 743                 obj->oo_inode = NULL;
 744                 iput(inode);
 745                 GOTO(out, result = -ENOMEM);
 746         }
 747         GOTO(out, result = 0);
 748
 749 out:
 750         LINVRNT(osd_invariant(obj));
 751         return result;
 752 }
 753
 754 /*
 755  * Concurrency: shouldn't matter.
 756  */
 757 static void osd_object_init0(struct osd_object *obj)
 758 {
 759         LASSERT(obj->oo_inode != NULL);
 760         obj->oo_dt.do_body_ops = &osd_body_ops;
 761         obj->oo_dt.do_lu.lo_header->loh_attr |=
 762                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 763 }
 764
 765 /*
 766  * Concurrency: no concurrent access is possible that early in object
 767  * life-cycle.
 768  */
 769 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 770                            const struct lu_object_conf *conf)
 771 {
 772         struct osd_object *obj = osd_obj(l);
 773         int result;
 774
 775         LINVRNT(osd_invariant(obj));
 776
 777         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 778                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 779                 l->lo_header->loh_attr |= LOHA_EXISTS;
 780                 return 0;
 781         }
 782
 783         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 784         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 785         if (result == 0 && obj->oo_inode != NULL)
 786                 osd_object_init0(obj);
 787
 788         LINVRNT(osd_invariant(obj));
 789         return result;
 790 }
 791
 792 /*
 793  * Concurrency: no concurrent access is possible that late in object
 794  * life-cycle.
 795  */
 796 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 797 {
 798         struct osd_object *obj = osd_obj(l);
 799
 800         LINVRNT(osd_invariant(obj));
 801
 802         dt_object_fini(&obj->oo_dt);
 803         if (obj->oo_hl_head != NULL)
 804                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 805         OBD_FREE_PTR(obj);
 806 }
 807
 808 /*
 809  * Concurrency: no concurrent access is possible that late in object
 810  * life-cycle.
 811  */
 812 static void osd_index_fini(struct osd_object *o)
 813 {
 814         struct iam_container *bag;
 815
 816         if (o->oo_dir != NULL) {
 817                 bag = &o->oo_dir->od_container;
 818                 if (o->oo_inode != NULL) {
 819                         if (bag->ic_object == o->oo_inode)
 820                                 iam_container_fini(bag);
 821                 }
 822                 OBD_FREE_PTR(o->oo_dir);
 823                 o->oo_dir = NULL;
 824         }
 825 }
 826
 827 /*
 828  * Concurrency: no concurrent access is possible that late in object
 829  * life-cycle (for all existing callers, that is. New callers have to provide
 830  * their own locking.)
 831  */
 832 static int osd_inode_unlinked(const struct inode *inode)
 833 {
 834         return inode->i_nlink == 0;
 835 }
 836
 837 enum {
 838         OSD_TXN_OI_DELETE_CREDITS    = 20,
 839         OSD_TXN_INODE_DELETE_CREDITS = 20
 840 };
 841
 842 /*
 843  * Journal
 844  */
 845
 846 #if OSD_THANDLE_STATS
 847 /**
 848  * Set time when the handle is allocated
 849  */
 850 static void osd_th_alloced(struct osd_thandle *oth)
 851 {
 852         oth->oth_alloced = cfs_time_current();
 853 }
 854
 855 /**
 856  * Set time when the handle started
 857  */
 858 static void osd_th_started(struct osd_thandle *oth)
 859 {
 860         oth->oth_started = cfs_time_current();
 861 }
 862
 863 /**
 864  * Helper function to convert time interval to microseconds packed in
 865  * long int.
 866  */
 867 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 868 {
 869         struct timeval val;
 870
 871         cfs_duration_usec(cfs_time_sub(end, start), &val);
 872         return val.tv_sec * 1000000 + val.tv_usec;
 873 }
 874
 875 /**
 876  * Check whether the we deal with this handle for too long.
 877  */
 878 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 879                                 cfs_time_t alloced, cfs_time_t started,
 880                                 cfs_time_t closed)
 881 {
 882         cfs_time_t now = cfs_time_current();
 883
 884         LASSERT(dev != NULL);
 885
 886         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 887                             interval_to_usec(alloced, started));
 888         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 889                             interval_to_usec(started, closed));
 890         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 891                             interval_to_usec(closed, now));
 892
 893         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 894                 CWARN("transaction handle %p was open for too long: "
 895                       "now "CFS_TIME_T" ,"
 896                       "alloced "CFS_TIME_T" ,"
 897                       "started "CFS_TIME_T" ,"
 898                       "closed "CFS_TIME_T"\n",
 899                       oth, now, alloced, started, closed);
 900                 libcfs_debug_dumpstack(NULL);
 901         }
 902 }
 903
 904 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 905 {                                                                       \
 906         cfs_time_t __closed = cfs_time_current();                       \
 907         cfs_time_t __alloced = oth->oth_alloced;                        \
 908         cfs_time_t __started = oth->oth_started;                        \
 909                                                                         \
 910         expr;                                                           \
 911         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 912 }
 913
 914 #else /* OSD_THANDLE_STATS */
 915
 916 #define osd_th_alloced(h)                  do {} while(0)
 917 #define osd_th_started(h)                  do {} while(0)
 918 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 919
 920 #endif /* OSD_THANDLE_STATS */
 921
 922 /*
 923  * Concurrency: doesn't access mutable data.
 924  */
 925 static int osd_param_is_not_sane(const struct osd_device *dev,
 926                                  const struct thandle *th)
 927 {
 928         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 929
 930         return oh->ot_credits > osd_transaction_size(dev);
 931 }
 932
 933 /*
 934  * Concurrency: shouldn't matter.
 935  */
 936 static void osd_trans_commit_cb(struct super_block *sb,
 937                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 938 {
 939         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 940         struct thandle     *th  = &oh->ot_super;
 941         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 942         struct dt_txn_commit_cb *dcb, *tmp;
 943
 944         LASSERT(oh->ot_handle == NULL);
 945
 946         if (error)
 947                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 948
 949         dt_txn_hook_commit(th);
 950
 951         /* call per-transaction callbacks if any */
 952         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 954                          "commit callback entry: magic=%x name='%s'\n",
 955                          dcb->dcb_magic, dcb->dcb_name);
 956                 list_del_init(&dcb->dcb_linkage);
 957                 dcb->dcb_func(NULL, th, dcb, error);
 958         }
 959
 960         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 961         lu_device_put(lud);
 962         th->th_dev = NULL;
 963
 964         lu_context_exit(&th->th_ctx);
 965         lu_context_fini(&th->th_ctx);
 966         OBD_FREE_PTR(oh);
 967 }
 968
 969 static struct thandle *osd_trans_create(const struct lu_env *env,
 970                                         struct dt_device *d)
 971 {
 972         struct osd_thread_info  *oti = osd_oti_get(env);
 973         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 974         struct osd_thandle      *oh;
 975         struct thandle          *th;
 976         ENTRY;
 977
 978         /* on pending IO in this thread should left from prev. request */
 979         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 980
 981         th = ERR_PTR(-ENOMEM);
 982         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 983         if (oh != NULL) {
 984                 oh->ot_quota_trans = &oti->oti_quota_trans;
 985                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 986                 th = &oh->ot_super;
 987                 th->th_dev = d;
 988                 th->th_result = 0;
 989                 th->th_tags = LCT_TX_HANDLE;
 990                 oh->ot_credits = 0;
 991                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 992                 osd_th_alloced(oh);
 993
 994                 memset(oti->oti_declare_ops, 0,
 995                                         sizeof(oti->oti_declare_ops));
 996                 memset(oti->oti_declare_ops_rb, 0,
 997                                         sizeof(oti->oti_declare_ops_rb));
 998                 memset(oti->oti_declare_ops_cred, 0,
 999                                         sizeof(oti->oti_declare_ops_cred));
1000                 oti->oti_rollback = false;
1001         }
1002         RETURN(th);
1003 }
1004
1005 /*
1006  * Concurrency: shouldn't matter.
1007  */
1008 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1009                            struct thandle *th)
1010 {
1011         struct osd_thread_info *oti = osd_oti_get(env);
1012         struct osd_device  *dev = osd_dt_dev(d);
1013         handle_t           *jh;
1014         struct osd_thandle *oh;
1015         int rc;
1016
1017         ENTRY;
1018
1019         LASSERT(current->journal_info == NULL);
1020
1021         oh = container_of0(th, struct osd_thandle, ot_super);
1022         LASSERT(oh != NULL);
1023         LASSERT(oh->ot_handle == NULL);
1024
1025         rc = dt_txn_hook_start(env, d, th);
1026         if (rc != 0)
1027                 GOTO(out, rc);
1028
1029         if (unlikely(osd_param_is_not_sane(dev, th))) {
1030                 static unsigned long last_printed;
1031                 static int last_credits;
1032
1033                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1034                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1035                       oh->ot_credits,
1036                       osd_journal(dev)->j_max_transaction_buffers);
1037                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1038                       oti->oti_declare_ops[OSD_OT_CREATE],
1039                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1040                       oti->oti_declare_ops[OSD_OT_DESTROY],
1041                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1042                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1043                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1044                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1045                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1046                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1047                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1048                       oti->oti_declare_ops[OSD_OT_WRITE],
1049                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1050                       oti->oti_declare_ops[OSD_OT_PUNCH],
1051                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1052                       oti->oti_declare_ops[OSD_OT_QUOTA],
1053                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1054                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1055                       oti->oti_declare_ops[OSD_OT_INSERT],
1056                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1057                       oti->oti_declare_ops[OSD_OT_DELETE],
1058                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1059                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1060                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1061                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1062                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1063                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1064
1065                 if (last_credits != oh->ot_credits &&
1066                     time_after(jiffies, last_printed +
1067                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1068                         libcfs_debug_dumpstack(NULL);
1069                         last_credits = oh->ot_credits;
1070                         last_printed = jiffies;
1071                 }
1072                 /* XXX Limit the credits to 'max_transaction_buffers', and
1073                  *     let the underlying filesystem to catch the error if
1074                  *     we really need so many credits.
1075                  *
1076                  *     This should be removed when we can calculate the
1077                  *     credits precisely. */
1078                 oh->ot_credits = osd_transaction_size(dev);
1079         }
1080
1081         /*
1082          * XXX temporary stuff. Some abstraction layer should
1083          * be used.
1084          */
1085         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1086         osd_th_started(oh);
1087         if (!IS_ERR(jh)) {
1088                 oh->ot_handle = jh;
1089                 LASSERT(oti->oti_txns == 0);
1090                 lu_context_init(&th->th_ctx, th->th_tags);
1091                 lu_context_enter(&th->th_ctx);
1092
1093                 lu_device_get(&d->dd_lu_dev);
1094                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1095                               "osd-tx", th);
1096                 oti->oti_txns++;
1097                 rc = 0;
1098         } else {
1099                 rc = PTR_ERR(jh);
1100         }
1101 out:
1102         RETURN(rc);
1103 }
1104
1105 static int osd_seq_exists(const struct lu_env *env,
1106                           struct osd_device *osd, u64 seq)
1107 {
1108         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1109         struct seq_server_site  *ss = osd_seq_site(osd);
1110         int                     rc;
1111         ENTRY;
1112
1113         LASSERT(ss != NULL);
1114         LASSERT(ss->ss_server_fld != NULL);
1115
1116         rc = osd_fld_lookup(env, osd, seq, range);
1117         if (rc != 0) {
1118                 if (rc != -ENOENT)
1119                         CERROR("%s: can't lookup FLD sequence "LPX64
1120                                ": rc = %d\n", osd_name(osd), seq, rc);
1121                 RETURN(0);
1122         }
1123
1124         RETURN(ss->ss_node_id == range->lsr_index);
1125 }
1126
1127 /*
1128  * Concurrency: shouldn't matter.
1129  */
1130 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1131                           struct thandle *th)
1132 {
1133         int                     rc = 0;
1134         struct osd_thandle     *oh;
1135         struct osd_thread_info *oti = osd_oti_get(env);
1136         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1137         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1138         struct qsd_instance    *qsd = osd->od_quota_slave;
1139         struct lquota_trans    *qtrans;
1140         ENTRY;
1141
1142         oh = container_of0(th, struct osd_thandle, ot_super);
1143
1144         qtrans = oh->ot_quota_trans;
1145         oh->ot_quota_trans = NULL;
1146
1147         if (oh->ot_handle != NULL) {
1148                 handle_t *hdl = oh->ot_handle;
1149
1150                 /*
1151                  * add commit callback
1152                  * notice we don't do this in osd_trans_start()
1153                  * as underlying transaction can change during truncate
1154                  */
1155                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1156                                          &oh->ot_jcb);
1157
1158                 LASSERT(oti->oti_txns == 1);
1159                 oti->oti_txns--;
1160                 rc = dt_txn_hook_stop(env, th);
1161                 if (rc != 0)
1162                         CERROR("%s: failed in transaction hook: rc = %d\n",
1163                                osd_name(osd), rc);
1164
1165                 /* hook functions might modify th_sync */
1166                 hdl->h_sync = th->th_sync;
1167
1168                 oh->ot_handle = NULL;
1169                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1170                 if (rc != 0)
1171                         CERROR("%s: failed to stop transaction: rc = %d\n",
1172                                osd_name(osd), rc);
1173         } else {
1174                 OBD_FREE_PTR(oh);
1175         }
1176
1177         /* inform the quota slave device that the transaction is stopping */
1178         qsd_op_end(env, qsd, qtrans);
1179
1180         /* as we want IO to journal and data IO be concurrent, we don't block
1181          * awaiting data IO completion in osd_do_bio(), instead we wait here
1182          * once transaction is submitted to the journal. all reqular requests
1183          * don't do direct IO (except read/write), thus this wait_event becomes
1184          * no-op for them.
1185          *
1186          * IMPORTANT: we have to wait till any IO submited by the thread is
1187          * completed otherwise iobuf may be corrupted by different request
1188          */
1189         wait_event(iobuf->dr_wait,
1190                        atomic_read(&iobuf->dr_numreqs) == 0);
1191         osd_fini_iobuf(osd, iobuf);
1192         if (!rc)
1193                 rc = iobuf->dr_error;
1194
1195         RETURN(rc);
1196 }
1197
1198 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1199 {
1200         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1201                                                ot_super);
1202
1203         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1204         LASSERT(&dcb->dcb_func != NULL);
1205         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * Called just before object is freed. Releases all resources except for
1212  * object itself (that is released by osd_object_free()).
1213  *
1214  * Concurrency: no concurrent access is possible that late in object
1215  * life-cycle.
1216  */
1217 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1218 {
1219         struct osd_object *obj   = osd_obj(l);
1220         struct inode      *inode = obj->oo_inode;
1221
1222         LINVRNT(osd_invariant(obj));
1223
1224         /*
1225          * If object is unlinked remove fid->ino mapping from object index.
1226          */
1227
1228         osd_index_fini(obj);
1229         if (inode != NULL) {
1230                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1231                 qid_t                    uid = i_uid_read(inode);
1232                 qid_t                    gid = i_gid_read(inode);
1233
1234                 iput(inode);
1235                 obj->oo_inode = NULL;
1236
1237                 if (qsd != NULL) {
1238                         struct osd_thread_info  *info = osd_oti_get(env);
1239                         struct lquota_id_info   *qi = &info->oti_qi;
1240
1241                         /* Release granted quota to master if necessary */
1242                         qi->lqi_id.qid_uid = uid;
1243                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1244
1245                         qi->lqi_id.qid_uid = gid;
1246                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1247                 }
1248         }
1249 }
1250
1251 /*
1252  * Concurrency: ->loo_object_release() is called under site spin-lock.
1253  */
1254 static void osd_object_release(const struct lu_env *env,
1255                                struct lu_object *l)
1256 {
1257 }
1258
1259 /*
1260  * Concurrency: shouldn't matter.
1261  */
1262 static int osd_object_print(const struct lu_env *env, void *cookie,
1263                             lu_printer_t p, const struct lu_object *l)
1264 {
1265         struct osd_object *o = osd_obj(l);
1266         struct iam_descr  *d;
1267
1268         if (o->oo_dir != NULL)
1269                 d = o->oo_dir->od_container.ic_descr;
1270         else
1271                 d = NULL;
1272         return (*p)(env, cookie,
1273                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1274                     o, o->oo_inode,
1275                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1276                     o->oo_inode ? o->oo_inode->i_generation : 0,
1277                     d ? d->id_ops->id_name : "plain");
1278 }
1279
1280 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1281
1282 /*
1283  * Concurrency: shouldn't matter.
1284  */
1285 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1286                struct obd_statfs *sfs)
1287 {
1288         struct osd_device  *osd = osd_dt_dev(d);
1289         struct super_block *sb = osd_sb(osd);
1290         struct kstatfs     *ksfs;
1291         int result = 0;
1292
1293         if (unlikely(osd->od_mnt == NULL))
1294                 return -EINPROGRESS;
1295
1296         /* osd_lproc.c call this without env, allocate ksfs for that case */
1297         if (unlikely(env == NULL)) {
1298                 OBD_ALLOC_PTR(ksfs);
1299                 if (ksfs == NULL)
1300                         return -ENOMEM;
1301         } else {
1302                 ksfs = &osd_oti_get(env)->oti_ksfs;
1303         }
1304
1305         spin_lock(&osd->od_osfs_lock);
1306         result = sb->s_op->statfs(sb->s_root, ksfs);
1307         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1308                 statfs_pack(sfs, ksfs);
1309                 if (sb->s_flags & MS_RDONLY)
1310                         sfs->os_state = OS_STATE_READONLY;
1311         }
1312
1313         spin_unlock(&osd->od_osfs_lock);
1314
1315         if (unlikely(env == NULL))
1316                 OBD_FREE_PTR(ksfs);
1317
1318         /* Reserve a small amount of space for local objects like last_rcvd,
1319          * llog, quota files, ... */
1320         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1321                 sfs->os_bavail = 0;
1322         } else {
1323                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1324                 /** Take out metadata overhead for indirect blocks */
1325                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1326         }
1327
1328         return result;
1329 }
1330
1331 /**
1332  * Estimate space needed for file creations. We assume the largest filename
1333  * which is 2^64 - 1, hence a filename of 20 chars.
1334  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1335  */
1336 #ifdef __LDISKFS_DIR_REC_LEN
1337 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1338 #else
1339 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1340 #endif
1341
1342 /*
1343  * Concurrency: doesn't access mutable data.
1344  */
1345 static void osd_conf_get(const struct lu_env *env,
1346                          const struct dt_device *dev,
1347                          struct dt_device_param *param)
1348 {
1349         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1350         int                ea_overhead;
1351
1352         /*
1353          * XXX should be taken from not-yet-existing fs abstraction layer.
1354          */
1355         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1356         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1357         param->ddp_block_shift  = sb->s_blocksize_bits;
1358         param->ddp_mount_type     = LDD_MT_LDISKFS;
1359         param->ddp_maxbytes       = sb->s_maxbytes;
1360         /* Overhead estimate should be fairly accurate, so we really take a tiny
1361          * error margin which also avoids fragmenting the filesystem too much */
1362         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1363         /* inode are statically allocated, so per-inode space consumption
1364          * is the space consumed by the directory entry */
1365         param->ddp_inodespace     = PER_OBJ_USAGE;
1366         /* per-fragment overhead to be used by the client code */
1367         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1368         param->ddp_mntopts      = 0;
1369         if (test_opt(sb, XATTR_USER))
1370                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1371         if (test_opt(sb, POSIX_ACL))
1372                 param->ddp_mntopts |= MNTOPT_ACL;
1373
1374         /* LOD might calculate the max stripe count based on max_ea_size,
1375          * so we need take account in the overhead as well,
1376          * xattr_header + magic + xattr_entry_head */
1377         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1378                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1379
1380 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1381         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1382                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1383                                                                 ea_overhead;
1384         else
1385 #endif
1386                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1387 }
1388
1389 /*
1390  * Concurrency: shouldn't matter.
1391  */
1392 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1393 {
1394         int rc;
1395
1396         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1397
1398         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1399
1400         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1401                LUSTRE_OSD_LDISKFS_NAME, rc);
1402
1403         return rc;
1404 }
1405
1406 /**
1407  * Start commit for OSD device.
1408  *
1409  * An implementation of dt_commit_async method for OSD device.
1410  * Asychronously starts underlayng fs sync and thereby a transaction
1411  * commit.
1412  *
1413  * \param env environment
1414  * \param d dt device
1415  *
1416  * \see dt_device_operations
1417  */
1418 static int osd_commit_async(const struct lu_env *env,
1419                             struct dt_device *d)
1420 {
1421         struct super_block *s = osd_sb(osd_dt_dev(d));
1422         ENTRY;
1423
1424         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1425         RETURN(s->s_op->sync_fs(s, 0));
1426 }
1427
1428 /*
1429  * Concurrency: shouldn't matter.
1430  */
1431
1432 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1433 {
1434         struct super_block *sb = osd_sb(osd_dt_dev(d));
1435         struct block_device *dev = sb->s_bdev;
1436 #ifdef HAVE_DEV_SET_RDONLY
1437         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1438         int rc = 0;
1439 #else
1440         int rc = -EOPNOTSUPP;
1441 #endif
1442         ENTRY;
1443
1444 #ifdef HAVE_DEV_SET_RDONLY
1445         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1446
1447         if (jdev && (jdev != dev)) {
1448                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1449                        (long)jdev);
1450                 dev_set_rdonly(jdev);
1451         }
1452         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1453         dev_set_rdonly(dev);
1454 #else
1455         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1456                osd_dt_dev(d)->od_svname, (long)dev, rc);
1457 #endif
1458         RETURN(rc);
1459 }
1460
1461 /**
1462  * Note: we do not count into QUOTA here.
1463  * If we mount with --data_journal we may need more.
1464  */
1465 const int osd_dto_credits_noquota[DTO_NR] = {
1466         /**
1467          * Insert.
1468          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1469          * SINGLEDATA_TRANS_BLOCKS(8)
1470          * XXX Note: maybe iam need more, since iam have more level than
1471          *           EXT3 htree.
1472          */
1473         [DTO_INDEX_INSERT]  = 16,
1474         /**
1475          * Delete
1476          * just modify a single entry, probably merge few within a block
1477          */
1478         [DTO_INDEX_DELETE]  = 1,
1479         /**
1480          * Used for OI scrub
1481          */
1482         [DTO_INDEX_UPDATE]  = 16,
1483         /**
1484          * 4(inode, inode bits, groups, GDT)
1485          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1486          */
1487         [DTO_OBJECT_CREATE] = 4,
1488         /**
1489          * 4(inode, inode bits, groups, GDT)
1490          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1491          */
1492         [DTO_OBJECT_DELETE] = 4,
1493         /**
1494          * Attr set credits (inode)
1495          */
1496         [DTO_ATTR_SET_BASE] = 1,
1497         /**
1498          * Xattr set. The same as xattr of EXT3.
1499          * DATA_TRANS_BLOCKS(14)
1500          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1501          * are also counted in. Do not know why?
1502          */
1503         [DTO_XATTR_SET]     = 14,
1504         /**
1505          * credits for inode change during write.
1506          */
1507         [DTO_WRITE_BASE]    = 3,
1508         /**
1509          * credits for single block write.
1510          */
1511         [DTO_WRITE_BLOCK]   = 14,
1512         /**
1513          * Attr set credits for chown.
1514          * This is extra credits for setattr, and it is null without quota
1515          */
1516         [DTO_ATTR_SET_CHOWN] = 0
1517 };
1518
1519 static const struct dt_device_operations osd_dt_ops = {
1520         .dt_root_get       = osd_root_get,
1521         .dt_statfs         = osd_statfs,
1522         .dt_trans_create   = osd_trans_create,
1523         .dt_trans_start    = osd_trans_start,
1524         .dt_trans_stop     = osd_trans_stop,
1525         .dt_trans_cb_add   = osd_trans_cb_add,
1526         .dt_conf_get       = osd_conf_get,
1527         .dt_sync           = osd_sync,
1528         .dt_ro             = osd_ro,
1529         .dt_commit_async   = osd_commit_async,
1530 };
1531
1532 static void osd_object_read_lock(const struct lu_env *env,
1533                                  struct dt_object *dt, unsigned role)
1534 {
1535         struct osd_object *obj = osd_dt_obj(dt);
1536         struct osd_thread_info *oti = osd_oti_get(env);
1537
1538         LINVRNT(osd_invariant(obj));
1539
1540         LASSERT(obj->oo_owner != env);
1541         down_read_nested(&obj->oo_sem, role);
1542
1543         LASSERT(obj->oo_owner == NULL);
1544         oti->oti_r_locks++;
1545 }
1546
1547 static void osd_object_write_lock(const struct lu_env *env,
1548                                   struct dt_object *dt, unsigned role)
1549 {
1550         struct osd_object *obj = osd_dt_obj(dt);
1551         struct osd_thread_info *oti = osd_oti_get(env);
1552
1553         LINVRNT(osd_invariant(obj));
1554
1555         LASSERT(obj->oo_owner != env);
1556         down_write_nested(&obj->oo_sem, role);
1557
1558         LASSERT(obj->oo_owner == NULL);
1559         obj->oo_owner = env;
1560         oti->oti_w_locks++;
1561 }
1562
1563 static void osd_object_read_unlock(const struct lu_env *env,
1564                                    struct dt_object *dt)
1565 {
1566         struct osd_object *obj = osd_dt_obj(dt);
1567         struct osd_thread_info *oti = osd_oti_get(env);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         LASSERT(oti->oti_r_locks > 0);
1572         oti->oti_r_locks--;
1573         up_read(&obj->oo_sem);
1574 }
1575
1576 static void osd_object_write_unlock(const struct lu_env *env,
1577                                     struct dt_object *dt)
1578 {
1579         struct osd_object *obj = osd_dt_obj(dt);
1580         struct osd_thread_info *oti = osd_oti_get(env);
1581
1582         LINVRNT(osd_invariant(obj));
1583
1584         LASSERT(obj->oo_owner == env);
1585         LASSERT(oti->oti_w_locks > 0);
1586         oti->oti_w_locks--;
1587         obj->oo_owner = NULL;
1588         up_write(&obj->oo_sem);
1589 }
1590
1591 static int osd_object_write_locked(const struct lu_env *env,
1592                                    struct dt_object *dt)
1593 {
1594         struct osd_object *obj = osd_dt_obj(dt);
1595
1596         LINVRNT(osd_invariant(obj));
1597
1598         return obj->oo_owner == env;
1599 }
1600
1601 static struct timespec *osd_inode_time(const struct lu_env *env,
1602                                        struct inode *inode, __u64 seconds)
1603 {
1604         struct osd_thread_info  *oti = osd_oti_get(env);
1605         struct timespec         *t   = &oti->oti_time;
1606
1607         t->tv_sec = seconds;
1608         t->tv_nsec = 0;
1609         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1610         return t;
1611 }
1612
1613
1614 static void osd_inode_getattr(const struct lu_env *env,
1615                               struct inode *inode, struct lu_attr *attr)
1616 {
1617         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1618                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1619                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1620                            LA_TYPE;
1621
1622         attr->la_atime   = LTIME_S(inode->i_atime);
1623         attr->la_mtime   = LTIME_S(inode->i_mtime);
1624         attr->la_ctime   = LTIME_S(inode->i_ctime);
1625         attr->la_mode    = inode->i_mode;
1626         attr->la_size    = i_size_read(inode);
1627         attr->la_blocks  = inode->i_blocks;
1628         attr->la_uid     = i_uid_read(inode);
1629         attr->la_gid     = i_gid_read(inode);
1630         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1631         attr->la_nlink   = inode->i_nlink;
1632         attr->la_rdev    = inode->i_rdev;
1633         attr->la_blksize = 1 << inode->i_blkbits;
1634         attr->la_blkbits = inode->i_blkbits;
1635 }
1636
1637 static int osd_attr_get(const struct lu_env *env,
1638                         struct dt_object *dt,
1639                         struct lu_attr *attr)
1640 {
1641         struct osd_object *obj = osd_dt_obj(dt);
1642
1643         if (!dt_object_exists(dt))
1644                 return -ENOENT;
1645
1646         LASSERT(!dt_object_remote(dt));
1647         LINVRNT(osd_invariant(obj));
1648
1649         spin_lock(&obj->oo_guard);
1650         osd_inode_getattr(env, obj->oo_inode, attr);
1651         spin_unlock(&obj->oo_guard);
1652         return 0;
1653 }
1654
1655 static int osd_declare_attr_set(const struct lu_env *env,
1656                                 struct dt_object *dt,
1657                                 const struct lu_attr *attr,
1658                                 struct thandle *handle)
1659 {
1660         struct osd_thandle     *oh;
1661         struct osd_object      *obj;
1662         struct osd_thread_info *info = osd_oti_get(env);
1663         struct lquota_id_info  *qi = &info->oti_qi;
1664         qid_t                   uid;
1665         qid_t                   gid;
1666         long long               bspace;
1667         int                     rc = 0;
1668         bool                    enforce;
1669         ENTRY;
1670
1671         LASSERT(dt != NULL);
1672         LASSERT(handle != NULL);
1673
1674         obj = osd_dt_obj(dt);
1675         LASSERT(osd_invariant(obj));
1676
1677         oh = container_of0(handle, struct osd_thandle, ot_super);
1678         LASSERT(oh->ot_handle == NULL);
1679
1680         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1681                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1682
1683         if (attr == NULL || obj->oo_inode == NULL)
1684                 RETURN(rc);
1685
1686         bspace   = obj->oo_inode->i_blocks;
1687         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1688         bspace   = toqb(bspace);
1689
1690         /* Changing ownership is always preformed by super user, it should not
1691          * fail with EDQUOT.
1692          *
1693          * We still need to call the osd_declare_qid() to calculate the journal
1694          * credits for updating quota accounting files and to trigger quota
1695          * space adjustment once the operation is completed.*/
1696         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1697                 /* USERQUOTA */
1698                 uid = i_uid_read(obj->oo_inode);
1699                 qi->lqi_type = USRQUOTA;
1700                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1701                 /* inode accounting */
1702                 qi->lqi_is_blk = false;
1703
1704                 /* one more inode for the new uid ... */
1705                 qi->lqi_id.qid_uid = attr->la_uid;
1706                 qi->lqi_space      = 1;
1707                 /* Reserve credits for the new uid */
1708                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1709                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1710                         rc = 0;
1711                 if (rc)
1712                         RETURN(rc);
1713
1714                 /* and one less inode for the current uid */
1715                 qi->lqi_id.qid_uid = uid;
1716                 qi->lqi_space      = -1;
1717                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1718                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1719                         rc = 0;
1720                 if (rc)
1721                         RETURN(rc);
1722
1723                 /* block accounting */
1724                 qi->lqi_is_blk = true;
1725
1726                 /* more blocks for the new uid ... */
1727                 qi->lqi_id.qid_uid = attr->la_uid;
1728                 qi->lqi_space      = bspace;
1729                 /*
1730                  * Credits for the new uid has been reserved, re-use "obj"
1731                  * to save credit reservation.
1732                  */
1733                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1734                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1735                         rc = 0;
1736                 if (rc)
1737                         RETURN(rc);
1738
1739                 /* and finally less blocks for the current uid */
1740                 qi->lqi_id.qid_uid = uid;
1741                 qi->lqi_space      = -bspace;
1742                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1743                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1744                         rc = 0;
1745                 if (rc)
1746                         RETURN(rc);
1747
1748                 /* GROUP QUOTA */
1749                 gid = i_gid_read(obj->oo_inode);
1750                 qi->lqi_type = GRPQUOTA;
1751                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1752
1753                 /* inode accounting */
1754                 qi->lqi_is_blk = false;
1755
1756                 /* one more inode for the new gid ... */
1757                 qi->lqi_id.qid_gid = attr->la_gid;
1758                 qi->lqi_space      = 1;
1759                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1760                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1761                         rc = 0;
1762                 if (rc)
1763                         RETURN(rc);
1764
1765                 /* and one less inode for the current gid */
1766                 qi->lqi_id.qid_gid = gid;
1767                 qi->lqi_space      = -1;
1768                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1769                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1770                         rc = 0;
1771                 if (rc)
1772                         RETURN(rc);
1773
1774                 /* block accounting */
1775                 qi->lqi_is_blk = true;
1776
1777                 /* more blocks for the new gid ... */
1778                 qi->lqi_id.qid_gid = attr->la_gid;
1779                 qi->lqi_space      = bspace;
1780                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1781                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1782                         rc = 0;
1783                 if (rc)
1784                         RETURN(rc);
1785
1786                 /* and finally less blocks for the current gid */
1787                 qi->lqi_id.qid_gid = gid;
1788                 qi->lqi_space      = -bspace;
1789                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1790                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1791                         rc = 0;
1792                 if (rc)
1793                         RETURN(rc);
1794         }
1795
1796         RETURN(rc);
1797 }
1798
1799 static int osd_inode_setattr(const struct lu_env *env,
1800                              struct inode *inode, const struct lu_attr *attr)
1801 {
1802         __u64 bits = attr->la_valid;
1803
1804         /* Only allow set size for regular file */
1805         if (!S_ISREG(inode->i_mode))
1806                 bits &= ~(LA_SIZE | LA_BLOCKS);
1807
1808         if (bits == 0)
1809                 return 0;
1810
1811         if (bits & LA_ATIME)
1812                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1813         if (bits & LA_CTIME)
1814                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1815         if (bits & LA_MTIME)
1816                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1817         if (bits & LA_SIZE) {
1818                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1819                 i_size_write(inode, attr->la_size);
1820         }
1821
1822 #if 0
1823         /* OSD should not change "i_blocks" which is used by quota.
1824          * "i_blocks" should be changed by ldiskfs only. */
1825         if (bits & LA_BLOCKS)
1826                 inode->i_blocks = attr->la_blocks;
1827 #endif
1828         if (bits & LA_MODE)
1829                 inode->i_mode = (inode->i_mode & S_IFMT) |
1830                                 (attr->la_mode & ~S_IFMT);
1831         if (bits & LA_UID)
1832                 i_uid_write(inode, attr->la_uid);
1833         if (bits & LA_GID)
1834                 i_gid_write(inode, attr->la_gid);
1835         if (bits & LA_NLINK)
1836                 set_nlink(inode, attr->la_nlink);
1837         if (bits & LA_RDEV)
1838                 inode->i_rdev = attr->la_rdev;
1839
1840         if (bits & LA_FLAGS) {
1841                 /* always keep S_NOCMTIME */
1842                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1843                                  S_NOCMTIME;
1844         }
1845         return 0;
1846 }
1847
1848 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1849 {
1850         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1851             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1852                 struct iattr    iattr;
1853                 int             rc;
1854
1855                 ll_vfs_dq_init(inode);
1856                 iattr.ia_valid = 0;
1857                 if (attr->la_valid & LA_UID)
1858                         iattr.ia_valid |= ATTR_UID;
1859                 if (attr->la_valid & LA_GID)
1860                         iattr.ia_valid |= ATTR_GID;
1861                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1862                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1863
1864                 rc = ll_vfs_dq_transfer(inode, &iattr);
1865                 if (rc) {
1866                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1867                                "enforcement enabled on the ldiskfs "
1868                                "filesystem?\n", inode->i_sb->s_id, rc);
1869                         return rc;
1870                 }
1871         }
1872         return 0;
1873 }
1874
1875 static int osd_attr_set(const struct lu_env *env,
1876                         struct dt_object *dt,
1877                         const struct lu_attr *attr,
1878                         struct thandle *handle)
1879 {
1880         struct osd_object *obj = osd_dt_obj(dt);
1881         struct inode      *inode;
1882         int rc;
1883
1884         if (!dt_object_exists(dt))
1885                 return -ENOENT;
1886
1887         LASSERT(handle != NULL);
1888         LASSERT(!dt_object_remote(dt));
1889         LASSERT(osd_invariant(obj));
1890
1891         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1892
1893         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1894                 struct osd_thread_info  *oti  = osd_oti_get(env);
1895                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1896                 struct lu_fid           *fid1 = &oti->oti_fid;
1897                 struct osd_inode_id     *id   = &oti->oti_id;
1898                 struct iam_path_descr   *ipd;
1899                 struct iam_container    *bag;
1900                 struct osd_thandle      *oh;
1901                 int                      rc;
1902
1903                 fid_cpu_to_be(fid1, fid0);
1904                 memset(id, 1, sizeof(*id));
1905                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1906                                   fid0)->oi_dir.od_container;
1907                 ipd = osd_idx_ipd_get(env, bag);
1908                 if (unlikely(ipd == NULL))
1909                         RETURN(-ENOMEM);
1910
1911                 oh = container_of0(handle, struct osd_thandle, ot_super);
1912                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1913                                 (const struct iam_rec *)id, ipd);
1914                 osd_ipd_put(env, bag, ipd);
1915                 return(rc > 0 ? 0 : rc);
1916         }
1917
1918         inode = obj->oo_inode;
1919
1920         rc = osd_quota_transfer(inode, attr);
1921         if (rc)
1922                 return rc;
1923
1924         spin_lock(&obj->oo_guard);
1925         rc = osd_inode_setattr(env, inode, attr);
1926         spin_unlock(&obj->oo_guard);
1927
1928         if (!rc)
1929                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1930         return rc;
1931 }
1932
1933 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1934                                            struct osd_object *obj,
1935                                            const char *name, const int namelen)
1936 {
1937         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1938 }
1939
1940 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1941                       umode_t mode, struct dt_allocation_hint *hint,
1942                       struct thandle *th)
1943 {
1944         int result;
1945         struct osd_device  *osd = osd_obj2dev(obj);
1946         struct osd_thandle *oth;
1947         struct dt_object   *parent = NULL;
1948         struct inode       *inode;
1949
1950         LINVRNT(osd_invariant(obj));
1951         LASSERT(obj->oo_inode == NULL);
1952         LASSERT(obj->oo_hl_head == NULL);
1953
1954         if (S_ISDIR(mode) && ldiskfs_pdo) {
1955                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1956                 if (obj->oo_hl_head == NULL)
1957                         return -ENOMEM;
1958         }
1959
1960         oth = container_of(th, struct osd_thandle, ot_super);
1961         LASSERT(oth->ot_handle->h_transaction != NULL);
1962
1963         if (hint && hint->dah_parent)
1964                 parent = hint->dah_parent;
1965
1966         inode = ldiskfs_create_inode(oth->ot_handle,
1967                                      parent ? osd_dt_obj(parent)->oo_inode :
1968                                               osd_sb(osd)->s_root->d_inode,
1969                                      mode);
1970         if (!IS_ERR(inode)) {
1971                 /* Do not update file c/mtime in ldiskfs. */
1972                 inode->i_flags |= S_NOCMTIME;
1973
1974                 /* For new created object, it must be consistent,
1975                  * and it is unnecessary to scrub against it. */
1976                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1977
1978                 obj->oo_inode = inode;
1979                 result = 0;
1980         } else {
1981                 if (obj->oo_hl_head != NULL) {
1982                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1983                         obj->oo_hl_head = NULL;
1984                 }
1985                 result = PTR_ERR(inode);
1986         }
1987         LINVRNT(osd_invariant(obj));
1988         return result;
1989 }
1990
1991 enum {
1992         OSD_NAME_LEN = 255
1993 };
1994
1995 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1996                      struct lu_attr *attr,
1997                      struct dt_allocation_hint *hint,
1998                      struct dt_object_format *dof,
1999                      struct thandle *th)
2000 {
2001         int result;
2002         struct osd_thandle *oth;
2003         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2004
2005         LASSERT(S_ISDIR(attr->la_mode));
2006
2007         oth = container_of(th, struct osd_thandle, ot_super);
2008         LASSERT(oth->ot_handle->h_transaction != NULL);
2009         result = osd_mkfile(info, obj, mode, hint, th);
2010
2011         return result;
2012 }
2013
2014 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2015                         struct lu_attr *attr,
2016                         struct dt_allocation_hint *hint,
2017                         struct dt_object_format *dof,
2018                         struct thandle *th)
2019 {
2020         int result;
2021         struct osd_thandle *oth;
2022         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2023
2024         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2025
2026         LASSERT(S_ISREG(attr->la_mode));
2027
2028         oth = container_of(th, struct osd_thandle, ot_super);
2029         LASSERT(oth->ot_handle->h_transaction != NULL);
2030
2031         result = osd_mkfile(info, obj, mode, hint, th);
2032         if (result == 0) {
2033                 LASSERT(obj->oo_inode != NULL);
2034                 if (feat->dif_flags & DT_IND_VARKEY)
2035                         result = iam_lvar_create(obj->oo_inode,
2036                                                  feat->dif_keysize_max,
2037                                                  feat->dif_ptrsize,
2038                                                  feat->dif_recsize_max,
2039                                                  oth->ot_handle);
2040                 else
2041                         result = iam_lfix_create(obj->oo_inode,
2042                                                  feat->dif_keysize_max,
2043                                                  feat->dif_ptrsize,
2044                                                  feat->dif_recsize_max,
2045                                                  oth->ot_handle);
2046
2047         }
2048         return result;
2049 }
2050
2051 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2052                      struct lu_attr *attr,
2053                      struct dt_allocation_hint *hint,
2054                      struct dt_object_format *dof,
2055                      struct thandle *th)
2056 {
2057         LASSERT(S_ISREG(attr->la_mode));
2058         return osd_mkfile(info, obj, (attr->la_mode &
2059                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2060 }
2061
2062 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2063                      struct lu_attr *attr,
2064                      struct dt_allocation_hint *hint,
2065                      struct dt_object_format *dof,
2066                      struct thandle *th)
2067 {
2068         LASSERT(S_ISLNK(attr->la_mode));
2069         return osd_mkfile(info, obj, (attr->la_mode &
2070                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2071 }
2072
2073 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2074                      struct lu_attr *attr,
2075                      struct dt_allocation_hint *hint,
2076                      struct dt_object_format *dof,
2077                      struct thandle *th)
2078 {
2079         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2080         int result;
2081
2082         LINVRNT(osd_invariant(obj));
2083         LASSERT(obj->oo_inode == NULL);
2084         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2085                 S_ISFIFO(mode) || S_ISSOCK(mode));
2086
2087         result = osd_mkfile(info, obj, mode, hint, th);
2088         if (result == 0) {
2089                 LASSERT(obj->oo_inode != NULL);
2090                 /*
2091                  * This inode should be marked dirty for i_rdev.  Currently
2092                  * that is done in the osd_attr_init().
2093                  */
2094                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2095                                    attr->la_rdev);
2096         }
2097         LINVRNT(osd_invariant(obj));
2098         return result;
2099 }
2100
2101 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2102                               struct lu_attr *,
2103                               struct dt_allocation_hint *hint,
2104                               struct dt_object_format *dof,
2105                               struct thandle *);
2106
2107 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2108 {
2109         osd_obj_type_f result;
2110
2111         switch (type) {
2112         case DFT_DIR:
2113                 result = osd_mkdir;
2114                 break;
2115         case DFT_REGULAR:
2116                 result = osd_mkreg;
2117                 break;
2118         case DFT_SYM:
2119                 result = osd_mksym;
2120                 break;
2121         case DFT_NODE:
2122                 result = osd_mknod;
2123                 break;
2124         case DFT_INDEX:
2125                 result = osd_mk_index;
2126                 break;
2127
2128         default:
2129                 LBUG();
2130                 break;
2131         }
2132         return result;
2133 }
2134
2135
2136 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2137                         struct dt_object *parent, struct dt_object *child,
2138                         umode_t child_mode)
2139 {
2140         LASSERT(ah);
2141
2142         ah->dah_parent = parent;
2143         ah->dah_mode = child_mode;
2144 }
2145
2146 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2147                           struct lu_attr *attr, struct dt_object_format *dof)
2148 {
2149         struct inode   *inode = obj->oo_inode;
2150         __u64           valid = attr->la_valid;
2151         int             result;
2152
2153         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2154
2155         if (dof->dof_type != DFT_NODE)
2156                 attr->la_valid &= ~LA_RDEV;
2157         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2158                 attr->la_valid &= ~LA_ATIME;
2159         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2160                 attr->la_valid &= ~LA_CTIME;
2161         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2162                 attr->la_valid &= ~LA_MTIME;
2163
2164         result = osd_quota_transfer(inode, attr);
2165         if (result)
2166                 return;
2167
2168         if (attr->la_valid != 0) {
2169                 result = osd_inode_setattr(info->oti_env, inode, attr);
2170                 /*
2171                  * The osd_inode_setattr() should always succeed here.  The
2172                  * only error that could be returned is EDQUOT when we are
2173                  * trying to change the UID or GID of the inode. However, this
2174                  * should not happen since quota enforcement is no longer
2175                  * enabled on ldiskfs (lquota takes care of it).
2176                  */
2177                 LASSERTF(result == 0, "%d\n", result);
2178                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2179         }
2180
2181         attr->la_valid = valid;
2182 }
2183
2184 /**
2185  * Helper function for osd_object_create()
2186  *
2187  * \retval 0, on success
2188  */
2189 static int __osd_object_create(struct osd_thread_info *info,
2190                                struct osd_object *obj, struct lu_attr *attr,
2191                                struct dt_allocation_hint *hint,
2192                                struct dt_object_format *dof,
2193                                struct thandle *th)
2194 {
2195         int     result;
2196         __u32   umask;
2197
2198         /* we drop umask so that permissions we pass are not affected */
2199         umask = current->fs->umask;
2200         current->fs->umask = 0;
2201
2202         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2203                                                   th);
2204         if (likely(obj->oo_inode != NULL)) {
2205                 LASSERT(obj->oo_inode->i_state & I_NEW);
2206
2207                 /* Unlock the inode before attr initialization to avoid
2208                  * unnecessary dqget operations. LU-6378 */
2209                 unlock_new_inode(obj->oo_inode);
2210         }
2211
2212         if (likely(result == 0)) {
2213                 osd_attr_init(info, obj, attr, dof);
2214                 osd_object_init0(obj);
2215         }
2216
2217         /* restore previous umask value */
2218         current->fs->umask = umask;
2219
2220         return result;
2221 }
2222
2223 /**
2224  * Helper function for osd_object_create()
2225  *
2226  * \retval 0, on success
2227  */
2228 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2229                            const struct lu_fid *fid, struct thandle *th)
2230 {
2231         struct osd_thread_info *info = osd_oti_get(env);
2232         struct osd_inode_id    *id   = &info->oti_id;
2233         struct osd_device      *osd  = osd_obj2dev(obj);
2234         struct osd_thandle     *oh;
2235
2236         LASSERT(obj->oo_inode != NULL);
2237
2238         oh = container_of0(th, struct osd_thandle, ot_super);
2239         LASSERT(oh->ot_handle);
2240
2241         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2242         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2243 }
2244
2245 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2246                    u64 seq, struct lu_seq_range *range)
2247 {
2248         struct seq_server_site  *ss = osd_seq_site(osd);
2249
2250         if (fid_seq_is_idif(seq)) {
2251                 fld_range_set_ost(range);
2252                 range->lsr_index = idif_ost_idx(seq);
2253                 return 0;
2254         }
2255
2256         if (!fid_seq_in_fldb(seq)) {
2257                 fld_range_set_mdt(range);
2258                 if (ss != NULL)
2259                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2260                          * at all */
2261                         range->lsr_index = ss->ss_node_id;
2262                 return 0;
2263         }
2264
2265         LASSERT(ss != NULL);
2266         fld_range_set_any(range);
2267         /* OSD will only do local fld lookup */
2268         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2269 }
2270
2271 /*
2272  * Concurrency: no external locking is necessary.
2273  */
2274 static int osd_declare_object_create(const struct lu_env *env,
2275                                      struct dt_object *dt,
2276                                      struct lu_attr *attr,
2277                                      struct dt_allocation_hint *hint,
2278                                      struct dt_object_format *dof,
2279                                      struct thandle *handle)
2280 {
2281         struct osd_thandle      *oh;
2282         int                      rc;
2283         ENTRY;
2284
2285         LASSERT(handle != NULL);
2286
2287         oh = container_of0(handle, struct osd_thandle, ot_super);
2288         LASSERT(oh->ot_handle == NULL);
2289
2290         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2291                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2292         /* Reuse idle OI block may cause additional one OI block
2293          * to be changed. */
2294         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2295                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2296
2297         /* If this is directory, then we expect . and .. to be inserted as
2298          * well. The one directory block always needs to be created for the
2299          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2300          * block), there is no danger of needing a tree for the first block.
2301          */
2302         if (attr && S_ISDIR(attr->la_mode)) {
2303                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2304                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2305                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2306         }
2307
2308         if (!attr)
2309                 RETURN(0);
2310
2311         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2312                                    osd_dt_obj(dt), false, NULL, false);
2313         if (rc != 0)
2314                 RETURN(rc);
2315
2316         RETURN(rc);
2317 }
2318
2319 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2320                              struct lu_attr *attr,
2321                              struct dt_allocation_hint *hint,
2322                              struct dt_object_format *dof, struct thandle *th)
2323 {
2324         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2325         struct osd_object       *obj    = osd_dt_obj(dt);
2326         struct osd_thread_info  *info   = osd_oti_get(env);
2327         int result;
2328         ENTRY;
2329
2330         if (dt_object_exists(dt))
2331                 return -EEXIST;
2332
2333         LINVRNT(osd_invariant(obj));
2334         LASSERT(!dt_object_remote(dt));
2335         LASSERT(osd_write_locked(env, obj));
2336         LASSERT(th != NULL);
2337
2338         if (unlikely(fid_is_acct(fid)))
2339                 /* Quota files can't be created from the kernel any more,
2340                  * 'tune2fs -O quota' will take care of creating them */
2341                 RETURN(-EPERM);
2342
2343         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2344         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2345
2346         result = __osd_object_create(info, obj, attr, hint, dof, th);
2347         if (result == 0)
2348                 result = __osd_oi_insert(env, obj, fid, th);
2349
2350         LASSERT(ergo(result == 0,
2351                      dt_object_exists(dt) && !dt_object_remote(dt)));
2352
2353         LASSERT(osd_invariant(obj));
2354         RETURN(result);
2355 }
2356
2357 /**
2358  * Called to destroy on-disk representation of the object
2359  *
2360  * Concurrency: must be locked
2361  */
2362 static int osd_declare_object_destroy(const struct lu_env *env,
2363                                       struct dt_object *dt,
2364                                       struct thandle *th)
2365 {
2366         struct osd_object  *obj = osd_dt_obj(dt);
2367         struct inode       *inode = obj->oo_inode;
2368         struct osd_thandle *oh;
2369         int                 rc;
2370         ENTRY;
2371
2372         oh = container_of0(th, struct osd_thandle, ot_super);
2373         LASSERT(oh->ot_handle == NULL);
2374         LASSERT(inode);
2375
2376         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2377                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2378         /* Recycle idle OI leaf may cause additional three OI blocks
2379          * to be changed. */
2380         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2381                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2382         /* one less inode */
2383         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2384                                    -1, oh, obj, false, NULL, false);
2385         if (rc)
2386                 RETURN(rc);
2387         /* data to be truncated */
2388         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2389                                    0, oh, obj, true, NULL, false);
2390         RETURN(rc);
2391 }
2392
2393 static int osd_object_destroy(const struct lu_env *env,
2394                               struct dt_object *dt,
2395                               struct thandle *th)
2396 {
2397         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2398         struct osd_object      *obj = osd_dt_obj(dt);
2399         struct inode           *inode = obj->oo_inode;
2400         struct osd_device      *osd = osd_obj2dev(obj);
2401         struct osd_thandle     *oh;
2402         int                     result;
2403         ENTRY;
2404
2405         oh = container_of0(th, struct osd_thandle, ot_super);
2406         LASSERT(oh->ot_handle);
2407         LASSERT(inode);
2408         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2409
2410         if (unlikely(fid_is_acct(fid)))
2411                 RETURN(-EPERM);
2412
2413         if (S_ISDIR(inode->i_mode)) {
2414                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2415                         inode->i_nlink == 2);
2416                 /* it will check/delete the inode from remote parent,
2417                  * how to optimize it? unlink performance impaction XXX */
2418                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2419                 if (result != 0 && result != -ENOENT) {
2420                         CERROR("%s: delete inode "DFID": rc = %d\n",
2421                                osd_name(osd), PFID(fid), result);
2422                 }
2423                 spin_lock(&obj->oo_guard);
2424                 clear_nlink(inode);
2425                 spin_unlock(&obj->oo_guard);
2426                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2427         }
2428
2429         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2430
2431         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2432                                OI_CHECK_FLD);
2433
2434         /* XXX: add to ext3 orphan list */
2435         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2436
2437         /* not needed in the cache anymore */
2438         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2439
2440         RETURN(0);
2441 }
2442
2443 /**
2444  * Put the fid into lustre_mdt_attrs, and then place the structure
2445  * inode's ea. This fid should not be altered during the life time
2446  * of the inode.
2447  *
2448  * \retval +ve, on success
2449  * \retval -ve, on error
2450  *
2451  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2452  */
2453 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2454                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2455 {
2456         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2457         int                      rc;
2458         ENTRY;
2459
2460         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2461                 RETURN(0);
2462
2463         lustre_lma_init(lma, fid, compat, incompat);
2464         lustre_lma_swab(lma);
2465
2466         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2467                              XATTR_CREATE);
2468         /* LMA may already exist, but we need to check that all the
2469          * desired compat/incompat flags have been added. */
2470         if (unlikely(rc == -EEXIST)) {
2471                 if (compat == 0 && incompat == 0)
2472                         RETURN(0);
2473
2474                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2475                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2476                                      LMA_OLD_SIZE);
2477                 if (rc <= 0)
2478                         RETURN(-EINVAL);
2479
2480                 lustre_lma_swab(lma);
2481                 if (!(~lma->lma_compat & compat) &&
2482                     !(~lma->lma_incompat & incompat))
2483                         RETURN(0);
2484
2485                 lma->lma_compat |= compat;
2486                 lma->lma_incompat |= incompat;
2487                 lustre_lma_swab(lma);
2488                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2489                                      sizeof(*lma), XATTR_REPLACE);
2490         }
2491
2492         RETURN(rc);
2493 }
2494
2495 /**
2496  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2497  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2498  * To have compatilibility with 1.8 ldiskfs driver we need to have
2499  * magic number at start of fid data.
2500  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2501  * its inmemory API.
2502  */
2503 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2504                                          const struct lu_fid *fid)
2505 {
2506         if (!fid_is_namespace_visible(fid) ||
2507             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2508                 param->edp_magic = 0;
2509                 return;
2510         }
2511
2512         param->edp_magic = LDISKFS_LUFID_MAGIC;
2513         param->edp_len =  sizeof(struct lu_fid) + 1;
2514         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2515 }
2516
2517 /**
2518  * Try to read the fid from inode ea into dt_rec.
2519  *
2520  * \param fid object fid.
2521  *
2522  * \retval 0 on success
2523  */
2524 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2525                           __u32 ino, struct lu_fid *fid,
2526                           struct osd_inode_id *id)
2527 {
2528         struct osd_thread_info *info  = osd_oti_get(env);
2529         struct inode           *inode;
2530         ENTRY;
2531
2532         osd_id_gen(id, ino, OSD_OII_NOGEN);
2533         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2534         if (IS_ERR(inode))
2535                 RETURN(PTR_ERR(inode));
2536
2537         iput(inode);
2538         RETURN(0);
2539 }
2540
2541 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2542                                         struct inode *dir,
2543                                         struct inode *parent_dir,
2544                                         const struct lu_fid *dot_fid,
2545                                         const struct lu_fid *dot_dot_fid,
2546                                         struct osd_thandle *oth)
2547 {
2548         struct ldiskfs_dentry_param *dot_ldp;
2549         struct ldiskfs_dentry_param *dot_dot_ldp;
2550
2551         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2552         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2553
2554         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2555         dot_ldp->edp_magic = 0;
2556         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2557                                         dir, dot_ldp, dot_dot_ldp);
2558 }
2559
2560 /**
2561  * Create an local agent inode for remote entry
2562  */
2563 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2564                                                   struct osd_device *osd,
2565                                                   struct osd_object *pobj,
2566                                                   const struct lu_fid *fid,
2567                                                   __u32 type,
2568                                                   struct thandle *th)
2569 {
2570         struct osd_thread_info  *info = osd_oti_get(env);
2571         struct inode            *local;
2572         struct osd_thandle      *oh;
2573         int                     rc;
2574         ENTRY;
2575
2576         LASSERT(th);
2577         oh = container_of(th, struct osd_thandle, ot_super);
2578         LASSERT(oh->ot_handle->h_transaction != NULL);
2579
2580         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2581         if (IS_ERR(local)) {
2582                 CERROR("%s: create local error %d\n", osd_name(osd),
2583                        (int)PTR_ERR(local));
2584                 RETURN(local);
2585         }
2586
2587         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2588         unlock_new_inode(local);
2589
2590         /* Set special LMA flag for local agent inode */
2591         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2592         if (rc != 0) {
2593                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2594                        osd_name(osd), PFID(fid), rc);
2595                 RETURN(ERR_PTR(rc));
2596         }
2597
2598         if (!S_ISDIR(type))
2599                 RETURN(local);
2600
2601         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2602                                          lu_object_fid(&pobj->oo_dt.do_lu),
2603                                          fid, oh);
2604         if (rc != 0) {
2605                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2606                         osd_name(osd), PFID(fid), rc);
2607                 RETURN(ERR_PTR(rc));
2608         }
2609
2610         RETURN(local);
2611 }
2612
2613 /**
2614  * Delete local agent inode for remote entry
2615  */
2616 static int osd_delete_local_agent_inode(const struct lu_env *env,
2617                                         struct osd_device *osd,
2618                                         const struct lu_fid *fid,
2619                                         __u32 ino, struct osd_thandle *oh)
2620 {
2621         struct osd_thread_info  *oti = osd_oti_get(env);
2622         struct osd_inode_id     *id = &oti->oti_id;
2623         struct inode            *inode;
2624         ENTRY;
2625
2626         id->oii_ino = le32_to_cpu(ino);
2627         id->oii_gen = OSD_OII_NOGEN;
2628         inode = osd_iget(oti, osd, id);
2629         if (IS_ERR(inode)) {
2630                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2631                        PFID(fid), id->oii_ino, id->oii_gen);
2632                 RETURN(PTR_ERR(inode));
2633         }
2634
2635         clear_nlink(inode);
2636         mark_inode_dirty(inode);
2637         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2638                 osd_name(osd), PFID(fid), inode->i_ino);
2639         iput(inode);
2640         RETURN(0);
2641 }
2642
2643 /**
2644  * OSD layer object create function for interoperability mode (b11826).
2645  * This is mostly similar to osd_object_create(). Only difference being, fid is
2646  * inserted into inode ea here.
2647  *
2648  * \retval   0, on success
2649  * \retval -ve, on error
2650  */
2651 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2652                                 struct lu_attr *attr,
2653                                 struct dt_allocation_hint *hint,
2654                                 struct dt_object_format *dof,
2655                                 struct thandle *th)
2656 {
2657         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2658         struct osd_object       *obj    = osd_dt_obj(dt);
2659         struct osd_thread_info  *info   = osd_oti_get(env);
2660         int                      result;
2661
2662         ENTRY;
2663
2664         if (dt_object_exists(dt))
2665                 RETURN(-EEXIST);
2666
2667         LASSERT(osd_invariant(obj));
2668         LASSERT(!dt_object_remote(dt));
2669         LASSERT(osd_write_locked(env, obj));
2670         LASSERT(th != NULL);
2671
2672         if (unlikely(fid_is_acct(fid)))
2673                 /* Quota files can't be created from the kernel any more,
2674                  * 'tune2fs -O quota' will take care of creating them */
2675                 RETURN(-EPERM);
2676
2677         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2678         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2679
2680         result = __osd_object_create(info, obj, attr, hint, dof, th);
2681         if (result == 0) {
2682                 if (fid_is_idif(fid) &&
2683                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2684                         struct lu_fid *tfid = &info->oti_fid;
2685                         struct ost_id *oi   = &info->oti_ostid;
2686
2687                         fid_to_ostid(fid, oi);
2688                         ostid_to_fid(tfid, oi, 0);
2689                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2690                                                 LMAC_FID_ON_OST, 0);
2691                 } else {
2692                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2693                                 fid_is_on_ost(info, osd_obj2dev(obj),
2694                                               fid, OI_CHECK_FLD) ?
2695                                 LMAC_FID_ON_OST : 0, 0);
2696                 }
2697         }
2698
2699         if (result == 0)
2700                 result = __osd_oi_insert(env, obj, fid, th);
2701
2702         LASSERT(ergo(result == 0,
2703                      dt_object_exists(dt) && !dt_object_remote(dt)));
2704         LINVRNT(osd_invariant(obj));
2705         RETURN(result);
2706 }
2707
2708 static int osd_declare_object_ref_add(const struct lu_env *env,
2709                                       struct dt_object *dt,
2710                                       struct thandle *handle)
2711 {
2712         struct osd_thandle       *oh;
2713
2714         /* it's possible that object doesn't exist yet */
2715         LASSERT(handle != NULL);
2716
2717         oh = container_of0(handle, struct osd_thandle, ot_super);
2718         LASSERT(oh->ot_handle == NULL);
2719
2720         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2721                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2722
2723         return 0;
2724 }
2725
2726 /*
2727  * Concurrency: @dt is write locked.
2728  */
2729 static int osd_object_ref_add(const struct lu_env *env,
2730                               struct dt_object *dt, struct thandle *th)
2731 {
2732         struct osd_object  *obj = osd_dt_obj(dt);
2733         struct inode       *inode = obj->oo_inode;
2734         struct osd_thandle *oh;
2735         int                 rc = 0;
2736
2737         if (!dt_object_exists(dt))
2738                 return -ENOENT;
2739
2740         LINVRNT(osd_invariant(obj));
2741         LASSERT(!dt_object_remote(dt));
2742         LASSERT(osd_write_locked(env, obj));
2743         LASSERT(th != NULL);
2744
2745         oh = container_of0(th, struct osd_thandle, ot_super);
2746         LASSERT(oh->ot_handle != NULL);
2747
2748         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2749
2750         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2751                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2752         /*
2753          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2754          * (65000) subdirectories by storing "1" in i_nlink if the link count
2755          * would otherwise overflow. Directory tranversal tools understand
2756          * that (st_nlink == 1) indicates that the filesystem dose not track
2757          * hard links count on the directory, and will not abort subdirectory
2758          * scanning early once (st_nlink - 2) subdirs have been found.
2759          *
2760          * This also has to properly handle the case of inodes with nlink == 0
2761          * in case they are being linked into the PENDING directory
2762          */
2763         spin_lock(&obj->oo_guard);
2764         if (unlikely(inode->i_nlink == 0))
2765                 /* inc_nlink from 0 may cause WARN_ON */
2766                 set_nlink(inode, 1);
2767         else {
2768                 ldiskfs_inc_count(oh->ot_handle, inode);
2769                 if (!S_ISDIR(inode->i_mode))
2770                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2771         }
2772         spin_unlock(&obj->oo_guard);
2773
2774         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2775         LINVRNT(osd_invariant(obj));
2776
2777         return rc;
2778 }
2779
2780 static int osd_declare_object_ref_del(const struct lu_env *env,
2781                                       struct dt_object *dt,
2782                                       struct thandle *handle)
2783 {
2784         struct osd_thandle *oh;
2785
2786         LASSERT(!dt_object_remote(dt));
2787         LASSERT(handle != NULL);
2788
2789         oh = container_of0(handle, struct osd_thandle, ot_super);
2790         LASSERT(oh->ot_handle == NULL);
2791
2792         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2793                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2794
2795         return 0;
2796 }
2797
2798 /*
2799  * Concurrency: @dt is write locked.
2800  */
2801 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2802                               struct thandle *th)
2803 {
2804         struct osd_object       *obj = osd_dt_obj(dt);
2805         struct inode            *inode = obj->oo_inode;
2806         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2807         struct osd_thandle      *oh;
2808
2809         if (!dt_object_exists(dt))
2810                 return -ENOENT;
2811
2812         LINVRNT(osd_invariant(obj));
2813         LASSERT(!dt_object_remote(dt));
2814         LASSERT(osd_write_locked(env, obj));
2815         LASSERT(th != NULL);
2816
2817         oh = container_of0(th, struct osd_thandle, ot_super);
2818         LASSERT(oh->ot_handle != NULL);
2819
2820         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2821
2822         spin_lock(&obj->oo_guard);
2823         /* That can be result of upgrade from old Lustre version and
2824          * applied only to local files.  Just skip this ref_del call.
2825          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2826         if (inode->i_nlink == 0) {
2827                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2828                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2829                              ", maybe an upgraded file? (LU-3915)\n",
2830                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2831                 spin_unlock(&obj->oo_guard);
2832                 return 0;
2833         }
2834
2835         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2836                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2837
2838         ldiskfs_dec_count(oh->ot_handle, inode);
2839         spin_unlock(&obj->oo_guard);
2840
2841         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2842         LINVRNT(osd_invariant(obj));
2843
2844         return 0;
2845 }
2846
2847 /*
2848  * Get the 64-bit version for an inode.
2849  */
2850 static int osd_object_version_get(const struct lu_env *env,
2851                                   struct dt_object *dt, dt_obj_version_t *ver)
2852 {
2853         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2854
2855         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2856                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2857         *ver = LDISKFS_I(inode)->i_fs_version;
2858         return 0;
2859 }
2860
2861 /*
2862  * Concurrency: @dt is read locked.
2863  */
2864 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2865                          struct lu_buf *buf, const char *name)
2866 {
2867         struct osd_object      *obj    = osd_dt_obj(dt);
2868         struct inode           *inode  = obj->oo_inode;
2869         struct osd_thread_info *info   = osd_oti_get(env);
2870         struct dentry          *dentry = &info->oti_obj_dentry;
2871
2872         /* version get is not real XATTR but uses xattr API */
2873         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2874                 /* for version we are just using xattr API but change inode
2875                  * field instead */
2876                 if (buf->lb_len == 0)
2877                         return sizeof(dt_obj_version_t);
2878
2879                 if (buf->lb_len < sizeof(dt_obj_version_t))
2880                         return -ERANGE;
2881
2882                 osd_object_version_get(env, dt, buf->lb_buf);
2883
2884                 return sizeof(dt_obj_version_t);
2885         }
2886
2887         if (!dt_object_exists(dt))
2888                 return -ENOENT;
2889
2890         LASSERT(!dt_object_remote(dt));
2891         LASSERT(inode->i_op != NULL);
2892         LASSERT(inode->i_op->getxattr != NULL);
2893
2894         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2895 }
2896
2897
2898 static int osd_declare_xattr_set(const struct lu_env *env,
2899                                  struct dt_object *dt,
2900                                  const struct lu_buf *buf, const char *name,
2901                                  int fl, struct thandle *handle)
2902 {
2903         struct osd_thandle *oh;
2904         int credits;
2905         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2906
2907         LASSERT(handle != NULL);
2908
2909         oh = container_of0(handle, struct osd_thandle, ot_super);
2910         LASSERT(oh->ot_handle == NULL);
2911
2912         /* optimistic optimization: LMA is set first and usually fit inode */
2913         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2914                 if (dt_object_exists(dt))
2915                         credits = 0;
2916                 else
2917                         credits = 1;
2918         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2919                 credits = 1;
2920         } else {
2921                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2922                 if (buf && buf->lb_len > sb->s_blocksize) {
2923                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2924                                         sb->s_blocksize_bits;
2925                 }
2926                 /*
2927                  * xattr set may involve inode quota change, reserve credits for
2928                  * dquot_initialize()
2929                  */
2930                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2931         }
2932
2933         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2934
2935         return 0;
2936 }
2937
2938 /*
2939  * Set the 64-bit version for object
2940  */
2941 static void osd_object_version_set(const struct lu_env *env,
2942                                    struct dt_object *dt,
2943                                    dt_obj_version_t *new_version)
2944 {
2945         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2946
2947         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2948                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2949
2950         LDISKFS_I(inode)->i_fs_version = *new_version;
2951         /** Version is set after all inode operations are finished,
2952          *  so we should mark it dirty here */
2953         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2954 }
2955
2956 /*
2957  * Concurrency: @dt is write locked.
2958  */
2959 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2960                          const struct lu_buf *buf, const char *name, int fl,
2961                          struct thandle *handle)
2962 {
2963         struct osd_object      *obj      = osd_dt_obj(dt);
2964         struct inode           *inode    = obj->oo_inode;
2965         struct osd_thread_info *info     = osd_oti_get(env);
2966         int                     fs_flags = 0;
2967         ENTRY;
2968
2969         LASSERT(handle != NULL);
2970
2971         /* version set is not real XATTR */
2972         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2973                 /* for version we are just using xattr API but change inode
2974                  * field instead */
2975                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2976                 osd_object_version_set(env, dt, buf->lb_buf);
2977                 return sizeof(dt_obj_version_t);
2978         }
2979
2980         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2981                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2982
2983         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2984         if (fl & LU_XATTR_REPLACE)
2985                 fs_flags |= XATTR_REPLACE;
2986
2987         if (fl & LU_XATTR_CREATE)
2988                 fs_flags |= XATTR_CREATE;
2989
2990         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2991                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2992                 int                      rc;
2993
2994                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2995                 if (rc != 0)
2996                         RETURN(rc);
2997
2998                 lma->lma_incompat |= LMAI_STRIPED;
2999                 lustre_lma_swab(lma);
3000                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3001                                      sizeof(*lma), XATTR_REPLACE);
3002                 if (rc != 0)
3003                         RETURN(rc);
3004         }
3005
3006         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3007             strcmp(name, XATTR_NAME_LINK) == 0)
3008                 return -ENOSPC;
3009
3010         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3011                                fs_flags);
3012 }
3013
3014 /*
3015  * Concurrency: @dt is read locked.
3016  */
3017 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3018                           const struct lu_buf *buf)
3019 {
3020         struct osd_object      *obj    = osd_dt_obj(dt);
3021         struct inode           *inode  = obj->oo_inode;
3022         struct osd_thread_info *info   = osd_oti_get(env);
3023         struct dentry          *dentry = &info->oti_obj_dentry;
3024
3025         if (!dt_object_exists(dt))
3026                 return -ENOENT;
3027
3028         LASSERT(!dt_object_remote(dt));
3029         LASSERT(inode->i_op != NULL);
3030         LASSERT(inode->i_op->listxattr != NULL);
3031
3032         dentry->d_inode = inode;
3033         dentry->d_sb = inode->i_sb;
3034         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3035 }
3036
3037 static int osd_declare_xattr_del(const struct lu_env *env,
3038                                  struct dt_object *dt, const char *name,
3039                                  struct thandle *handle)
3040 {
3041         struct osd_thandle *oh;
3042         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3043
3044         LASSERT(!dt_object_remote(dt));
3045         LASSERT(handle != NULL);
3046
3047         oh = container_of0(handle, struct osd_thandle, ot_super);
3048         LASSERT(oh->ot_handle == NULL);
3049
3050         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3051                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3052         /*
3053          * xattr del may involve inode quota change, reserve credits for
3054          * dquot_initialize()
3055          */
3056         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3057
3058         return 0;
3059 }
3060
3061 /*
3062  * Concurrency: @dt is write locked.
3063  */
3064 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3065                          const char *name, struct thandle *handle)
3066 {
3067         struct osd_object      *obj    = osd_dt_obj(dt);
3068         struct inode           *inode  = obj->oo_inode;
3069         struct osd_thread_info *info   = osd_oti_get(env);
3070         struct dentry          *dentry = &info->oti_obj_dentry;
3071         int                     rc;
3072
3073         if (!dt_object_exists(dt))
3074                 return -ENOENT;
3075
3076         LASSERT(!dt_object_remote(dt));
3077         LASSERT(inode->i_op != NULL);
3078         LASSERT(inode->i_op->removexattr != NULL);
3079         LASSERT(handle != NULL);
3080
3081         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3082
3083         ll_vfs_dq_init(inode);
3084         dentry->d_inode = inode;
3085         dentry->d_sb = inode->i_sb;
3086         rc = inode->i_op->removexattr(dentry, name);
3087         return rc;
3088 }
3089
3090 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3091                            __u64 start, __u64 end)
3092 {
3093         struct osd_object       *obj    = osd_dt_obj(dt);
3094         struct inode            *inode  = obj->oo_inode;
3095         struct osd_thread_info  *info   = osd_oti_get(env);
3096         struct dentry           *dentry = &info->oti_obj_dentry;
3097         struct file             *file   = &info->oti_file;
3098         int                     rc;
3099
3100         ENTRY;
3101
3102         dentry->d_inode = inode;
3103         dentry->d_sb = inode->i_sb;
3104         file->f_dentry = dentry;
3105         file->f_mapping = inode->i_mapping;
3106         file->f_op = inode->i_fop;
3107         set_file_inode(file, inode);
3108
3109         rc = ll_vfs_fsync_range(file, start, end, 0);
3110
3111         RETURN(rc);
3112 }
3113
3114 /*
3115  * Index operations.
3116  */
3117
3118 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3119                            const struct dt_index_features *feat)
3120 {
3121         struct iam_descr *descr;
3122
3123         if (osd_object_is_root(o))
3124                 return feat == &dt_directory_features;
3125
3126         LASSERT(o->oo_dir != NULL);
3127
3128         descr = o->oo_dir->od_container.ic_descr;
3129         if (feat == &dt_directory_features) {
3130                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3131                         return 1;
3132                 else
3133                         return 0;
3134         } else {
3135                 return
3136                         feat->dif_keysize_min <= descr->id_key_size &&
3137                         descr->id_key_size <= feat->dif_keysize_max &&
3138                         feat->dif_recsize_min <= descr->id_rec_size &&
3139                         descr->id_rec_size <= feat->dif_recsize_max &&
3140                         !(feat->dif_flags & (DT_IND_VARKEY |
3141                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3142                         ergo(feat->dif_flags & DT_IND_UPDATE,
3143                              1 /* XXX check that object (and file system) is
3144                                 * writable */);
3145         }
3146 }
3147
3148 static int osd_iam_container_init(const struct lu_env *env,
3149                                   struct osd_object *obj,
3150                                   struct osd_directory *dir)
3151 {
3152         struct iam_container *bag = &dir->od_container;
3153         int result;
3154
3155         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3156         if (result != 0)
3157                 return result;
3158
3159         result = iam_container_setup(bag);
3160         if (result == 0)
3161                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3162         else
3163                 iam_container_fini(bag);
3164
3165         return result;
3166 }
3167
3168