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
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76 #include <lustre_linkea.h>
  77
  78 int ldiskfs_pdo = 1;
  79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  80                 "ldiskfs with parallel directory operations");
  81
  82 int ldiskfs_track_declares_assert;
  83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  84                 "LBUG during tracking of declares");
  85
  86 /* Slab to allocate dynlocks */
  87 struct kmem_cache *dynlock_cachep;
  88
  89 /* Slab to allocate osd_it_ea */
  90 struct kmem_cache *osd_itea_cachep;
  91
  92 static struct lu_kmem_descr ldiskfs_caches[] = {
  93         {
  94                 .ckd_cache = &dynlock_cachep,
  95                 .ckd_name  = "dynlock_cache",
  96                 .ckd_size  = sizeof(struct dynlock_handle)
  97         },
  98         {
  99                 .ckd_cache = &osd_itea_cachep,
 100                 .ckd_name  = "osd_itea_cache",
 101                 .ckd_size  = sizeof(struct osd_it_ea)
 102         },
 103         {
 104                 .ckd_cache = NULL
 105         }
 106 };
 107
 108 static const char dot[] = ".";
 109 static const char dotdot[] = "..";
 110 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 111
 112 static const struct lu_object_operations      osd_lu_obj_ops;
 113 static const struct dt_object_operations      osd_obj_ops;
 114 static const struct dt_object_operations      osd_obj_ea_ops;
 115 static const struct dt_object_operations      osd_obj_otable_it_ops;
 116 static const struct dt_index_operations       osd_index_iam_ops;
 117 static const struct dt_index_operations       osd_index_ea_ops;
 118
 119 int osd_trans_declare_op2rb[] = {
 120         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 121         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 122         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 123         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 124         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 125         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 126         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 127         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 128         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 129         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 130         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 131         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 132 };
 133
 134 static int osd_has_index(const struct osd_object *obj)
 135 {
 136         return obj->oo_dt.do_index_ops != NULL;
 137 }
 138
 139 static int osd_object_invariant(const struct lu_object *l)
 140 {
 141         return osd_invariant(osd_obj(l));
 142 }
 143
 144 /*
 145  * Concurrency: doesn't matter
 146  */
 147
 148 /*
 149  * Concurrency: doesn't matter
 150  */
 151 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 152 {
 153         struct osd_thread_info *oti = osd_oti_get(env);
 154         return oti->oti_w_locks > 0 && o->oo_owner == env;
 155 }
 156
 157 /*
 158  * Concurrency: doesn't access mutable data
 159  */
 160 static int osd_root_get(const struct lu_env *env,
 161                         struct dt_device *dev, struct lu_fid *f)
 162 {
 163         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 164         return 0;
 165 }
 166
 167 /*
 168  * OSD object methods.
 169  */
 170
 171 /*
 172  * Concurrency: no concurrent access is possible that early in object
 173  * life-cycle.
 174  */
 175 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 176                                           const struct lu_object_header *hdr,
 177                                           struct lu_device *d)
 178 {
 179         struct osd_object *mo;
 180
 181         OBD_ALLOC_PTR(mo);
 182         if (mo != NULL) {
 183                 struct lu_object *l;
 184
 185                 l = &mo->oo_dt.do_lu;
 186                 dt_object_init(&mo->oo_dt, NULL, d);
 187                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 188                 l->lo_ops = &osd_lu_obj_ops;
 189                 init_rwsem(&mo->oo_sem);
 190                 init_rwsem(&mo->oo_ext_idx_sem);
 191                 spin_lock_init(&mo->oo_guard);
 192                 return l;
 193         } else {
 194                 return NULL;
 195         }
 196 }
 197
 198 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 199                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 200 {
 201         int rc;
 202
 203         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 204         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 205                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 206         if (rc > 0) {
 207                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 208                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 209                 rc = 0;
 210                 lustre_lma_swab(lma);
 211                 /* Check LMA compatibility */
 212                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 213                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 214                               "for fid = "DFID", ino = %lu\n",
 215                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 216                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 217                               PFID(&lma->lma_self_fid), inode->i_ino);
 218                         rc = -EOPNOTSUPP;
 219                 }
 220         } else if (rc == 0) {
 221                 rc = -ENODATA;
 222         }
 223
 224         return rc;
 225 }
 226
 227 /*
 228  * retrieve object from backend ext fs.
 229  **/
 230 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 231                        struct osd_inode_id *id)
 232 {
 233         struct inode *inode = NULL;
 234
 235         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 236         if (IS_ERR(inode)) {
 237                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 238                        id->oii_ino, PTR_ERR(inode));
 239         } else if (id->oii_gen != OSD_OII_NOGEN &&
 240                    inode->i_generation != id->oii_gen) {
 241                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 242                        "i_generation = %u\n",
 243                        id->oii_ino, id->oii_gen, inode->i_generation);
 244                 iput(inode);
 245                 inode = ERR_PTR(-ESTALE);
 246         } else if (inode->i_nlink == 0) {
 247                 /* due to parallel readdir and unlink,
 248                 * we can have dead inode here. */
 249                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 250                 iput(inode);
 251                 inode = ERR_PTR(-ESTALE);
 252         } else if (is_bad_inode(inode)) {
 253                 CWARN("%.16s: bad inode: ino = %u\n",
 254                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 255                 iput(inode);
 256                 inode = ERR_PTR(-ENOENT);
 257         } else {
 258                 if (id->oii_gen == OSD_OII_NOGEN)
 259                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 260
 261                 /* Do not update file c/mtime in ldiskfs.
 262                  * NB: we don't have any lock to protect this because we don't
 263                  * have reference on osd_object now, but contention with
 264                  * another lookup + attr_set can't happen in the tiny window
 265                  * between if (...) and set S_NOCMTIME. */
 266                 if (!(inode->i_flags & S_NOCMTIME))
 267                         inode->i_flags |= S_NOCMTIME;
 268         }
 269         return inode;
 270 }
 271
 272 static struct inode *
 273 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 274              struct osd_inode_id *id, struct lu_fid *fid)
 275 {
 276         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 277         struct inode            *inode;
 278         int                      rc;
 279
 280         inode = osd_iget(info, dev, id);
 281         if (IS_ERR(inode))
 282                 return inode;
 283
 284         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 285         if (rc == 0) {
 286                 *fid = lma->lma_self_fid;
 287         } else if (rc == -ENODATA) {
 288                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 289                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 290                 else
 291                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 292         } else {
 293                 iput(inode);
 294                 inode = ERR_PTR(rc);
 295         }
 296         return inode;
 297 }
 298
 299 static struct inode *osd_iget_check(struct osd_thread_info *info,
 300                                     struct osd_device *dev,
 301                                     const struct lu_fid *fid,
 302                                     struct osd_inode_id *id,
 303                                     bool in_oi)
 304 {
 305         struct inode    *inode;
 306         int              rc     = 0;
 307         ENTRY;
 308
 309         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 310         if (IS_ERR(inode)) {
 311                 rc = PTR_ERR(inode);
 312                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 313                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 314                                id->oii_ino, rc);
 315
 316                         GOTO(put, rc);
 317                 }
 318
 319                 goto check_oi;
 320         }
 321
 322         if (is_bad_inode(inode)) {
 323                 rc = -ENOENT;
 324                 if (!in_oi) {
 325                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 326
 327                         GOTO(put, rc);
 328                 }
 329
 330                 goto check_oi;
 331         }
 332
 333         if (id->oii_gen != OSD_OII_NOGEN &&
 334             inode->i_generation != id->oii_gen) {
 335                 rc = -ESTALE;
 336                 if (!in_oi) {
 337                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 338                                "oii_gen = %u, i_generation = %u\n",
 339                                id->oii_ino, id->oii_gen, inode->i_generation);
 340
 341                         GOTO(put, rc);
 342                 }
 343
 344                 goto check_oi;
 345         }
 346
 347         if (inode->i_nlink == 0) {
 348                 rc = -ENOENT;
 349                 if (!in_oi) {
 350                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 351
 352                         GOTO(put, rc);
 353                 }
 354
 355                 goto check_oi;
 356         }
 357
 358 check_oi:
 359         if (rc != 0) {
 360                 struct osd_inode_id saved_id = *id;
 361
 362                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 363
 364                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 365                 /* XXX: There are some possible cases:
 366                  *      1. rc = 0.
 367                  *         Backup/restore caused the OI invalid.
 368                  *      2. rc = 0.
 369                  *         Someone unlinked the object but NOT removed
 370                  *         the OI mapping, such as mount target device
 371                  *         as ldiskfs, and modify something directly.
 372                  *      3. rc = -ENOENT.
 373                  *         Someone just removed the object between the
 374                  *         former oi_lookup and the iget. It is normal.
 375                  *      4. Other failure cases.
 376                  *
 377                  *      Generally, when the device is mounted, it will
 378                  *      auto check whether the system is restored from
 379                  *      file-level backup or not. We trust such detect
 380                  *      to distinguish the 1st case from the 2nd case. */
 381                 if (rc == 0) {
 382                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 383                             inode->i_generation == id->oii_gen) {
 384                                 rc = -ENOENT;
 385                         } else {
 386                                 __u32 level = D_LFSCK;
 387
 388                                 rc = -EREMCHG;
 389                                 if (!thread_is_running(&dev->od_scrub.os_thread))
 390                                         level |= D_CONSOLE;
 391
 392                                 CDEBUG(level, "%s: the OI mapping for the FID "
 393                                        DFID" become inconsistent, the given ID "
 394                                        "%u/%u, the ID in OI mapping %u/%u\n",
 395                                        osd_name(dev), PFID(fid),
 396                                        saved_id.oii_ino, saved_id.oii_gen,
 397                                        id->oii_ino, id->oii_ino);
 398                         }
 399                 }
 400         } else {
 401                 if (id->oii_gen == OSD_OII_NOGEN)
 402                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 403
 404                 /* Do not update file c/mtime in ldiskfs.
 405                  * NB: we don't have any lock to protect this because we don't
 406                  * have reference on osd_object now, but contention with
 407                  * another lookup + attr_set can't happen in the tiny window
 408                  * between if (...) and set S_NOCMTIME. */
 409                 if (!(inode->i_flags & S_NOCMTIME))
 410                         inode->i_flags |= S_NOCMTIME;
 411         }
 412
 413         GOTO(put, rc);
 414
 415 put:
 416         if (rc != 0) {
 417                 if (!IS_ERR(inode))
 418                         iput(inode);
 419
 420                 inode = ERR_PTR(rc);
 421         }
 422
 423         return inode;
 424 }
 425
 426 /**
 427  * \retval +v: new filter_fid, does not contain self-fid
 428  * \retval 0:  filter_fid_old, contains self-fid
 429  * \retval -v: other failure cases
 430  */
 431 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 432                  struct dentry *dentry, struct lu_fid *fid)
 433 {
 434         struct filter_fid_old   *ff     = &info->oti_ff;
 435         struct ost_id           *ostid  = &info->oti_ostid;
 436         int                      rc;
 437
 438         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 439         if (rc == sizeof(*ff)) {
 440                 rc = 0;
 441                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 442                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 443                 /* XXX: should use real OST index in the future. LU-3569 */
 444                 ostid_to_fid(fid, ostid, 0);
 445         } else if (rc == sizeof(struct filter_fid)) {
 446                 rc = 1;
 447         } else if (rc >= 0) {
 448                 rc = -EINVAL;
 449         }
 450
 451         return rc;
 452 }
 453
 454 static int osd_lma_self_repair(struct osd_thread_info *info,
 455                                struct osd_device *osd, struct inode *inode,
 456                                const struct lu_fid *fid, __u32 compat)
 457 {
 458         handle_t *jh;
 459         int       rc;
 460
 461         LASSERT(current->journal_info == NULL);
 462
 463         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 464                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 465         if (IS_ERR(jh)) {
 466                 rc = PTR_ERR(jh);
 467                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 468                       osd_name(osd), rc);
 469                 return rc;
 470         }
 471
 472         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 473         if (rc != 0)
 474                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 475                       osd_name(osd), rc);
 476         ldiskfs_journal_stop(jh);
 477         return rc;
 478 }
 479
 480 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 481 {
 482         struct osd_thread_info  *info   = osd_oti_get(env);
 483         struct osd_device       *osd    = osd_obj2dev(obj);
 484         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 485         struct inode            *inode  = obj->oo_inode;
 486         struct dentry           *dentry = &info->oti_obj_dentry;
 487         struct lu_fid           *fid    = NULL;
 488         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 489         int                      rc;
 490         ENTRY;
 491
 492         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 493         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 494                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 495         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 496                 fid = &lma->lma_self_fid;
 497                 rc = osd_get_idif(info, inode, dentry, fid);
 498                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 499                         /* For the given OST-object, if it has neither LMA nor
 500                          * FID in XATTR_NAME_FID, then the given FID (which is
 501                          * contained in the @obj, from client RPC for locating
 502                          * the OST-object) is trusted. We use it to generate
 503                          * the LMA. */
 504                         osd_lma_self_repair(info, osd, inode, rfid,
 505                                             LMAC_FID_ON_OST);
 506                         RETURN(0);
 507                 }
 508         }
 509
 510         if (unlikely(rc == -ENODATA))
 511                 RETURN(0);
 512
 513         if (rc < 0)
 514                 RETURN(rc);
 515
 516         if (rc > 0) {
 517                 rc = 0;
 518                 lustre_lma_swab(lma);
 519                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 520                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 521                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 522                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 523                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 524                               PFID(rfid), inode->i_ino);
 525                         rc = -EOPNOTSUPP;
 526                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 527                         fid = &lma->lma_self_fid;
 528                 }
 529         }
 530
 531         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 532                 __u32 level = D_LFSCK;
 533
 534                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 535                         struct ost_id   *oi   = &info->oti_ostid;
 536                         struct lu_fid   *fid1 = &info->oti_fid3;
 537                         __u32            idx  = fid_idif_ost_idx(rfid);
 538
 539                         /* For old IDIF, the OST index is not part of the IDIF,
 540                          * Means that different OSTs may have the same IDIFs.
 541                          * Under such case, we need to make some compatible
 542                          * check to make sure to trigger OI scrub properly. */
 543                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 544                                 /* Given @rfid is new, LMA is old. */
 545                                 fid_to_ostid(fid, oi);
 546                                 ostid_to_fid(fid1, oi, idx);
 547                                 if (lu_fid_eq(fid1, rfid)) {
 548                                         if (osd->od_index_in_idif)
 549                                                 osd_lma_self_repair(info, osd,
 550                                                         inode, rfid,
 551                                                         LMAC_FID_ON_OST);
 552                                         RETURN(0);
 553                                 }
 554                         }
 555                 }
 556
 557
 558                 rc = -EREMCHG;
 559                 if (!thread_is_running(&osd->od_scrub.os_thread))
 560                         level |= D_CONSOLE;
 561
 562                 CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
 563                        osd_name(osd), PFID(rfid), PFID(fid));
 564         }
 565
 566         RETURN(rc);
 567 }
 568
 569 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 570                           const struct lu_fid *fid,
 571                           const struct lu_object_conf *conf)
 572 {
 573         struct osd_thread_info *info;
 574         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 575         struct osd_device      *dev;
 576         struct osd_idmap_cache *oic;
 577         struct osd_inode_id    *id;
 578         struct inode           *inode;
 579         struct osd_scrub       *scrub;
 580         struct scrub_file      *sf;
 581         int                     result;
 582         int                     saved  = 0;
 583         bool                    in_oi  = false;
 584         bool                    triggered = false;
 585         ENTRY;
 586
 587         LINVRNT(osd_invariant(obj));
 588         LASSERT(obj->oo_inode == NULL);
 589         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 590
 591         dev = osd_dev(ldev);
 592         scrub = &dev->od_scrub;
 593         sf = &scrub->os_file;
 594         info = osd_oti_get(env);
 595         LASSERT(info);
 596         oic = &info->oti_cache;
 597
 598         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 599                 RETURN(-ENOENT);
 600
 601         /* For the object is created as locking anchor, or for the object to
 602          * be created on disk. No need to osd_oi_lookup() at here because FID
 603          * shouldn't never be re-used, if it's really a duplicate FID from
 604          * unexpected reason, we should be able to detect it later by calling
 605          * do_create->osd_oi_insert(). */
 606         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 607                 GOTO(out, result = 0);
 608
 609         /* Search order: 1. per-thread cache. */
 610         if (lu_fid_eq(fid, &oic->oic_fid) &&
 611             likely(oic->oic_dev == dev)) {
 612                 id = &oic->oic_lid;
 613                 goto iget;
 614         }
 615
 616         id = &info->oti_id;
 617         if (!list_empty(&scrub->os_inconsistent_items)) {
 618                 /* Search order: 2. OI scrub pending list. */
 619                 result = osd_oii_lookup(dev, fid, id);
 620                 if (result == 0)
 621                         goto iget;
 622         }
 623
 624         /* Search order: 3. OI files. */
 625         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 626         if (result == -ENOENT) {
 627                 if (!fid_is_norm(fid) ||
 628                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 629                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 630                                       sf->sf_oi_bitmap))
 631                         GOTO(out, result = 0);
 632
 633                 goto trigger;
 634         }
 635
 636         if (result != 0)
 637                 GOTO(out, result);
 638
 639         in_oi = true;
 640
 641 iget:
 642         inode = osd_iget_check(info, dev, fid, id, in_oi);
 643         if (IS_ERR(inode)) {
 644                 result = PTR_ERR(inode);
 645                 if (result == -ENOENT || result == -ESTALE) {
 646                         if (!in_oi)
 647                                 fid_zero(&oic->oic_fid);
 648
 649                         GOTO(out, result = -ENOENT);
 650                 } else if (result == -EREMCHG) {
 651
 652 trigger:
 653                         if (!in_oi)
 654                                 fid_zero(&oic->oic_fid);
 655
 656                         if (unlikely(triggered))
 657                                 GOTO(out, result = saved);
 658
 659                         triggered = true;
 660                         if (thread_is_running(&scrub->os_thread)) {
 661                                 result = -EINPROGRESS;
 662                         } else if (!dev->od_noscrub) {
 663                                 /* Since we do not know the right OI mapping,
 664                                  * we have to trigger OI scrub to scan the
 665                                  * whole device. */
 666                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
 667                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 668                                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
 669                                        "scrub by RPC for "DFID", rc = %d [1]\n",
 670                                        osd_name(dev), PFID(fid),result);
 671                                 if (result == 0 || result == -EALREADY)
 672                                         result = -EINPROGRESS;
 673                                 else
 674                                         result = -EREMCHG;
 675                         }
 676
 677                         /* We still have chance to get the valid inode: for the
 678                          * object which is referenced by remote name entry, the
 679                          * object on the local MDT will be linked under the dir
 680                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 681                          *
 682                          * We do not know whether the object for the given FID
 683                          * is referenced by some remote name entry or not, and
 684                          * especially for DNE II, a multiple-linked object may
 685                          * have many name entries reside on many MDTs.
 686                          *
 687                          * To simplify the operation, OSD will not distinguish
 688                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 689                          * only happened for the RPC from other MDT during the
 690                          * OI scrub, or for the client side RPC with FID only,
 691                          * such as FID to path, or from old connected client. */
 692                         saved = result;
 693                         result = osd_lookup_in_remote_parent(info, dev,
 694                                                              fid, id);
 695                         if (result == 0) {
 696                                 in_oi = false;
 697                                 goto iget;
 698                         }
 699
 700                         result = saved;
 701                 }
 702
 703                 GOTO(out, result);
 704         }
 705
 706         obj->oo_inode = inode;
 707         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 708
 709         result = osd_check_lma(env, obj);
 710         if (result != 0) {
 711                 iput(inode);
 712                 obj->oo_inode = NULL;
 713                 if (result == -EREMCHG) {
 714                         if (!in_oi) {
 715                                 result = osd_oi_lookup(info, dev, fid, id,
 716                                                        OI_CHECK_FLD);
 717                                 if (result != 0) {
 718                                         fid_zero(&oic->oic_fid);
 719                                         GOTO(out, result);
 720                                 }
 721                         }
 722
 723                         goto trigger;
 724                 }
 725
 726                 GOTO(out, result);
 727         }
 728
 729         obj->oo_compat_dot_created = 1;
 730         obj->oo_compat_dotdot_created = 1;
 731
 732         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 733                 GOTO(out, result = 0);
 734
 735         LASSERT(obj->oo_hl_head == NULL);
 736         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 737         if (obj->oo_hl_head == NULL) {
 738                 obj->oo_inode = NULL;
 739                 iput(inode);
 740                 GOTO(out, result = -ENOMEM);
 741         }
 742         GOTO(out, result = 0);
 743
 744 out:
 745         LINVRNT(osd_invariant(obj));
 746         return result;
 747 }
 748
 749 /*
 750  * Concurrency: shouldn't matter.
 751  */
 752 static void osd_object_init0(struct osd_object *obj)
 753 {
 754         LASSERT(obj->oo_inode != NULL);
 755         obj->oo_dt.do_body_ops = &osd_body_ops;
 756         obj->oo_dt.do_lu.lo_header->loh_attr |=
 757                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 758 }
 759
 760 /*
 761  * Concurrency: no concurrent access is possible that early in object
 762  * life-cycle.
 763  */
 764 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 765                            const struct lu_object_conf *conf)
 766 {
 767         struct osd_object *obj = osd_obj(l);
 768         int result;
 769
 770         LINVRNT(osd_invariant(obj));
 771
 772         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 773                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 774                 l->lo_header->loh_attr |= LOHA_EXISTS;
 775                 return 0;
 776         }
 777
 778         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 779         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 780         if (result == 0 && obj->oo_inode != NULL)
 781                 osd_object_init0(obj);
 782
 783         LINVRNT(osd_invariant(obj));
 784         return result;
 785 }
 786
 787 /*
 788  * Concurrency: no concurrent access is possible that late in object
 789  * life-cycle.
 790  */
 791 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 792 {
 793         struct osd_object *obj = osd_obj(l);
 794
 795         LINVRNT(osd_invariant(obj));
 796
 797         dt_object_fini(&obj->oo_dt);
 798         if (obj->oo_hl_head != NULL)
 799                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 800         OBD_FREE_PTR(obj);
 801 }
 802
 803 /*
 804  * Concurrency: no concurrent access is possible that late in object
 805  * life-cycle.
 806  */
 807 static void osd_index_fini(struct osd_object *o)
 808 {
 809         struct iam_container *bag;
 810
 811         if (o->oo_dir != NULL) {
 812                 bag = &o->oo_dir->od_container;
 813                 if (o->oo_inode != NULL) {
 814                         if (bag->ic_object == o->oo_inode)
 815                                 iam_container_fini(bag);
 816                 }
 817                 OBD_FREE_PTR(o->oo_dir);
 818                 o->oo_dir = NULL;
 819         }
 820 }
 821
 822 /*
 823  * Concurrency: no concurrent access is possible that late in object
 824  * life-cycle (for all existing callers, that is. New callers have to provide
 825  * their own locking.)
 826  */
 827 static int osd_inode_unlinked(const struct inode *inode)
 828 {
 829         return inode->i_nlink == 0;
 830 }
 831
 832 enum {
 833         OSD_TXN_OI_DELETE_CREDITS    = 20,
 834         OSD_TXN_INODE_DELETE_CREDITS = 20
 835 };
 836
 837 /*
 838  * Journal
 839  */
 840
 841 #if OSD_THANDLE_STATS
 842 /**
 843  * Set time when the handle is allocated
 844  */
 845 static void osd_th_alloced(struct osd_thandle *oth)
 846 {
 847         oth->oth_alloced = cfs_time_current();
 848 }
 849
 850 /**
 851  * Set time when the handle started
 852  */
 853 static void osd_th_started(struct osd_thandle *oth)
 854 {
 855         oth->oth_started = cfs_time_current();
 856 }
 857
 858 /**
 859  * Helper function to convert time interval to microseconds packed in
 860  * long int.
 861  */
 862 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 863 {
 864         struct timeval val;
 865
 866         cfs_duration_usec(cfs_time_sub(end, start), &val);
 867         return val.tv_sec * 1000000 + val.tv_usec;
 868 }
 869
 870 /**
 871  * Check whether the we deal with this handle for too long.
 872  */
 873 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 874                                 cfs_time_t alloced, cfs_time_t started,
 875                                 cfs_time_t closed)
 876 {
 877         cfs_time_t now = cfs_time_current();
 878
 879         LASSERT(dev != NULL);
 880
 881         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 882                             interval_to_usec(alloced, started));
 883         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 884                             interval_to_usec(started, closed));
 885         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 886                             interval_to_usec(closed, now));
 887
 888         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 889                 CWARN("transaction handle %p was open for too long: "
 890                       "now "CFS_TIME_T" ,"
 891                       "alloced "CFS_TIME_T" ,"
 892                       "started "CFS_TIME_T" ,"
 893                       "closed "CFS_TIME_T"\n",
 894                       oth, now, alloced, started, closed);
 895                 libcfs_debug_dumpstack(NULL);
 896         }
 897 }
 898
 899 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 900 {                                                                       \
 901         cfs_time_t __closed = cfs_time_current();                       \
 902         cfs_time_t __alloced = oth->oth_alloced;                        \
 903         cfs_time_t __started = oth->oth_started;                        \
 904                                                                         \
 905         expr;                                                           \
 906         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 907 }
 908
 909 #else /* OSD_THANDLE_STATS */
 910
 911 #define osd_th_alloced(h)                  do {} while(0)
 912 #define osd_th_started(h)                  do {} while(0)
 913 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 914
 915 #endif /* OSD_THANDLE_STATS */
 916
 917 /*
 918  * Concurrency: doesn't access mutable data.
 919  */
 920 static int osd_param_is_not_sane(const struct osd_device *dev,
 921                                  const struct thandle *th)
 922 {
 923         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 924
 925         return oh->ot_credits > osd_transaction_size(dev);
 926 }
 927
 928 /*
 929  * Concurrency: shouldn't matter.
 930  */
 931 static void osd_trans_commit_cb(struct super_block *sb,
 932                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 933 {
 934         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 935         struct thandle     *th  = &oh->ot_super;
 936         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 937         struct dt_txn_commit_cb *dcb, *tmp;
 938
 939         LASSERT(oh->ot_handle == NULL);
 940
 941         if (error)
 942                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 943
 944         dt_txn_hook_commit(th);
 945
 946         /* call per-transaction callbacks if any */
 947         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 948                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 949                          "commit callback entry: magic=%x name='%s'\n",
 950                          dcb->dcb_magic, dcb->dcb_name);
 951                 list_del_init(&dcb->dcb_linkage);
 952                 dcb->dcb_func(NULL, th, dcb, error);
 953         }
 954
 955         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 956         lu_device_put(lud);
 957         th->th_dev = NULL;
 958
 959         lu_context_exit(&th->th_ctx);
 960         lu_context_fini(&th->th_ctx);
 961         thandle_put(th);
 962 }
 963
 964 static struct thandle *osd_trans_create(const struct lu_env *env,
 965                                         struct dt_device *d)
 966 {
 967         struct osd_thread_info  *oti = osd_oti_get(env);
 968         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 969         struct osd_thandle      *oh;
 970         struct thandle          *th;
 971         ENTRY;
 972
 973         /* on pending IO in this thread should left from prev. request */
 974         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 975
 976         th = ERR_PTR(-ENOMEM);
 977         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 978         if (oh != NULL) {
 979                 oh->ot_quota_trans = &oti->oti_quota_trans;
 980                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 981                 th = &oh->ot_super;
 982                 th->th_dev = d;
 983                 th->th_result = 0;
 984                 th->th_tags = LCT_TX_HANDLE;
 985                 oh->ot_credits = 0;
 986                 atomic_set(&th->th_refc, 1);
 987                 th->th_alloc_size = sizeof(*oh);
 988                 oti->oti_dev = osd_dt_dev(d);
 989                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 990                 osd_th_alloced(oh);
 991
 992                 memset(oti->oti_declare_ops, 0,
 993                                         sizeof(oti->oti_declare_ops));
 994                 memset(oti->oti_declare_ops_rb, 0,
 995                                         sizeof(oti->oti_declare_ops_rb));
 996                 memset(oti->oti_declare_ops_cred, 0,
 997                                         sizeof(oti->oti_declare_ops_cred));
 998                 oti->oti_rollback = false;
 999         }
1000         RETURN(th);
1001 }
1002
1003 /*
1004  * Concurrency: shouldn't matter.
1005  */
1006 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1007                            struct thandle *th)
1008 {
1009         struct osd_thread_info *oti = osd_oti_get(env);
1010         struct osd_device  *dev = osd_dt_dev(d);
1011         handle_t           *jh;
1012         struct osd_thandle *oh;
1013         int rc;
1014
1015         ENTRY;
1016
1017         LASSERT(current->journal_info == NULL);
1018
1019         oh = container_of0(th, struct osd_thandle, ot_super);
1020         LASSERT(oh != NULL);
1021         LASSERT(oh->ot_handle == NULL);
1022
1023         rc = dt_txn_hook_start(env, d, th);
1024         if (rc != 0)
1025                 GOTO(out, rc);
1026
1027         if (unlikely(osd_param_is_not_sane(dev, th))) {
1028                 static unsigned long last_printed;
1029                 static int last_credits;
1030
1031                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1032                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1033                       oh->ot_credits,
1034                       osd_journal(dev)->j_max_transaction_buffers);
1035                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1036                       oti->oti_declare_ops[OSD_OT_CREATE],
1037                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1038                       oti->oti_declare_ops[OSD_OT_DESTROY],
1039                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1040                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1041                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1042                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1043                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1044                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1045                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1046                       oti->oti_declare_ops[OSD_OT_WRITE],
1047                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1048                       oti->oti_declare_ops[OSD_OT_PUNCH],
1049                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1050                       oti->oti_declare_ops[OSD_OT_QUOTA],
1051                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1052                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1053                       oti->oti_declare_ops[OSD_OT_INSERT],
1054                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1055                       oti->oti_declare_ops[OSD_OT_DELETE],
1056                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1057                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1058                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1059                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1060                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1061                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1062
1063                 if (last_credits != oh->ot_credits &&
1064                     time_after(jiffies, last_printed +
1065                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1066                         libcfs_debug_dumpstack(NULL);
1067                         last_credits = oh->ot_credits;
1068                         last_printed = jiffies;
1069                 }
1070                 /* XXX Limit the credits to 'max_transaction_buffers', and
1071                  *     let the underlying filesystem to catch the error if
1072                  *     we really need so many credits.
1073                  *
1074                  *     This should be removed when we can calculate the
1075                  *     credits precisely. */
1076                 oh->ot_credits = osd_transaction_size(dev);
1077         }
1078
1079         /*
1080          * XXX temporary stuff. Some abstraction layer should
1081          * be used.
1082          */
1083         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1084         osd_th_started(oh);
1085         if (!IS_ERR(jh)) {
1086                 oh->ot_handle = jh;
1087                 LASSERT(oti->oti_txns == 0);
1088                 lu_context_init(&th->th_ctx, th->th_tags);
1089                 lu_context_enter(&th->th_ctx);
1090
1091                 lu_device_get(&d->dd_lu_dev);
1092                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1093                               "osd-tx", th);
1094                 oti->oti_txns++;
1095                 rc = 0;
1096         } else {
1097                 rc = PTR_ERR(jh);
1098         }
1099 out:
1100         RETURN(rc);
1101 }
1102
1103 static int osd_seq_exists(const struct lu_env *env,
1104                           struct osd_device *osd, u64 seq)
1105 {
1106         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1107         struct seq_server_site  *ss = osd_seq_site(osd);
1108         int                     rc;
1109         ENTRY;
1110
1111         LASSERT(ss != NULL);
1112         LASSERT(ss->ss_server_fld != NULL);
1113
1114         rc = osd_fld_lookup(env, osd, seq, range);
1115         if (rc != 0) {
1116                 if (rc != -ENOENT)
1117                         CERROR("%s: can't lookup FLD sequence "LPX64
1118                                ": rc = %d\n", osd_name(osd), seq, rc);
1119                 RETURN(0);
1120         }
1121
1122         RETURN(ss->ss_node_id == range->lsr_index);
1123 }
1124
1125 /*
1126  * Concurrency: shouldn't matter.
1127  */
1128 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1129                           struct thandle *th)
1130 {
1131         int                     rc = 0;
1132         struct osd_thandle     *oh;
1133         struct osd_thread_info *oti = osd_oti_get(env);
1134         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1135         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1136         struct lquota_trans    *qtrans;
1137         ENTRY;
1138
1139         oh = container_of0(th, struct osd_thandle, ot_super);
1140
1141         qtrans = oh->ot_quota_trans;
1142         oh->ot_quota_trans = NULL;
1143
1144         if (oh->ot_handle != NULL) {
1145                 handle_t *hdl = oh->ot_handle;
1146
1147                 /*
1148                  * add commit callback
1149                  * notice we don't do this in osd_trans_start()
1150                  * as underlying transaction can change during truncate
1151                  */
1152                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1153                                          &oh->ot_jcb);
1154
1155                 LASSERT(oti->oti_txns == 1);
1156                 oti->oti_txns--;
1157                 rc = dt_txn_hook_stop(env, th);
1158                 if (rc != 0)
1159                         CERROR("Failure in transaction hook: %d\n", rc);
1160
1161                 /* hook functions might modify th_sync */
1162                 hdl->h_sync = th->th_sync;
1163
1164                 oh->ot_handle = NULL;
1165                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1166                                   rc = ldiskfs_journal_stop(hdl));
1167                 if (rc != 0)
1168                         CERROR("Failure to stop transaction: %d\n", rc);
1169         } else {
1170                 thandle_put(&oh->ot_super);
1171         }
1172
1173         /* inform the quota slave device that the transaction is stopping */
1174         qsd_op_end(env, qsd, qtrans);
1175
1176         /* as we want IO to journal and data IO be concurrent, we don't block
1177          * awaiting data IO completion in osd_do_bio(), instead we wait here
1178          * once transaction is submitted to the journal. all reqular requests
1179          * don't do direct IO (except read/write), thus this wait_event becomes
1180          * no-op for them.
1181          *
1182          * IMPORTANT: we have to wait till any IO submited by the thread is
1183          * completed otherwise iobuf may be corrupted by different request
1184          */
1185         wait_event(iobuf->dr_wait,
1186                        atomic_read(&iobuf->dr_numreqs) == 0);
1187         osd_fini_iobuf(oti->oti_dev, iobuf);
1188         if (!rc)
1189                 rc = iobuf->dr_error;
1190
1191         RETURN(rc);
1192 }
1193
1194 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1195 {
1196         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1197                                                ot_super);
1198
1199         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1200         LASSERT(&dcb->dcb_func != NULL);
1201         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1202
1203         return 0;
1204 }
1205
1206 /*
1207  * Called just before object is freed. Releases all resources except for
1208  * object itself (that is released by osd_object_free()).
1209  *
1210  * Concurrency: no concurrent access is possible that late in object
1211  * life-cycle.
1212  */
1213 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1214 {
1215         struct osd_object *obj   = osd_obj(l);
1216         struct inode      *inode = obj->oo_inode;
1217
1218         LINVRNT(osd_invariant(obj));
1219
1220         /*
1221          * If object is unlinked remove fid->ino mapping from object index.
1222          */
1223
1224         osd_index_fini(obj);
1225         if (inode != NULL) {
1226                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1227                 qid_t                    uid = i_uid_read(inode);
1228                 qid_t                    gid = i_gid_read(inode);
1229
1230                 iput(inode);
1231                 obj->oo_inode = NULL;
1232
1233                 if (qsd != NULL) {
1234                         struct osd_thread_info  *info = osd_oti_get(env);
1235                         struct lquota_id_info   *qi = &info->oti_qi;
1236
1237                         /* Release granted quota to master if necessary */
1238                         qi->lqi_id.qid_uid = uid;
1239                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1240
1241                         qi->lqi_id.qid_uid = gid;
1242                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1243                 }
1244         }
1245 }
1246
1247 /*
1248  * Concurrency: ->loo_object_release() is called under site spin-lock.
1249  */
1250 static void osd_object_release(const struct lu_env *env,
1251                                struct lu_object *l)
1252 {
1253 }
1254
1255 /*
1256  * Concurrency: shouldn't matter.
1257  */
1258 static int osd_object_print(const struct lu_env *env, void *cookie,
1259                             lu_printer_t p, const struct lu_object *l)
1260 {
1261         struct osd_object *o = osd_obj(l);
1262         struct iam_descr  *d;
1263
1264         if (o->oo_dir != NULL)
1265                 d = o->oo_dir->od_container.ic_descr;
1266         else
1267                 d = NULL;
1268         return (*p)(env, cookie,
1269                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1270                     o, o->oo_inode,
1271                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1272                     o->oo_inode ? o->oo_inode->i_generation : 0,
1273                     d ? d->id_ops->id_name : "plain");
1274 }
1275
1276 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1277
1278 /*
1279  * Concurrency: shouldn't matter.
1280  */
1281 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1282                struct obd_statfs *sfs)
1283 {
1284         struct osd_device  *osd = osd_dt_dev(d);
1285         struct super_block *sb = osd_sb(osd);
1286         struct kstatfs     *ksfs;
1287         int result = 0;
1288
1289         if (unlikely(osd->od_mnt == NULL))
1290                 return -EINPROGRESS;
1291
1292         /* osd_lproc.c call this without env, allocate ksfs for that case */
1293         if (unlikely(env == NULL)) {
1294                 OBD_ALLOC_PTR(ksfs);
1295                 if (ksfs == NULL)
1296                         return -ENOMEM;
1297         } else {
1298                 ksfs = &osd_oti_get(env)->oti_ksfs;
1299         }
1300
1301         spin_lock(&osd->od_osfs_lock);
1302         result = sb->s_op->statfs(sb->s_root, ksfs);
1303         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1304                 statfs_pack(sfs, ksfs);
1305                 if (sb->s_flags & MS_RDONLY)
1306                         sfs->os_state = OS_STATE_READONLY;
1307         }
1308
1309         spin_unlock(&osd->od_osfs_lock);
1310
1311         if (unlikely(env == NULL))
1312                 OBD_FREE_PTR(ksfs);
1313
1314         /* Reserve a small amount of space for local objects like last_rcvd,
1315          * llog, quota files, ... */
1316         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1317                 sfs->os_bavail = 0;
1318         } else {
1319                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1320                 /** Take out metadata overhead for indirect blocks */
1321                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1322         }
1323
1324         return result;
1325 }
1326
1327 /**
1328  * Estimate space needed for file creations. We assume the largest filename
1329  * which is 2^64 - 1, hence a filename of 20 chars.
1330  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1331  */
1332 #ifdef __LDISKFS_DIR_REC_LEN
1333 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1334 #else
1335 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1336 #endif
1337
1338 /*
1339  * Concurrency: doesn't access mutable data.
1340  */
1341 static void osd_conf_get(const struct lu_env *env,
1342                          const struct dt_device *dev,
1343                          struct dt_device_param *param)
1344 {
1345         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1346         int                ea_overhead;
1347
1348         /*
1349          * XXX should be taken from not-yet-existing fs abstraction layer.
1350          */
1351         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1352         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1353         param->ddp_block_shift  = sb->s_blocksize_bits;
1354         param->ddp_mount_type     = LDD_MT_LDISKFS;
1355         param->ddp_maxbytes       = sb->s_maxbytes;
1356         /* Overhead estimate should be fairly accurate, so we really take a tiny
1357          * error margin which also avoids fragmenting the filesystem too much */
1358         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1359         /* inode are statically allocated, so per-inode space consumption
1360          * is the space consumed by the directory entry */
1361         param->ddp_inodespace     = PER_OBJ_USAGE;
1362         /* per-fragment overhead to be used by the client code */
1363         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1364         param->ddp_mntopts      = 0;
1365         if (test_opt(sb, XATTR_USER))
1366                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1367         if (test_opt(sb, POSIX_ACL))
1368                 param->ddp_mntopts |= MNTOPT_ACL;
1369
1370         /* LOD might calculate the max stripe count based on max_ea_size,
1371          * so we need take account in the overhead as well,
1372          * xattr_header + magic + xattr_entry_head */
1373         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1374                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1375
1376 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1377         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1378                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1379                                                                 ea_overhead;
1380         else
1381 #endif
1382                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1383 }
1384
1385 /*
1386  * Concurrency: shouldn't matter.
1387  */
1388 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1389 {
1390         int rc;
1391
1392         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1393
1394         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1395
1396         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1397                LUSTRE_OSD_LDISKFS_NAME, rc);
1398
1399         return rc;
1400 }
1401
1402 /**
1403  * Start commit for OSD device.
1404  *
1405  * An implementation of dt_commit_async method for OSD device.
1406  * Asychronously starts underlayng fs sync and thereby a transaction
1407  * commit.
1408  *
1409  * \param env environment
1410  * \param d dt device
1411  *
1412  * \see dt_device_operations
1413  */
1414 static int osd_commit_async(const struct lu_env *env,
1415                             struct dt_device *d)
1416 {
1417         struct super_block *s = osd_sb(osd_dt_dev(d));
1418         ENTRY;
1419
1420         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1421         RETURN(s->s_op->sync_fs(s, 0));
1422 }
1423
1424 /*
1425  * Concurrency: shouldn't matter.
1426  */
1427
1428 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1429 {
1430         struct super_block *sb = osd_sb(osd_dt_dev(d));
1431         struct block_device *dev = sb->s_bdev;
1432 #ifdef HAVE_DEV_SET_RDONLY
1433         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1434         int rc = 0;
1435 #else
1436         int rc = -EOPNOTSUPP;
1437 #endif
1438         ENTRY;
1439
1440 #ifdef HAVE_DEV_SET_RDONLY
1441         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1442
1443         if (jdev && (jdev != dev)) {
1444                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1445                        (long)jdev);
1446                 dev_set_rdonly(jdev);
1447         }
1448         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1449         dev_set_rdonly(dev);
1450 #else
1451         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1452                osd_dt_dev(d)->od_svname, (long)dev, rc);
1453 #endif
1454         RETURN(rc);
1455 }
1456
1457 /*
1458  * Concurrency: serialization provided by callers.
1459  */
1460 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1461                               int mode, unsigned long timeout, __u32 alg,
1462                               struct lustre_capa_key *keys)
1463 {
1464         struct osd_device *dev = osd_dt_dev(d);
1465         ENTRY;
1466
1467         dev->od_fl_capa = mode;
1468         dev->od_capa_timeout = timeout;
1469         dev->od_capa_alg = alg;
1470         dev->od_capa_keys = keys;
1471         RETURN(0);
1472 }
1473
1474 /**
1475  * Note: we do not count into QUOTA here.
1476  * If we mount with --data_journal we may need more.
1477  */
1478 const int osd_dto_credits_noquota[DTO_NR] = {
1479         /**
1480          * Insert.
1481          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1482          * SINGLEDATA_TRANS_BLOCKS(8)
1483          * XXX Note: maybe iam need more, since iam have more level than
1484          *           EXT3 htree.
1485          */
1486         [DTO_INDEX_INSERT]  = 16,
1487         /**
1488          * Delete
1489          * just modify a single entry, probably merge few within a block
1490          */
1491         [DTO_INDEX_DELETE]  = 1,
1492         /**
1493          * Used for OI scrub
1494          */
1495         [DTO_INDEX_UPDATE]  = 16,
1496         /**
1497          * 4(inode, inode bits, groups, GDT)
1498          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1499          */
1500         [DTO_OBJECT_CREATE] = 4,
1501         /**
1502          * 4(inode, inode bits, groups, GDT)
1503          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1504          */
1505         [DTO_OBJECT_DELETE] = 4,
1506         /**
1507          * Attr set credits (inode)
1508          */
1509         [DTO_ATTR_SET_BASE] = 1,
1510         /**
1511          * Xattr set. The same as xattr of EXT3.
1512          * DATA_TRANS_BLOCKS(14)
1513          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1514          * are also counted in. Do not know why?
1515          */
1516         [DTO_XATTR_SET]     = 14,
1517         /**
1518          * credits for inode change during write.
1519          */
1520         [DTO_WRITE_BASE]    = 3,
1521         /**
1522          * credits for single block write.
1523          */
1524         [DTO_WRITE_BLOCK]   = 14,
1525         /**
1526          * Attr set credits for chown.
1527          * This is extra credits for setattr, and it is null without quota
1528          */
1529         [DTO_ATTR_SET_CHOWN] = 0
1530 };
1531
1532 static const struct dt_device_operations osd_dt_ops = {
1533         .dt_root_get       = osd_root_get,
1534         .dt_statfs         = osd_statfs,
1535         .dt_trans_create   = osd_trans_create,
1536         .dt_trans_start    = osd_trans_start,
1537         .dt_trans_stop     = osd_trans_stop,
1538         .dt_trans_cb_add   = osd_trans_cb_add,
1539         .dt_conf_get       = osd_conf_get,
1540         .dt_sync           = osd_sync,
1541         .dt_ro             = osd_ro,
1542         .dt_commit_async   = osd_commit_async,
1543         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1544 };
1545
1546 static void osd_object_read_lock(const struct lu_env *env,
1547                                  struct dt_object *dt, unsigned role)
1548 {
1549         struct osd_object *obj = osd_dt_obj(dt);
1550         struct osd_thread_info *oti = osd_oti_get(env);
1551
1552         LINVRNT(osd_invariant(obj));
1553
1554         LASSERT(obj->oo_owner != env);
1555         down_read_nested(&obj->oo_sem, role);
1556
1557         LASSERT(obj->oo_owner == NULL);
1558         oti->oti_r_locks++;
1559 }
1560
1561 static void osd_object_write_lock(const struct lu_env *env,
1562                                   struct dt_object *dt, unsigned role)
1563 {
1564         struct osd_object *obj = osd_dt_obj(dt);
1565         struct osd_thread_info *oti = osd_oti_get(env);
1566
1567         LINVRNT(osd_invariant(obj));
1568
1569         LASSERT(obj->oo_owner != env);
1570         down_write_nested(&obj->oo_sem, role);
1571
1572         LASSERT(obj->oo_owner == NULL);
1573         obj->oo_owner = env;
1574         oti->oti_w_locks++;
1575 }
1576
1577 static void osd_object_read_unlock(const struct lu_env *env,
1578                                    struct dt_object *dt)
1579 {
1580         struct osd_object *obj = osd_dt_obj(dt);
1581         struct osd_thread_info *oti = osd_oti_get(env);
1582
1583         LINVRNT(osd_invariant(obj));
1584
1585         LASSERT(oti->oti_r_locks > 0);
1586         oti->oti_r_locks--;
1587         up_read(&obj->oo_sem);
1588 }
1589
1590 static void osd_object_write_unlock(const struct lu_env *env,
1591                                     struct dt_object *dt)
1592 {
1593         struct osd_object *obj = osd_dt_obj(dt);
1594         struct osd_thread_info *oti = osd_oti_get(env);
1595
1596         LINVRNT(osd_invariant(obj));
1597
1598         LASSERT(obj->oo_owner == env);
1599         LASSERT(oti->oti_w_locks > 0);
1600         oti->oti_w_locks--;
1601         obj->oo_owner = NULL;
1602         up_write(&obj->oo_sem);
1603 }
1604
1605 static int osd_object_write_locked(const struct lu_env *env,
1606                                    struct dt_object *dt)
1607 {
1608         struct osd_object *obj = osd_dt_obj(dt);
1609
1610         LINVRNT(osd_invariant(obj));
1611
1612         return obj->oo_owner == env;
1613 }
1614
1615 static int capa_is_sane(const struct lu_env *env,
1616                         struct osd_device *dev,
1617                         struct lustre_capa *capa,
1618                         struct lustre_capa_key *keys)
1619 {
1620         struct osd_thread_info *oti = osd_oti_get(env);
1621         struct lustre_capa *tcapa = &oti->oti_capa;
1622         struct obd_capa *oc;
1623         int i, rc = 0;
1624         ENTRY;
1625
1626         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1627         if (oc) {
1628                 if (capa_is_expired(oc)) {
1629                         DEBUG_CAPA(D_ERROR, capa, "expired");
1630                         rc = -ESTALE;
1631                 }
1632                 capa_put(oc);
1633                 RETURN(rc);
1634         }
1635
1636         if (capa_is_expired_sec(capa)) {
1637                 DEBUG_CAPA(D_ERROR, capa, "expired");
1638                 RETURN(-ESTALE);
1639         }
1640
1641         spin_lock(&capa_lock);
1642         for (i = 0; i < 2; i++) {
1643                 if (keys[i].lk_keyid == capa->lc_keyid) {
1644                         oti->oti_capa_key = keys[i];
1645                         break;
1646                 }
1647         }
1648         spin_unlock(&capa_lock);
1649
1650         if (i == 2) {
1651                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1652                 RETURN(-ESTALE);
1653         }
1654
1655         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1656         if (rc)
1657                 RETURN(rc);
1658
1659         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1660                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1661                 RETURN(-EACCES);
1662         }
1663
1664         oc = capa_add(dev->od_capa_hash, capa);
1665         capa_put(oc);
1666
1667         RETURN(0);
1668 }
1669
1670 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1671                     struct lustre_capa *capa, __u64 opc)
1672 {
1673         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1674         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1675         struct lu_capainfo *lci;
1676         int rc;
1677
1678         if (!osd->od_fl_capa)
1679                 return 0;
1680
1681         if (capa == BYPASS_CAPA)
1682                 return 0;
1683
1684         lci = lu_capainfo_get(env);
1685         if (unlikely(lci == NULL))
1686                 return 0;
1687
1688         if (lci->lci_auth == LC_ID_NONE)
1689                 return 0;
1690
1691         if (capa == NULL) {
1692                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1693                        osd_name(osd), PFID(fid), -EACCES);
1694                 return -EACCES;
1695         }
1696
1697         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1698                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1699                            PFID(fid));
1700                 return -EACCES;
1701         }
1702
1703         if (!capa_opc_supported(capa, opc)) {
1704                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1705                 return -EACCES;
1706         }
1707
1708         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1709         if (rc != 0) {
1710                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1711                 return -EACCES;
1712         }
1713
1714         return 0;
1715 }
1716
1717 static struct timespec *osd_inode_time(const struct lu_env *env,
1718                                        struct inode *inode, __u64 seconds)
1719 {
1720         struct osd_thread_info  *oti = osd_oti_get(env);
1721         struct timespec         *t   = &oti->oti_time;
1722
1723         t->tv_sec = seconds;
1724         t->tv_nsec = 0;
1725         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1726         return t;
1727 }
1728
1729
1730 static void osd_inode_getattr(const struct lu_env *env,
1731                               struct inode *inode, struct lu_attr *attr)
1732 {
1733         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1734                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1735                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1736                            LA_TYPE;
1737
1738         attr->la_atime   = LTIME_S(inode->i_atime);
1739         attr->la_mtime   = LTIME_S(inode->i_mtime);
1740         attr->la_ctime   = LTIME_S(inode->i_ctime);
1741         attr->la_mode    = inode->i_mode;
1742         attr->la_size    = i_size_read(inode);
1743         attr->la_blocks  = inode->i_blocks;
1744         attr->la_uid     = i_uid_read(inode);
1745         attr->la_gid     = i_gid_read(inode);
1746         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1747         attr->la_nlink   = inode->i_nlink;
1748         attr->la_rdev    = inode->i_rdev;
1749         attr->la_blksize = 1 << inode->i_blkbits;
1750         attr->la_blkbits = inode->i_blkbits;
1751 }
1752
1753 static int osd_attr_get(const struct lu_env *env,
1754                         struct dt_object *dt,
1755                         struct lu_attr *attr,
1756                         struct lustre_capa *capa)
1757 {
1758         struct osd_object *obj = osd_dt_obj(dt);
1759
1760         if (!dt_object_exists(dt))
1761                 return -ENOENT;
1762
1763         LASSERT(!dt_object_remote(dt));
1764         LINVRNT(osd_invariant(obj));
1765
1766         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1767                 return -EACCES;
1768
1769         spin_lock(&obj->oo_guard);
1770         osd_inode_getattr(env, obj->oo_inode, attr);
1771         spin_unlock(&obj->oo_guard);
1772         return 0;
1773 }
1774
1775 static int osd_declare_attr_set(const struct lu_env *env,
1776                                 struct dt_object *dt,
1777                                 const struct lu_attr *attr,
1778                                 struct thandle *handle)
1779 {
1780         struct osd_thandle     *oh;
1781         struct osd_object      *obj;
1782         struct osd_thread_info *info = osd_oti_get(env);
1783         struct lquota_id_info  *qi = &info->oti_qi;
1784         qid_t                   uid;
1785         qid_t                   gid;
1786         long long               bspace;
1787         int                     rc = 0;
1788         bool                    enforce;
1789         ENTRY;
1790
1791         LASSERT(dt != NULL);
1792         LASSERT(handle != NULL);
1793
1794         obj = osd_dt_obj(dt);
1795         LASSERT(osd_invariant(obj));
1796
1797         oh = container_of0(handle, struct osd_thandle, ot_super);
1798         LASSERT(oh->ot_handle == NULL);
1799
1800         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1801                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1802
1803         if (attr == NULL || obj->oo_inode == NULL)
1804                 RETURN(rc);
1805
1806         bspace   = obj->oo_inode->i_blocks;
1807         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1808         bspace   = toqb(bspace);
1809
1810         /* Changing ownership is always preformed by super user, it should not
1811          * fail with EDQUOT.
1812          *
1813          * We still need to call the osd_declare_qid() to calculate the journal
1814          * credits for updating quota accounting files and to trigger quota
1815          * space adjustment once the operation is completed.*/
1816         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1817                 /* USERQUOTA */
1818                 uid = i_uid_read(obj->oo_inode);
1819                 qi->lqi_type = USRQUOTA;
1820                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1821                 /* inode accounting */
1822                 qi->lqi_is_blk = false;
1823
1824                 /* one more inode for the new uid ... */
1825                 qi->lqi_id.qid_uid = attr->la_uid;
1826                 qi->lqi_space      = 1;
1827                 /* Reserve credits for the new uid */
1828                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1829                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1830                         rc = 0;
1831                 if (rc)
1832                         RETURN(rc);
1833
1834                 /* and one less inode for the current uid */
1835                 qi->lqi_id.qid_uid = uid;
1836                 qi->lqi_space      = -1;
1837                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1838                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1839                         rc = 0;
1840                 if (rc)
1841                         RETURN(rc);
1842
1843                 /* block accounting */
1844                 qi->lqi_is_blk = true;
1845
1846                 /* more blocks for the new uid ... */
1847                 qi->lqi_id.qid_uid = attr->la_uid;
1848                 qi->lqi_space      = bspace;
1849                 /*
1850                  * Credits for the new uid has been reserved, re-use "obj"
1851                  * to save credit reservation.
1852                  */
1853                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1854                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1855                         rc = 0;
1856                 if (rc)
1857                         RETURN(rc);
1858
1859                 /* and finally less blocks for the current uid */
1860                 qi->lqi_id.qid_uid = uid;
1861                 qi->lqi_space      = -bspace;
1862                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1863                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1864                         rc = 0;
1865                 if (rc)
1866                         RETURN(rc);
1867
1868                 /* GROUP QUOTA */
1869                 gid = i_gid_read(obj->oo_inode);
1870                 qi->lqi_type = GRPQUOTA;
1871                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1872
1873                 /* inode accounting */
1874                 qi->lqi_is_blk = false;
1875
1876                 /* one more inode for the new gid ... */
1877                 qi->lqi_id.qid_gid = attr->la_gid;
1878                 qi->lqi_space      = 1;
1879                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1880                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1881                         rc = 0;
1882                 if (rc)
1883                         RETURN(rc);
1884
1885                 /* and one less inode for the current gid */
1886                 qi->lqi_id.qid_gid = gid;
1887                 qi->lqi_space      = -1;
1888                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1889                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1890                         rc = 0;
1891                 if (rc)
1892                         RETURN(rc);
1893
1894                 /* block accounting */
1895                 qi->lqi_is_blk = true;
1896
1897                 /* more blocks for the new gid ... */
1898                 qi->lqi_id.qid_gid = attr->la_gid;
1899                 qi->lqi_space      = bspace;
1900                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1901                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1902                         rc = 0;
1903                 if (rc)
1904                         RETURN(rc);
1905
1906                 /* and finally less blocks for the current gid */
1907                 qi->lqi_id.qid_gid = gid;
1908                 qi->lqi_space      = -bspace;
1909                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1910                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1911                         rc = 0;
1912                 if (rc)
1913                         RETURN(rc);
1914         }
1915
1916         RETURN(rc);
1917 }
1918
1919 static int osd_inode_setattr(const struct lu_env *env,
1920                              struct inode *inode, const struct lu_attr *attr)
1921 {
1922         __u64 bits = attr->la_valid;
1923
1924         /* Only allow set size for regular file */
1925         if (!S_ISREG(inode->i_mode))
1926                 bits &= ~(LA_SIZE | LA_BLOCKS);
1927
1928         if (bits == 0)
1929                 return 0;
1930
1931         if (bits & LA_ATIME)
1932                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1933         if (bits & LA_CTIME)
1934                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1935         if (bits & LA_MTIME)
1936                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1937         if (bits & LA_SIZE) {
1938                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1939                 i_size_write(inode, attr->la_size);
1940         }
1941
1942 #if 0
1943         /* OSD should not change "i_blocks" which is used by quota.
1944          * "i_blocks" should be changed by ldiskfs only. */
1945         if (bits & LA_BLOCKS)
1946                 inode->i_blocks = attr->la_blocks;
1947 #endif
1948         if (bits & LA_MODE)
1949                 inode->i_mode = (inode->i_mode & S_IFMT) |
1950                                 (attr->la_mode & ~S_IFMT);
1951         if (bits & LA_UID)
1952                 i_uid_write(inode, attr->la_uid);
1953         if (bits & LA_GID)
1954                 i_gid_write(inode, attr->la_gid);
1955         if (bits & LA_NLINK)
1956                 set_nlink(inode, attr->la_nlink);
1957         if (bits & LA_RDEV)
1958                 inode->i_rdev = attr->la_rdev;
1959
1960         if (bits & LA_FLAGS) {
1961                 /* always keep S_NOCMTIME */
1962                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1963                                  S_NOCMTIME;
1964         }
1965         return 0;
1966 }
1967
1968 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1969 {
1970         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1971             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1972                 struct iattr    iattr;
1973                 int             rc;
1974
1975                 ll_vfs_dq_init(inode);
1976                 iattr.ia_valid = 0;
1977                 if (attr->la_valid & LA_UID)
1978                         iattr.ia_valid |= ATTR_UID;
1979                 if (attr->la_valid & LA_GID)
1980                         iattr.ia_valid |= ATTR_GID;
1981                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1982                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1983
1984                 rc = ll_vfs_dq_transfer(inode, &iattr);
1985                 if (rc) {
1986                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1987                                "enforcement enabled on the ldiskfs "
1988                                "filesystem?\n", inode->i_sb->s_id, rc);
1989                         return rc;
1990                 }
1991         }
1992         return 0;
1993 }
1994
1995 static int osd_attr_set(const struct lu_env *env,
1996                         struct dt_object *dt,
1997                         const struct lu_attr *attr,
1998                         struct thandle *handle,
1999                         struct lustre_capa *capa)
2000 {
2001         struct osd_object *obj = osd_dt_obj(dt);
2002         struct inode      *inode;
2003         int rc;
2004
2005         if (!dt_object_exists(dt))
2006                 return -ENOENT;
2007
2008         LASSERT(handle != NULL);
2009         LASSERT(!dt_object_remote(dt));
2010         LASSERT(osd_invariant(obj));
2011
2012         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2013                 return -EACCES;
2014
2015         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2016
2017         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2018                 struct osd_thread_info  *oti  = osd_oti_get(env);
2019                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2020                 struct lu_fid           *fid1 = &oti->oti_fid;
2021                 struct osd_inode_id     *id   = &oti->oti_id;
2022                 struct iam_path_descr   *ipd;
2023                 struct iam_container    *bag;
2024                 struct osd_thandle      *oh;
2025                 int                      rc;
2026
2027                 fid_cpu_to_be(fid1, fid0);
2028                 memset(id, 1, sizeof(*id));
2029                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2030                                   fid0)->oi_dir.od_container;
2031                 ipd = osd_idx_ipd_get(env, bag);
2032                 if (unlikely(ipd == NULL))
2033                         RETURN(-ENOMEM);
2034
2035                 oh = container_of0(handle, struct osd_thandle, ot_super);
2036                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2037                                 (const struct iam_rec *)id, ipd);
2038                 osd_ipd_put(env, bag, ipd);
2039                 return(rc > 0 ? 0 : rc);
2040         }
2041
2042         inode = obj->oo_inode;
2043
2044         rc = osd_quota_transfer(inode, attr);
2045         if (rc)
2046                 return rc;
2047
2048         spin_lock(&obj->oo_guard);
2049         rc = osd_inode_setattr(env, inode, attr);
2050         spin_unlock(&obj->oo_guard);
2051
2052         if (!rc)
2053                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2054         return rc;
2055 }
2056
2057 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2058                                            struct osd_object *obj,
2059                                            const char *name, const int namelen)
2060 {
2061         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2062 }
2063
2064 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2065                       umode_t mode, struct dt_allocation_hint *hint,
2066                       struct thandle *th)
2067 {
2068         int result;
2069         struct osd_device  *osd = osd_obj2dev(obj);
2070         struct osd_thandle *oth;
2071         struct dt_object   *parent = NULL;
2072         struct inode       *inode;
2073
2074         LINVRNT(osd_invariant(obj));
2075         LASSERT(obj->oo_inode == NULL);
2076         LASSERT(obj->oo_hl_head == NULL);
2077
2078         if (S_ISDIR(mode) && ldiskfs_pdo) {
2079                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2080                 if (obj->oo_hl_head == NULL)
2081                         return -ENOMEM;
2082         }
2083
2084         oth = container_of(th, struct osd_thandle, ot_super);
2085         LASSERT(oth->ot_handle->h_transaction != NULL);
2086
2087         if (hint && hint->dah_parent)
2088                 parent = hint->dah_parent;
2089
2090         inode = ldiskfs_create_inode(oth->ot_handle,
2091                                      parent ? osd_dt_obj(parent)->oo_inode :
2092                                               osd_sb(osd)->s_root->d_inode,
2093                                      mode);
2094         if (!IS_ERR(inode)) {
2095                 /* Do not update file c/mtime in ldiskfs.
2096                  * NB: don't need any lock because no contention at this
2097                  * early stage */
2098                 inode->i_flags |= S_NOCMTIME;
2099
2100                 /* For new created object, it must be consistent,
2101                  * and it is unnecessary to scrub against it. */
2102                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2103                 obj->oo_inode = inode;
2104                 result = 0;
2105         } else {
2106                 if (obj->oo_hl_head != NULL) {
2107                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2108                         obj->oo_hl_head = NULL;
2109                 }
2110                 result = PTR_ERR(inode);
2111         }
2112         LINVRNT(osd_invariant(obj));
2113         return result;
2114 }
2115
2116 enum {
2117         OSD_NAME_LEN = 255
2118 };
2119
2120 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2121                      struct lu_attr *attr,
2122                      struct dt_allocation_hint *hint,
2123                      struct dt_object_format *dof,
2124                      struct thandle *th)
2125 {
2126         int result;
2127         struct osd_thandle *oth;
2128         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2129
2130         LASSERT(S_ISDIR(attr->la_mode));
2131
2132         oth = container_of(th, struct osd_thandle, ot_super);
2133         LASSERT(oth->ot_handle->h_transaction != NULL);
2134         result = osd_mkfile(info, obj, mode, hint, th);
2135
2136         return result;
2137 }
2138
2139 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2140                         struct lu_attr *attr,
2141                         struct dt_allocation_hint *hint,
2142                         struct dt_object_format *dof,
2143                         struct thandle *th)
2144 {
2145         int result;
2146         struct osd_thandle *oth;
2147         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2148
2149         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2150
2151         LASSERT(S_ISREG(attr->la_mode));
2152
2153         oth = container_of(th, struct osd_thandle, ot_super);
2154         LASSERT(oth->ot_handle->h_transaction != NULL);
2155
2156         result = osd_mkfile(info, obj, mode, hint, th);
2157         if (result == 0) {
2158                 LASSERT(obj->oo_inode != NULL);
2159                 if (feat->dif_flags & DT_IND_VARKEY)
2160                         result = iam_lvar_create(obj->oo_inode,
2161                                                  feat->dif_keysize_max,
2162                                                  feat->dif_ptrsize,
2163                                                  feat->dif_recsize_max,
2164                                                  oth->ot_handle);
2165                 else
2166                         result = iam_lfix_create(obj->oo_inode,
2167                                                  feat->dif_keysize_max,
2168                                                  feat->dif_ptrsize,
2169                                                  feat->dif_recsize_max,
2170                                                  oth->ot_handle);
2171
2172         }
2173         return result;
2174 }
2175
2176 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2177                      struct lu_attr *attr,
2178                      struct dt_allocation_hint *hint,
2179                      struct dt_object_format *dof,
2180                      struct thandle *th)
2181 {
2182         LASSERT(S_ISREG(attr->la_mode));
2183         return osd_mkfile(info, obj, (attr->la_mode &
2184                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2185 }
2186
2187 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2188                      struct lu_attr *attr,
2189                      struct dt_allocation_hint *hint,
2190                      struct dt_object_format *dof,
2191                      struct thandle *th)
2192 {
2193         LASSERT(S_ISLNK(attr->la_mode));
2194         return osd_mkfile(info, obj, (attr->la_mode &
2195                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2196 }
2197
2198 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2199                      struct lu_attr *attr,
2200                      struct dt_allocation_hint *hint,
2201                      struct dt_object_format *dof,
2202                      struct thandle *th)
2203 {
2204         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2205         int result;
2206
2207         LINVRNT(osd_invariant(obj));
2208         LASSERT(obj->oo_inode == NULL);
2209         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2210                 S_ISFIFO(mode) || S_ISSOCK(mode));
2211
2212         result = osd_mkfile(info, obj, mode, hint, th);
2213         if (result == 0) {
2214                 LASSERT(obj->oo_inode != NULL);
2215                 /*
2216                  * This inode should be marked dirty for i_rdev.  Currently
2217                  * that is done in the osd_attr_init().
2218                  */
2219                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2220                                    attr->la_rdev);
2221         }
2222         LINVRNT(osd_invariant(obj));
2223         return result;
2224 }
2225
2226 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2227                               struct lu_attr *,
2228                               struct dt_allocation_hint *hint,
2229                               struct dt_object_format *dof,
2230                               struct thandle *);
2231
2232 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2233 {
2234         osd_obj_type_f result;
2235
2236         switch (type) {
2237         case DFT_DIR:
2238                 result = osd_mkdir;
2239                 break;
2240         case DFT_REGULAR:
2241                 result = osd_mkreg;
2242                 break;
2243         case DFT_SYM:
2244                 result = osd_mksym;
2245                 break;
2246         case DFT_NODE:
2247                 result = osd_mknod;
2248                 break;
2249         case DFT_INDEX:
2250                 result = osd_mk_index;
2251                 break;
2252
2253         default:
2254                 LBUG();
2255                 break;
2256         }
2257         return result;
2258 }
2259
2260
2261 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2262                         struct dt_object *parent, struct dt_object *child,
2263                         umode_t child_mode)
2264 {
2265         LASSERT(ah);
2266
2267         ah->dah_parent = parent;
2268         ah->dah_mode = child_mode;
2269 }
2270
2271 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2272                           struct lu_attr *attr, struct dt_object_format *dof)
2273 {
2274         struct inode   *inode = obj->oo_inode;
2275         __u64           valid = attr->la_valid;
2276         int             result;
2277
2278         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2279
2280         if (dof->dof_type != DFT_NODE)
2281                 attr->la_valid &= ~LA_RDEV;
2282         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2283                 attr->la_valid &= ~LA_ATIME;
2284         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2285                 attr->la_valid &= ~LA_CTIME;
2286         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2287                 attr->la_valid &= ~LA_MTIME;
2288
2289         result = osd_quota_transfer(inode, attr);
2290         if (result)
2291                 return;
2292
2293         if (attr->la_valid != 0) {
2294                 result = osd_inode_setattr(info->oti_env, inode, attr);
2295                 /*
2296                  * The osd_inode_setattr() should always succeed here.  The
2297                  * only error that could be returned is EDQUOT when we are
2298                  * trying to change the UID or GID of the inode. However, this
2299                  * should not happen since quota enforcement is no longer
2300                  * enabled on ldiskfs (lquota takes care of it).
2301                  */
2302                 LASSERTF(result == 0, "%d\n", result);
2303                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2304         }
2305
2306         attr->la_valid = valid;
2307 }
2308
2309 /**
2310  * Helper function for osd_object_create()
2311  *
2312  * \retval 0, on success
2313  */
2314 static int __osd_object_create(struct osd_thread_info *info,
2315                                struct osd_object *obj, struct lu_attr *attr,
2316                                struct dt_allocation_hint *hint,
2317                                struct dt_object_format *dof,
2318                                struct thandle *th)
2319 {
2320         int     result;
2321         __u32   umask;
2322
2323         /* we drop umask so that permissions we pass are not affected */
2324         umask = current->fs->umask;
2325         current->fs->umask = 0;
2326
2327         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2328                                                   th);
2329         if (result == 0) {
2330                 osd_attr_init(info, obj, attr, dof);
2331                 osd_object_init0(obj);
2332                 /* bz 24037 */
2333                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2334                         unlock_new_inode(obj->oo_inode);
2335         }
2336
2337         /* restore previous umask value */
2338         current->fs->umask = umask;
2339
2340         return result;
2341 }
2342
2343 /**
2344  * Helper function for osd_object_create()
2345  *
2346  * \retval 0, on success
2347  */
2348 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2349                            const struct lu_fid *fid, struct thandle *th)
2350 {
2351         struct osd_thread_info *info = osd_oti_get(env);
2352         struct osd_inode_id    *id   = &info->oti_id;
2353         struct osd_device      *osd  = osd_obj2dev(obj);
2354         struct osd_thandle     *oh;
2355
2356         LASSERT(obj->oo_inode != NULL);
2357
2358         oh = container_of0(th, struct osd_thandle, ot_super);
2359         LASSERT(oh->ot_handle);
2360
2361         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2362         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2363 }
2364
2365 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2366                    u64 seq, struct lu_seq_range *range)
2367 {
2368         struct seq_server_site  *ss = osd_seq_site(osd);
2369
2370         if (fid_seq_is_idif(seq)) {
2371                 fld_range_set_ost(range);
2372                 range->lsr_index = idif_ost_idx(seq);
2373                 return 0;
2374         }
2375
2376         if (!fid_seq_in_fldb(seq)) {
2377                 fld_range_set_mdt(range);
2378                 if (ss != NULL)
2379                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2380                          * at all */
2381                         range->lsr_index = ss->ss_node_id;
2382                 return 0;
2383         }
2384
2385         LASSERT(ss != NULL);
2386         fld_range_set_any(range);
2387         /* OSD will only do local fld lookup */
2388         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2389 }
2390
2391 /*
2392  * Concurrency: no external locking is necessary.
2393  */
2394 static int osd_declare_object_create(const struct lu_env *env,
2395                                      struct dt_object *dt,
2396                                      struct lu_attr *attr,
2397                                      struct dt_allocation_hint *hint,
2398                                      struct dt_object_format *dof,
2399                                      struct thandle *handle)
2400 {
2401         struct osd_thandle      *oh;
2402         int                      rc;
2403         ENTRY;
2404
2405         LASSERT(handle != NULL);
2406
2407         oh = container_of0(handle, struct osd_thandle, ot_super);
2408         LASSERT(oh->ot_handle == NULL);
2409
2410         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2411                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2412         /* Reuse idle OI block may cause additional one OI block
2413          * to be changed. */
2414         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2415                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2416
2417         /* If this is directory, then we expect . and .. to be inserted as
2418          * well. The one directory block always needs to be created for the
2419          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2420          * block), there is no danger of needing a tree for the first block.
2421          */
2422         if (attr && S_ISDIR(attr->la_mode)) {
2423                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2424                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2425                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2426         }
2427
2428         if (!attr)
2429                 RETURN(0);
2430
2431         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2432                                    osd_dt_obj(dt), false, NULL, false);
2433         if (rc != 0)
2434                 RETURN(rc);
2435
2436         RETURN(rc);
2437 }
2438
2439 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2440                              struct lu_attr *attr,
2441                              struct dt_allocation_hint *hint,
2442                              struct dt_object_format *dof, struct thandle *th)
2443 {
2444         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2445         struct osd_object       *obj    = osd_dt_obj(dt);
2446         struct osd_thread_info  *info   = osd_oti_get(env);
2447         int result;
2448         ENTRY;
2449
2450         if (dt_object_exists(dt))
2451                 return -EEXIST;
2452
2453         LINVRNT(osd_invariant(obj));
2454         LASSERT(!dt_object_remote(dt));
2455         LASSERT(osd_write_locked(env, obj));
2456         LASSERT(th != NULL);
2457
2458         if (unlikely(fid_is_acct(fid)))
2459                 /* Quota files can't be created from the kernel any more,
2460                  * 'tune2fs -O quota' will take care of creating them */
2461                 RETURN(-EPERM);
2462
2463         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2464         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2465
2466         result = __osd_object_create(info, obj, attr, hint, dof, th);
2467         if (result == 0)
2468                 result = __osd_oi_insert(env, obj, fid, th);
2469
2470         LASSERT(ergo(result == 0,
2471                      dt_object_exists(dt) && !dt_object_remote(dt)));
2472
2473         LASSERT(osd_invariant(obj));
2474         RETURN(result);
2475 }
2476
2477 /**
2478  * Called to destroy on-disk representation of the object
2479  *
2480  * Concurrency: must be locked
2481  */
2482 static int osd_declare_object_destroy(const struct lu_env *env,
2483                                       struct dt_object *dt,
2484                                       struct thandle *th)
2485 {
2486         struct osd_object  *obj = osd_dt_obj(dt);
2487         struct inode       *inode = obj->oo_inode;
2488         struct osd_thandle *oh;
2489         int                 rc;
2490         ENTRY;
2491
2492         oh = container_of0(th, struct osd_thandle, ot_super);
2493         LASSERT(oh->ot_handle == NULL);
2494         LASSERT(inode);
2495
2496         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2497                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2498         /* Recycle idle OI leaf may cause additional three OI blocks
2499          * to be changed. */
2500         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2501                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2502         /* one less inode */
2503         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2504                                    -1, oh, obj, false, NULL, false);
2505         if (rc)
2506                 RETURN(rc);
2507         /* data to be truncated */
2508         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2509                                    0, oh, obj, true, NULL, false);
2510         RETURN(rc);
2511 }
2512
2513 static int osd_object_destroy(const struct lu_env *env,
2514                               struct dt_object *dt,
2515                               struct thandle *th)
2516 {
2517         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2518         struct osd_object      *obj = osd_dt_obj(dt);
2519         struct inode           *inode = obj->oo_inode;
2520         struct osd_device      *osd = osd_obj2dev(obj);
2521         struct osd_thandle     *oh;
2522         int                     result;
2523         ENTRY;
2524
2525         oh = container_of0(th, struct osd_thandle, ot_super);
2526         LASSERT(oh->ot_handle);
2527         LASSERT(inode);
2528         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2529
2530         if (unlikely(fid_is_acct(fid)))
2531                 RETURN(-EPERM);
2532
2533         if (S_ISDIR(inode->i_mode)) {
2534                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2535                         inode->i_nlink == 2);
2536                 /* it will check/delete the inode from remote parent,
2537                  * how to optimize it? unlink performance impaction XXX */
2538                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2539                 if (result != 0 && result != -ENOENT) {
2540                         CERROR("%s: delete inode "DFID": rc = %d\n",
2541                                osd_name(osd), PFID(fid), result);
2542                 }
2543                 spin_lock(&obj->oo_guard);
2544                 clear_nlink(inode);
2545                 spin_unlock(&obj->oo_guard);
2546                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2547         }
2548
2549         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2550
2551         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2552                                OI_CHECK_FLD);
2553
2554         /* XXX: add to ext3 orphan list */
2555         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2556
2557         /* not needed in the cache anymore */
2558         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2559
2560         RETURN(0);
2561 }
2562
2563 /**
2564  * Put the fid into lustre_mdt_attrs, and then place the structure
2565  * inode's ea. This fid should not be altered during the life time
2566  * of the inode.
2567  *
2568  * \retval +ve, on success
2569  * \retval -ve, on error
2570  *
2571  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2572  */
2573 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2574                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2575 {
2576         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2577         int                      rc;
2578         ENTRY;
2579
2580         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2581                 RETURN(0);
2582
2583         lustre_lma_init(lma, fid, compat, incompat);
2584         lustre_lma_swab(lma);
2585
2586         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2587                              XATTR_CREATE);
2588         /* LMA may already exist, but we need to check that all the
2589          * desired compat/incompat flags have been added. */
2590         if (unlikely(rc == -EEXIST)) {
2591                 if (compat == 0 && incompat == 0)
2592                         RETURN(0);
2593
2594                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2595                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2596                                      LMA_OLD_SIZE);
2597                 if (rc <= 0)
2598                         RETURN(-EINVAL);
2599
2600                 lustre_lma_swab(lma);
2601                 if (!(~lma->lma_compat & compat) &&
2602                     !(~lma->lma_incompat & incompat))
2603                         RETURN(0);
2604
2605                 lma->lma_compat |= compat;
2606                 lma->lma_incompat |= incompat;
2607                 lustre_lma_swab(lma);
2608                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2609                                      sizeof(*lma), XATTR_REPLACE);
2610         }
2611
2612         RETURN(rc);
2613 }
2614
2615 /**
2616  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2617  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2618  * To have compatilibility with 1.8 ldiskfs driver we need to have
2619  * magic number at start of fid data.
2620  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2621  * its inmemory API.
2622  */
2623 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2624                                          const struct lu_fid *fid)
2625 {
2626         if (!fid_is_namespace_visible(fid) ||
2627             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2628                 param->edp_magic = 0;
2629                 return;
2630         }
2631
2632         param->edp_magic = LDISKFS_LUFID_MAGIC;
2633         param->edp_len =  sizeof(struct lu_fid) + 1;
2634         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2635 }
2636
2637 /**
2638  * Try to read the fid from inode ea into dt_rec.
2639  *
2640  * \param fid object fid.
2641  *
2642  * \retval 0 on success
2643  */
2644 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2645                           __u32 ino, struct lu_fid *fid,
2646                           struct osd_inode_id *id)
2647 {
2648         struct osd_thread_info *info  = osd_oti_get(env);
2649         struct inode           *inode;
2650         ENTRY;
2651
2652         osd_id_gen(id, ino, OSD_OII_NOGEN);
2653         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2654         if (IS_ERR(inode))
2655                 RETURN(PTR_ERR(inode));
2656
2657         iput(inode);
2658         RETURN(0);
2659 }
2660
2661 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2662                                         struct inode *dir,
2663                                         struct inode *parent_dir,
2664                                         const struct lu_fid *dot_fid,
2665                                         const struct lu_fid *dot_dot_fid,
2666                                         struct osd_thandle *oth)
2667 {
2668         struct ldiskfs_dentry_param *dot_ldp;
2669         struct ldiskfs_dentry_param *dot_dot_ldp;
2670
2671         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2672         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2673
2674         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2675         dot_ldp->edp_magic = 0;
2676         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2677                                         dir, dot_ldp, dot_dot_ldp);
2678 }
2679
2680 /**
2681  * Create an local agent inode for remote entry
2682  */
2683 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2684                                                   struct osd_device *osd,
2685                                                   struct osd_object *pobj,
2686                                                   const struct lu_fid *fid,
2687                                                   __u32 type,
2688                                                   struct thandle *th)
2689 {
2690         struct osd_thread_info  *info = osd_oti_get(env);
2691         struct inode            *local;
2692         struct osd_thandle      *oh;
2693         int                     rc;
2694         ENTRY;
2695
2696         LASSERT(th);
2697         oh = container_of(th, struct osd_thandle, ot_super);
2698         LASSERT(oh->ot_handle->h_transaction != NULL);
2699
2700         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2701         if (IS_ERR(local)) {
2702                 CERROR("%s: create local error %d\n", osd_name(osd),
2703                        (int)PTR_ERR(local));
2704                 RETURN(local);
2705         }
2706
2707         /* Set special LMA flag for local agent inode */
2708         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2709         if (rc != 0) {
2710                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2711                        osd_name(osd), PFID(fid), rc);
2712                 RETURN(ERR_PTR(rc));
2713         }
2714
2715         if (!S_ISDIR(type))
2716                 RETURN(local);
2717
2718         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2719                                          lu_object_fid(&pobj->oo_dt.do_lu),
2720                                          fid, oh);
2721         if (rc != 0) {
2722                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2723                         osd_name(osd), PFID(fid), rc);
2724                 RETURN(ERR_PTR(rc));
2725         }
2726
2727         RETURN(local);
2728 }
2729
2730 /**
2731  * Delete local agent inode for remote entry
2732  */
2733 static int osd_delete_local_agent_inode(const struct lu_env *env,
2734                                         struct osd_device *osd,
2735                                         const struct lu_fid *fid,
2736                                         __u32 ino, struct osd_thandle *oh)
2737 {
2738         struct osd_thread_info  *oti = osd_oti_get(env);
2739         struct osd_inode_id     *id = &oti->oti_id;
2740         struct inode            *inode;
2741         ENTRY;
2742
2743         id->oii_ino = le32_to_cpu(ino);
2744         id->oii_gen = OSD_OII_NOGEN;
2745         inode = osd_iget(oti, osd, id);
2746         if (IS_ERR(inode)) {
2747                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2748                        PFID(fid), id->oii_ino, id->oii_gen);
2749                 RETURN(PTR_ERR(inode));
2750         }
2751
2752         clear_nlink(inode);
2753         mark_inode_dirty(inode);
2754         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2755                 osd_name(osd), PFID(fid), inode->i_ino);
2756         iput(inode);
2757         RETURN(0);
2758 }
2759
2760 /**
2761  * OSD layer object create function for interoperability mode (b11826).
2762  * This is mostly similar to osd_object_create(). Only difference being, fid is
2763  * inserted into inode ea here.
2764  *
2765  * \retval   0, on success
2766  * \retval -ve, on error
2767  */
2768 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2769                                 struct lu_attr *attr,
2770                                 struct dt_allocation_hint *hint,
2771                                 struct dt_object_format *dof,
2772                                 struct thandle *th)
2773 {
2774         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2775         struct osd_object       *obj    = osd_dt_obj(dt);
2776         struct osd_thread_info  *info   = osd_oti_get(env);
2777         int                      result;
2778
2779         ENTRY;
2780
2781         if (dt_object_exists(dt))
2782                 RETURN(-EEXIST);
2783
2784         LASSERT(osd_invariant(obj));
2785         LASSERT(!dt_object_remote(dt));
2786         LASSERT(osd_write_locked(env, obj));
2787         LASSERT(th != NULL);
2788
2789         if (unlikely(fid_is_acct(fid)))
2790                 /* Quota files can't be created from the kernel any more,
2791                  * 'tune2fs -O quota' will take care of creating them */
2792                 RETURN(-EPERM);
2793
2794         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2795         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2796
2797         result = __osd_object_create(info, obj, attr, hint, dof, th);
2798         if (result == 0) {
2799                 if (fid_is_idif(fid) &&
2800                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2801                         struct lu_fid *tfid = &info->oti_fid;
2802                         struct ost_id *oi   = &info->oti_ostid;
2803
2804                         fid_to_ostid(fid, oi);
2805                         ostid_to_fid(tfid, oi, 0);
2806                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2807                                                 LMAC_FID_ON_OST, 0);
2808                 } else {
2809                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2810                                 fid_is_on_ost(info, osd_obj2dev(obj),
2811                                               fid, OI_CHECK_FLD) ?
2812                                 LMAC_FID_ON_OST : 0, 0);
2813                 }
2814         }
2815
2816         if (result == 0)
2817                 result = __osd_oi_insert(env, obj, fid, th);
2818
2819         LASSERT(ergo(result == 0,
2820                      dt_object_exists(dt) && !dt_object_remote(dt)));
2821         LINVRNT(osd_invariant(obj));
2822         RETURN(result);
2823 }
2824
2825 static int osd_declare_object_ref_add(const struct lu_env *env,
2826                                       struct dt_object *dt,
2827                                       struct thandle *handle)
2828 {
2829         struct osd_thandle       *oh;
2830
2831         /* it's possible that object doesn't exist yet */
2832         LASSERT(handle != NULL);
2833
2834         oh = container_of0(handle, struct osd_thandle, ot_super);
2835         LASSERT(oh->ot_handle == NULL);
2836
2837         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2838                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2839
2840         return 0;
2841 }
2842
2843 /*
2844  * Concurrency: @dt is write locked.
2845  */
2846 static int osd_object_ref_add(const struct lu_env *env,
2847                               struct dt_object *dt, struct thandle *th)
2848 {
2849         struct osd_object  *obj = osd_dt_obj(dt);
2850         struct inode       *inode = obj->oo_inode;
2851         struct osd_thandle *oh;
2852         int                 rc = 0;
2853
2854         if (!dt_object_exists(dt))
2855                 return -ENOENT;
2856
2857         LINVRNT(osd_invariant(obj));
2858         LASSERT(!dt_object_remote(dt));
2859         LASSERT(osd_write_locked(env, obj));
2860         LASSERT(th != NULL);
2861
2862         oh = container_of0(th, struct osd_thandle, ot_super);
2863         LASSERT(oh->ot_handle != NULL);
2864
2865         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2866
2867         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2868                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2869         /*
2870          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2871          * (65000) subdirectories by storing "1" in i_nlink if the link count
2872          * would otherwise overflow. Directory tranversal tools understand
2873          * that (st_nlink == 1) indicates that the filesystem dose not track
2874          * hard links count on the directory, and will not abort subdirectory
2875          * scanning early once (st_nlink - 2) subdirs have been found.
2876          *
2877          * This also has to properly handle the case of inodes with nlink == 0
2878          * in case they are being linked into the PENDING directory
2879          */
2880         spin_lock(&obj->oo_guard);
2881         if (unlikely(inode->i_nlink == 0))
2882                 /* inc_nlink from 0 may cause WARN_ON */
2883                 set_nlink(inode, 1);
2884         else {
2885                 ldiskfs_inc_count(oh->ot_handle, inode);
2886                 if (!S_ISDIR(inode->i_mode))
2887                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2888         }
2889         spin_unlock(&obj->oo_guard);
2890
2891         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2892         LINVRNT(osd_invariant(obj));
2893
2894         return rc;
2895 }
2896
2897 static int osd_declare_object_ref_del(const struct lu_env *env,
2898                                       struct dt_object *dt,
2899                                       struct thandle *handle)
2900 {
2901         struct osd_thandle *oh;
2902
2903         LASSERT(!dt_object_remote(dt));
2904         LASSERT(handle != NULL);
2905
2906         oh = container_of0(handle, struct osd_thandle, ot_super);
2907         LASSERT(oh->ot_handle == NULL);
2908
2909         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2910                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2911
2912         return 0;
2913 }
2914
2915 /*
2916  * Concurrency: @dt is write locked.
2917  */
2918 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2919                               struct thandle *th)
2920 {
2921         struct osd_object       *obj = osd_dt_obj(dt);
2922         struct inode            *inode = obj->oo_inode;
2923         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2924         struct osd_thandle      *oh;
2925
2926         if (!dt_object_exists(dt))
2927                 return -ENOENT;
2928
2929         LINVRNT(osd_invariant(obj));
2930         LASSERT(!dt_object_remote(dt));
2931         LASSERT(osd_write_locked(env, obj));
2932         LASSERT(th != NULL);
2933
2934         oh = container_of0(th, struct osd_thandle, ot_super);
2935         LASSERT(oh->ot_handle != NULL);
2936
2937         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2938
2939         spin_lock(&obj->oo_guard);
2940         /* That can be result of upgrade from old Lustre version and
2941          * applied only to local files.  Just skip this ref_del call.
2942          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2943         if (inode->i_nlink == 0) {
2944                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2945                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2946                              ", maybe an upgraded file? (LU-3915)\n",
2947                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2948                 spin_unlock(&obj->oo_guard);
2949                 return 0;
2950         }
2951
2952         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2953                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2954
2955         ldiskfs_dec_count(oh->ot_handle, inode);
2956         spin_unlock(&obj->oo_guard);
2957
2958         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2959         LINVRNT(osd_invariant(obj));
2960
2961         return 0;
2962 }
2963
2964 /*
2965  * Get the 64-bit version for an inode.
2966  */
2967 static int osd_object_version_get(const struct lu_env *env,
2968                                   struct dt_object *dt, dt_obj_version_t *ver)
2969 {
2970         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2971
2972         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2973                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2974         *ver = LDISKFS_I(inode)->i_fs_version;
2975         return 0;
2976 }
2977
2978 /*
2979  * Concurrency: @dt is read locked.
2980  */
2981 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2982                          struct lu_buf *buf, const char *name,
2983                          struct lustre_capa *capa)
2984 {
2985         struct osd_object      *obj    = osd_dt_obj(dt);
2986         struct inode           *inode  = obj->oo_inode;
2987         struct osd_thread_info *info   = osd_oti_get(env);
2988         struct dentry          *dentry = &info->oti_obj_dentry;
2989
2990         /* version get is not real XATTR but uses xattr API */
2991         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2992                 /* for version we are just using xattr API but change inode
2993                  * field instead */
2994                 if (buf->lb_len == 0)
2995                         return sizeof(dt_obj_version_t);
2996
2997                 if (buf->lb_len < sizeof(dt_obj_version_t))
2998                         return -ERANGE;
2999
3000                 osd_object_version_get(env, dt, buf->lb_buf);
3001
3002                 return sizeof(dt_obj_version_t);
3003         }
3004
3005         if (!dt_object_exists(dt))
3006                 return -ENOENT;
3007
3008         LASSERT(!dt_object_remote(dt));
3009         LASSERT(inode->i_op != NULL);
3010         LASSERT(inode->i_op->getxattr != NULL);
3011
3012         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3013                 return -EACCES;
3014
3015         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3016 }
3017
3018
3019 static int osd_declare_xattr_set(const struct lu_env *env,
3020                                  struct dt_object *dt,
3021                                  const struct lu_buf *buf, const char *name,
3022                                  int fl, struct thandle *handle)
3023 {
3024         struct osd_thandle *oh;
3025         int credits;
3026         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3027
3028         LASSERT(handle != NULL);
3029
3030         oh = container_of0(handle, struct osd_thandle, ot_super);
3031         LASSERT(oh->ot_handle == NULL);
3032
3033         /* optimistic optimization: LMA is set first and usually fit inode */
3034         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3035                 if (dt_object_exists(dt))
3036                         credits = 0;
3037                 else
3038                         credits = 1;
3039         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3040                 credits = 1;
3041         } else {
3042                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3043                 if (buf && buf->lb_len > sb->s_blocksize) {
3044                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
3045                                         sb->s_blocksize_bits;
3046                 }
3047                 /*
3048                  * xattr set may involve inode quota change, reserve credits for
3049                  * dquot_initialize()
3050                  */
3051                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3052         }
3053
3054         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3055
3056         return 0;
3057 }
3058
3059 /*
3060  * Set the 64-bit version for object
3061  */
3062 static void osd_object_version_set(const struct lu_env *env,
3063                                    struct dt_object *dt,
3064                                    dt_obj_version_t *new_version)
3065 {
3066         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3067
3068         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3069                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3070
3071         LDISKFS_I(inode)->i_fs_version = *new_version;
3072         /** Version is set after all inode operations are finished,
3073          *  so we should mark it dirty here */
3074         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3075 }
3076
3077 /*
3078  * Concurrency: @dt is write locked.
3079  */
3080 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3081                          const struct lu_buf *buf, const char *name, int fl,
3082                          struct thandle *handle, struct lustre_capa *capa)
3083 {
3084         struct osd_object      *obj      = osd_dt_obj(dt);
3085         struct inode           *inode    = obj->oo_inode;
3086         struct osd_thread_info *info     = osd_oti_get(env);
3087         int                     fs_flags = 0;
3088         ENTRY;
3089
3090         LASSERT(handle != NULL);
3091
3092         /* version set is not real XATTR */
3093         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3094                 /* for version we are just using xattr API but change inode
3095                  * field instead */
3096                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3097                 osd_object_version_set(env, dt, buf->lb_buf);
3098                 return sizeof(dt_obj_version_t);
3099         }
3100
3101         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3102                 return -EACCES;
3103
3104         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3105                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3106
3107         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3108         if (fl & LU_XATTR_REPLACE)
3109                 fs_flags |= XATTR_REPLACE;
3110
3111         if (fl & LU_XATTR_CREATE)
3112                 fs_flags |= XATTR_CREATE;
3113
3114         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3115                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3116                 int                      rc;
3117
3118                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3119                 if (rc != 0)
3120                         RETURN(rc);
3121
3122                 lma->lma_incompat |= LMAI_STRIPED;
3123                 lustre_lma_swab(lma);
3124                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3125                                      sizeof(*lma), XATTR_REPLACE);
3126                 if (rc != 0)
3127                         RETURN(rc);
3128         }
3129
3130         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3131             strcmp(name, XATTR_NAME_LINK) == 0)
3132                 return -ENOSPC;
3133
3134         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3135                                fs_flags);
3136 }
3137
3138 /*
3139  * Concurrency: @dt is read locked.
3140  */
3141 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3142                           const struct lu_buf *buf, struct lustre_capa *capa)
3143 {
3144         struct osd_object      *obj    = osd_dt_obj(dt);
3145         struct inode           *inode  = obj->oo_inode;
3146         struct osd_thread_info *info   = osd_oti_get(env);
3147         struct dentry          *dentry = &info->oti_obj_dentry;
3148
3149         if (!dt_object_exists(dt))
3150                 return -ENOENT;
3151
3152         LASSERT(!dt_object_remote(dt));
3153         LASSERT(inode->i_op != NULL);
3154         LASSERT(inode->i_op->listxattr != NULL);
3155
3156         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3157                 return -EACCES;
3158
3159         dentry->d_inode = inode;
3160         dentry->d_sb = inode->i_sb;
3161         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3162 }
3163
3164 static int osd_declare_xattr_del(const struct lu_env *env,
3165                                  struct dt_object *dt, const char *name,
3166                                  struct thandle *handle)