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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/module.h>
  43 #include <linux/user_namespace.h>
  44 #ifdef HAVE_UIDGID_HEADER
  45 # include <linux/uidgid.h>
  46 #endif
  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 #include <ldiskfs/ldiskfs.h>
  58 #include <ldiskfs/xattr.h>
  59 #include <ldiskfs/ldiskfs_extents.h>
  60 #undef ENTRY
  61 /*
  62  * struct OBD_{ALLOC,FREE}*()
  63  * OBD_FAIL_CHECK
  64  */
  65 #include <obd_support.h>
  66 /* struct ptlrpc_thread */
  67 #include <lustre_net.h>
  68 #include <lustre_fid.h>
  69 /* process_config */
  70 #include <lustre_param.h>
  71
  72 #include "osd_internal.h"
  73 #include "osd_dynlocks.h"
  74
  75 /* llo_* api support */
  76 #include <md_object.h>
  77 #include <lustre_quota.h>
  78
  79 #include <lustre_linkea.h>
  80
  81 #define PFID_STRIPE_IDX_BITS    16
  82 #define PFID_STRIPE_COUNT_MASK  ((1 << PFID_STRIPE_IDX_BITS) - 1)
  83
  84 int ldiskfs_pdo = 1;
  85 module_param(ldiskfs_pdo, int, 0644);
  86 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  87
  88 int ldiskfs_track_declares_assert;
  89 module_param(ldiskfs_track_declares_assert, int, 0644);
  90 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  91
  92 /* Slab to allocate dynlocks */
  93 struct kmem_cache *dynlock_cachep;
  94
  95 /* Slab to allocate osd_it_ea */
  96 struct kmem_cache *osd_itea_cachep;
  97
  98 static struct lu_kmem_descr ldiskfs_caches[] = {
  99         {
 100                 .ckd_cache = &dynlock_cachep,
 101                 .ckd_name  = "dynlock_cache",
 102                 .ckd_size  = sizeof(struct dynlock_handle)
 103         },
 104         {
 105                 .ckd_cache = &osd_itea_cachep,
 106                 .ckd_name  = "osd_itea_cache",
 107                 .ckd_size  = sizeof(struct osd_it_ea)
 108         },
 109         {
 110                 .ckd_cache = NULL
 111         }
 112 };
 113
 114 static const char dot[] = ".";
 115 static const char dotdot[] = "..";
 116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 117
 118 static const struct lu_object_operations      osd_lu_obj_ops;
 119 static const struct dt_object_operations      osd_obj_ops;
 120 static const struct dt_object_operations      osd_obj_ea_ops;
 121 static const struct dt_object_operations      osd_obj_otable_it_ops;
 122 static const struct dt_index_operations       osd_index_iam_ops;
 123 static const struct dt_index_operations       osd_index_ea_ops;
 124
 125 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 126                           const struct lu_fid *fid);
 127 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 128                                                 struct osd_device *osd);
 129
 130 int osd_trans_declare_op2rb[] = {
 131         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 132         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 133         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 134         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 135         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 136         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 137         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 138         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 139         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 140         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 141         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 142 };
 143
 144 static int osd_has_index(const struct osd_object *obj)
 145 {
 146         return obj->oo_dt.do_index_ops != NULL;
 147 }
 148
 149 static int osd_object_invariant(const struct lu_object *l)
 150 {
 151         return osd_invariant(osd_obj(l));
 152 }
 153
 154 /*
 155  * Concurrency: doesn't matter
 156  */
 157
 158 /*
 159  * Concurrency: doesn't matter
 160  */
 161 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 162 {
 163         struct osd_thread_info *oti = osd_oti_get(env);
 164         return oti->oti_w_locks > 0 && o->oo_owner == env;
 165 }
 166
 167 /*
 168  * Concurrency: doesn't access mutable data
 169  */
 170 static int osd_root_get(const struct lu_env *env,
 171                         struct dt_device *dev, struct lu_fid *f)
 172 {
 173         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 174         return 0;
 175 }
 176
 177 /*
 178  * the following set of functions are used to maintain per-thread
 179  * cache of FID->ino mapping. this mechanism is needed to resolve
 180  * FID to inode at dt_insert() which in turn stores ino in the
 181  * directory entries to keep ldiskfs compatible with ext[34].
 182  * due to locking-originated restrictions we can't lookup ino
 183  * using LU cache (deadlock is possible). lookup using OI is quite
 184  * expensive. so instead we maintain this cache and methods like
 185  * dt_create() fill it. so in the majority of cases dt_insert() is
 186  * able to find needed mapping in lockless manner.
 187  */
 188 static struct osd_idmap_cache *
 189 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 190              const struct lu_fid *fid)
 191 {
 192         struct osd_thread_info  *oti   = osd_oti_get(env);
 193         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
 194         int i;
 195         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 196                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 197                         continue;
 198                 if (idc[i].oic_dev != osd)
 199                         continue;
 200
 201                 return idc + i;
 202         }
 203
 204         return NULL;
 205 }
 206
 207 static struct osd_idmap_cache *
 208 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 209             const struct lu_fid *fid)
 210 {
 211         struct osd_thread_info  *oti   = osd_oti_get(env);
 212         struct osd_idmap_cache  *idc;
 213         int i;
 214
 215         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 216                 i = oti->oti_ins_cache_size * 2;
 217                 if (i == 0)
 218                         i = OSD_INS_CACHE_SIZE;
 219                 OBD_ALLOC(idc, sizeof(*idc) * i);
 220                 if (idc == NULL)
 221                         return ERR_PTR(-ENOMEM);
 222                 if (oti->oti_ins_cache != NULL) {
 223                         memcpy(idc, oti->oti_ins_cache,
 224                                oti->oti_ins_cache_used * sizeof(*idc));
 225                         OBD_FREE(oti->oti_ins_cache,
 226                                  oti->oti_ins_cache_used * sizeof(*idc));
 227                 }
 228                 oti->oti_ins_cache = idc;
 229                 oti->oti_ins_cache_size = i;
 230         }
 231
 232         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 233         idc->oic_fid = *fid;
 234         idc->oic_dev = osd;
 235         idc->oic_lid.oii_ino = 0;
 236         idc->oic_lid.oii_gen = 0;
 237         idc->oic_remote = 0;
 238
 239         return idc;
 240 }
 241
 242 /*
 243  * lookup mapping for the given fid in the cache, initialize a
 244  * new one if not found. the initialization checks whether the
 245  * object is local or remote. for local objects, OI is used to
 246  * learn ino/generation. the function is used when the caller
 247  * has no information about the object, e.g. at dt_insert().
 248  */
 249 static struct osd_idmap_cache *
 250 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 251                      const struct lu_fid *fid)
 252 {
 253         struct osd_idmap_cache *idc;
 254         int rc;
 255
 256         idc = osd_idc_find(env, osd, fid);
 257         LASSERT(!IS_ERR(idc));
 258         if (idc != NULL)
 259                 return idc;
 260
 261         /* new mapping is needed */
 262         idc = osd_idc_add(env, osd, fid);
 263         if (IS_ERR(idc))
 264                 return idc;
 265
 266         /* initialize it */
 267         rc = osd_remote_fid(env, osd, fid);
 268         if (unlikely(rc < 0))
 269                 return ERR_PTR(rc);
 270
 271         if (rc == 0) {
 272                 /* the object is local, lookup in OI */
 273                 /* XXX: probably cheaper to lookup in LU first? */
 274                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 275                                    &idc->oic_lid, 0);
 276                 if (unlikely(rc < 0)) {
 277                         CERROR("can't lookup: rc = %d\n", rc);
 278                         return ERR_PTR(rc);
 279                 }
 280         } else {
 281                 /* the object is remote */
 282                 idc->oic_remote = 1;
 283         }
 284
 285         return idc;
 286 }
 287
 288 /*
 289  * lookup mapping for given FID and fill it from the given object.
 290  * the object is lolcal by definition.
 291  */
 292 static int osd_idc_find_and_init(const struct lu_env *env,
 293                                  struct osd_device *osd,
 294                                  struct osd_object *obj)
 295 {
 296         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 297         struct osd_idmap_cache  *idc;
 298
 299         idc = osd_idc_find(env, osd, fid);
 300         LASSERT(!IS_ERR(idc));
 301         if (idc != NULL) {
 302                 if (obj->oo_inode == NULL)
 303                         return 0;
 304                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 305                         LASSERT(idc->oic_lid.oii_ino == 0);
 306                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 307                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 308                 }
 309                 return 0;
 310         }
 311
 312         /* new mapping is needed */
 313         idc = osd_idc_add(env, osd, fid);
 314         if (IS_ERR(idc))
 315                 return PTR_ERR(idc);
 316
 317         if (obj->oo_inode != NULL) {
 318                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 319                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 320         }
 321         return 0;
 322 }
 323
 324 /*
 325  * OSD object methods.
 326  */
 327
 328 /*
 329  * Concurrency: no concurrent access is possible that early in object
 330  * life-cycle.
 331  */
 332 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 333                                           const struct lu_object_header *hdr,
 334                                           struct lu_device *d)
 335 {
 336         struct osd_object *mo;
 337
 338         OBD_ALLOC_PTR(mo);
 339         if (mo != NULL) {
 340                 struct lu_object *l;
 341
 342                 l = &mo->oo_dt.do_lu;
 343                 dt_object_init(&mo->oo_dt, NULL, d);
 344                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 345                 l->lo_ops = &osd_lu_obj_ops;
 346                 init_rwsem(&mo->oo_sem);
 347                 init_rwsem(&mo->oo_ext_idx_sem);
 348                 spin_lock_init(&mo->oo_guard);
 349                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 350                 return l;
 351         } else {
 352                 return NULL;
 353         }
 354 }
 355
 356 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 357                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 358 {
 359         int rc;
 360
 361         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 362                              (void *)loa, sizeof(*loa));
 363         if (rc > 0) {
 364                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 365
 366                 if (rc < sizeof(*lma))
 367                         return -EINVAL;
 368
 369                 rc = 0;
 370                 lustre_loa_swab(loa, true);
 371                 /* Check LMA compatibility */
 372                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 373                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 374                               "for fid = "DFID", ino = %lu\n",
 375                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 376                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 377                               PFID(&lma->lma_self_fid), inode->i_ino);
 378                         rc = -EOPNOTSUPP;
 379                 }
 380         } else if (rc == 0) {
 381                 rc = -ENODATA;
 382         }
 383
 384         return rc;
 385 }
 386
 387 /*
 388  * retrieve object from backend ext fs.
 389  **/
 390 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 391                        struct osd_inode_id *id)
 392 {
 393         int rc;
 394         struct inode *inode = NULL;
 395
 396         /* if we look for an inode withing a running
 397          * transaction, then we risk to deadlock */
 398         /* osd_dirent_check_repair() breaks this */
 399         /*LASSERT(current->journal_info == NULL);*/
 400
 401         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 402         if (IS_ERR(inode)) {
 403                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 404                        id->oii_ino, PTR_ERR(inode));
 405         } else if (id->oii_gen != OSD_OII_NOGEN &&
 406                    inode->i_generation != id->oii_gen) {
 407                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 408                        "i_generation = %u\n",
 409                        id->oii_ino, id->oii_gen, inode->i_generation);
 410                 iput(inode);
 411                 inode = ERR_PTR(-ESTALE);
 412         } else if (inode->i_nlink == 0) {
 413                 /* due to parallel readdir and unlink,
 414                 * we can have dead inode here. */
 415                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 416                 iput(inode);
 417                 inode = ERR_PTR(-ESTALE);
 418         } else if (is_bad_inode(inode)) {
 419                 CWARN("%.16s: bad inode: ino = %u\n",
 420                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 421                 iput(inode);
 422                 inode = ERR_PTR(-ENOENT);
 423         } else if ((rc = osd_attach_jinode(inode))) {
 424                 iput(inode);
 425                 inode = ERR_PTR(rc);
 426         } else {
 427                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 428                 if (id->oii_gen == OSD_OII_NOGEN)
 429                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 430
 431                 /* Do not update file c/mtime in ldiskfs.
 432                  * NB: we don't have any lock to protect this because we don't
 433                  * have reference on osd_object now, but contention with
 434                  * another lookup + attr_set can't happen in the tiny window
 435                  * between if (...) and set S_NOCMTIME. */
 436                 if (!(inode->i_flags & S_NOCMTIME))
 437                         inode->i_flags |= S_NOCMTIME;
 438         }
 439         return inode;
 440 }
 441
 442 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 443                           handle_t *handle, struct dentry *child,
 444                           struct inode *inode, struct htree_lock *hlock)
 445 {
 446         int rc, rc2;
 447
 448         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 449         if (rc == -ENOBUFS || rc == -ENOSPC) {
 450                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 451                 struct inode *parent = child->d_parent->d_inode;
 452                 struct lu_fid *fid = NULL;
 453
 454                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 455                 if (!rc2) {
 456                         fid = &loa->loa_lma.lma_self_fid;
 457                 } else if (rc2 == -ENODATA) {
 458                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 459                                 fid = &info->oti_fid3;
 460                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 461                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 462                                 fid = &info->oti_fid3;
 463                                 lu_igif_build(fid, parent->i_ino,
 464                                               parent->i_generation);
 465                         }
 466                 }
 467
 468                 if (fid != NULL)
 469                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 470                               "maximum entry limit\n",
 471                               osd_name(osd), parent->i_ino, PFID(fid),
 472                               rc == -ENOSPC ? "has reached" : "is approaching");
 473                 else
 474                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 475                               "maximum entry limit\n",
 476                               osd_name(osd), parent->i_ino,
 477                               rc == -ENOSPC ? "has reached" : "is approaching");
 478
 479                 /* ignore such error now */
 480                 if (rc == -ENOBUFS)
 481                         rc = 0;
 482         }
 483
 484         return rc;
 485 }
 486
 487
 488 static struct inode *
 489 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 490              struct osd_inode_id *id, struct lu_fid *fid)
 491 {
 492         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 493         struct inode *inode;
 494         int rc;
 495
 496         inode = osd_iget(info, dev, id);
 497         if (IS_ERR(inode))
 498                 return inode;
 499
 500         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 501         if (!rc) {
 502                 *fid = loa->loa_lma.lma_self_fid;
 503         } else if (rc == -ENODATA) {
 504                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 505                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 506                 else
 507                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 508         } else {
 509                 iput(inode);
 510                 inode = ERR_PTR(rc);
 511         }
 512         return inode;
 513 }
 514
 515 static struct inode *osd_iget_check(struct osd_thread_info *info,
 516                                     struct osd_device *dev,
 517                                     const struct lu_fid *fid,
 518                                     struct osd_inode_id *id,
 519                                     bool trusted)
 520 {
 521         struct inode *inode;
 522         int rc = 0;
 523         ENTRY;
 524
 525         /* The cached OI mapping is trustable. If we cannot locate the inode
 526          * via the cached OI mapping, then return the failure to the caller
 527          * directly without further OI checking. */
 528
 529 again:
 530         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 531         if (IS_ERR(inode)) {
 532                 rc = PTR_ERR(inode);
 533                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 534                         goto check_oi;
 535
 536                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 537                        PFID(fid), id->oii_ino, rc);
 538                 GOTO(put, rc);
 539         }
 540
 541         if (is_bad_inode(inode)) {
 542                 rc = -ENOENT;
 543                 if (!trusted)
 544                         goto check_oi;
 545
 546                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 547                        PFID(fid), id->oii_ino);
 548                 GOTO(put, rc);
 549         }
 550
 551         if (id->oii_gen != OSD_OII_NOGEN &&
 552             inode->i_generation != id->oii_gen) {
 553                 rc = -ESTALE;
 554                 if (!trusted)
 555                         goto check_oi;
 556
 557                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 558                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 559                        id->oii_ino, id->oii_gen, inode->i_generation);
 560                 GOTO(put, rc);
 561         }
 562
 563         if (inode->i_nlink == 0) {
 564                 rc = -ENOENT;
 565                 if (!trusted)
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 569                        PFID(fid), id->oii_ino);
 570                 GOTO(put, rc);
 571         }
 572
 573         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 574
 575 check_oi:
 576         if (rc != 0) {
 577                 __u32 saved_ino = id->oii_ino;
 578                 __u32 saved_gen = id->oii_gen;
 579
 580                 LASSERT(!trusted);
 581                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 582
 583                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 584                 /* XXX: There are four possible cases:
 585                  *      1. rc = 0.
 586                  *         Backup/restore caused the OI invalid.
 587                  *      2. rc = 0.
 588                  *         Someone unlinked the object but NOT removed
 589                  *         the OI mapping, such as mount target device
 590                  *         as ldiskfs, and modify something directly.
 591                  *      3. rc = -ENOENT.
 592                  *         Someone just removed the object between the
 593                  *         former oi_lookup and the iget. It is normal.
 594                  *      4. Other failure cases.
 595                  *
 596                  *      Generally, when the device is mounted, it will
 597                  *      auto check whether the system is restored from
 598                  *      file-level backup or not. We trust such detect
 599                  *      to distinguish the 1st case from the 2nd case:
 600                  *      if the OI files are consistent but may contain
 601                  *      stale OI mappings because of case 2, if iget()
 602                  *      returns -ENOENT or -ESTALE, then it should be
 603                  *      the case 2. */
 604                 if (rc != 0)
 605                         /* If the OI mapping was in OI file before the
 606                          * osd_iget_check(), but now, it is disappear,
 607                          * then it must be removed by race. That is a
 608                          * normal race case. */
 609                         GOTO(put, rc);
 610
 611                 /* It is the OI scrub updated the OI mapping by race.
 612                  * The new OI mapping must be valid. */
 613                 if (saved_ino != id->oii_ino || saved_gen != id->oii_gen) {
 614                         trusted = true;
 615                         goto again;
 616                 }
 617
 618                 if (IS_ERR(inode)) {
 619                         if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
 620                                 /* It still can be the case 2, but we cannot
 621                                  * distinguish it from the case 1. So return
 622                                  * -EREMCHG to block current operation until
 623                                  *  OI scrub rebuilt the OI mappings. */
 624                                 rc = -EREMCHG;
 625                         else
 626                                 rc = -ENOENT;
 627
 628                         GOTO(put, rc);
 629                 }
 630
 631                 if (inode->i_generation == id->oii_gen)
 632                         rc = -ENOENT;
 633                 else
 634                         rc = -EREMCHG;
 635         } else {
 636                 if (id->oii_gen == OSD_OII_NOGEN)
 637                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 638
 639                 /* Do not update file c/mtime in ldiskfs.
 640                  * NB: we don't have any lock to protect this because we don't
 641                  * have reference on osd_object now, but contention with
 642                  * another lookup + attr_set can't happen in the tiny window
 643                  * between if (...) and set S_NOCMTIME. */
 644                 if (!(inode->i_flags & S_NOCMTIME))
 645                         inode->i_flags |= S_NOCMTIME;
 646         }
 647
 648         GOTO(put, rc);
 649
 650 put:
 651         if (rc != 0) {
 652                 if (!IS_ERR(inode))
 653                         iput(inode);
 654
 655                 inode = ERR_PTR(rc);
 656         }
 657
 658         return inode;
 659 }
 660
 661 /**
 662  * \retval +v: new filter_fid, does not contain self-fid
 663  * \retval 0:  filter_fid_old, contains self-fid
 664  * \retval -v: other failure cases
 665  */
 666 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 667                  struct dentry *dentry, struct lu_fid *fid)
 668 {
 669         struct filter_fid_old   *ff     = &info->oti_ff;
 670         struct ost_id           *ostid  = &info->oti_ostid;
 671         int                      rc;
 672
 673         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 674         if (rc == sizeof(*ff)) {
 675                 rc = 0;
 676                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 677                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 678                 /*
 679                  * XXX: use 0 as the index for compatibility, the caller will
 680                  *      handle index related issues when necessary.
 681                  */
 682                 if (!rc)
 683                         ostid_to_fid(fid, ostid, 0);
 684         } else if (rc == sizeof(struct filter_fid)) {
 685                 rc = 1;
 686         } else if (rc >= 0) {
 687                 rc = -EINVAL;
 688         }
 689
 690         return rc;
 691 }
 692
 693 static int osd_lma_self_repair(struct osd_thread_info *info,
 694                                struct osd_device *osd, struct inode *inode,
 695                                const struct lu_fid *fid, __u32 compat)
 696 {
 697         handle_t *jh;
 698         int       rc;
 699
 700         LASSERT(current->journal_info == NULL);
 701
 702         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 703                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 704         if (IS_ERR(jh)) {
 705                 rc = PTR_ERR(jh);
 706                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 707                       osd_name(osd), rc);
 708                 return rc;
 709         }
 710
 711         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 712         if (rc != 0)
 713                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 714                       osd_name(osd), rc);
 715         ldiskfs_journal_stop(jh);
 716         return rc;
 717 }
 718
 719 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 720 {
 721         struct osd_thread_info  *info   = osd_oti_get(env);
 722         struct osd_device       *osd    = osd_obj2dev(obj);
 723         struct lustre_ost_attrs *loa    = &info->oti_ost_attrs;
 724         struct lustre_mdt_attrs *lma    = &loa->loa_lma;
 725         struct inode            *inode  = obj->oo_inode;
 726         struct dentry           *dentry = &info->oti_obj_dentry;
 727         struct lu_fid           *fid    = NULL;
 728         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 729         int                      rc;
 730         ENTRY;
 731
 732         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 733                              (void *)loa, sizeof(*loa));
 734         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 735                 fid = &lma->lma_self_fid;
 736                 rc = osd_get_idif(info, inode, dentry, fid);
 737                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 738                         /* For the given OST-object, if it has neither LMA nor
 739                          * FID in XATTR_NAME_FID, then the given FID (which is
 740                          * contained in the @obj, from client RPC for locating
 741                          * the OST-object) is trusted. We use it to generate
 742                          * the LMA. */
 743                         osd_lma_self_repair(info, osd, inode, rfid,
 744                                             LMAC_FID_ON_OST);
 745                         RETURN(0);
 746                 }
 747         }
 748
 749         if (rc < 0)
 750                 RETURN(rc);
 751
 752         if (rc > 0) {
 753                 rc = 0;
 754                 lustre_lma_swab(lma);
 755                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 756                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 757                               S_ISREG(inode->i_mode)))) {
 758                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 759                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 760                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 761                               PFID(rfid), inode->i_ino);
 762                         rc = -EOPNOTSUPP;
 763                 } else {
 764                         fid = &lma->lma_self_fid;
 765                 }
 766         }
 767
 768         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 769                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 770                         struct ost_id   *oi   = &info->oti_ostid;
 771                         struct lu_fid   *fid1 = &info->oti_fid3;
 772                         __u32            idx  = fid_idif_ost_idx(rfid);
 773
 774                         /* For old IDIF, the OST index is not part of the IDIF,
 775                          * Means that different OSTs may have the same IDIFs.
 776                          * Under such case, we need to make some compatible
 777                          * check to make sure to trigger OI scrub properly. */
 778                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 779                                 /* Given @rfid is new, LMA is old. */
 780                                 fid_to_ostid(fid, oi);
 781                                 ostid_to_fid(fid1, oi, idx);
 782                                 if (lu_fid_eq(fid1, rfid)) {
 783                                         if (osd->od_index_in_idif)
 784                                                 osd_lma_self_repair(info, osd,
 785                                                         inode, rfid,
 786                                                         LMAC_FID_ON_OST);
 787                                         RETURN(0);
 788                                 }
 789                         }
 790                 }
 791
 792                 rc = -EREMCHG;
 793         }
 794
 795         RETURN(rc);
 796 }
 797
 798 struct osd_check_lmv_buf {
 799 #ifdef HAVE_DIR_CONTEXT
 800         /* please keep it as first member */
 801         struct dir_context       ctx;
 802 #endif
 803         struct osd_thread_info  *oclb_info;
 804         struct osd_device       *oclb_dev;
 805         struct osd_idmap_cache  *oclb_oic;
 806 };
 807
 808 /**
 809  * It is called internally by ->readdir() to filter out the
 810  * local slave object's FID of the striped directory.
 811  *
 812  * \retval      1 found the local slave's FID
 813  * \retval      0 continue to check next item
 814  * \retval      -ve for failure
 815  */
 816 #ifdef HAVE_FILLDIR_USE_CTX
 817 static int osd_stripe_dir_filldir(struct dir_context *buf,
 818 #else
 819 static int osd_stripe_dir_filldir(void *buf,
 820 #endif
 821                                   const char *name, int namelen,
 822                                   loff_t offset, __u64 ino, unsigned d_type)
 823 {
 824         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 825         struct osd_thread_info *oti = oclb->oclb_info;
 826         struct lu_fid *fid = &oti->oti_fid3;
 827         struct osd_inode_id *id = &oti->oti_id3;
 828         struct osd_device *dev = oclb->oclb_dev;
 829         struct osd_idmap_cache *oic = oclb->oclb_oic;
 830         struct inode *inode;
 831         int rc;
 832
 833         if (name[0] == '.')
 834                 return 0;
 835
 836         fid_zero(fid);
 837         sscanf(name + 1, SFID, RFID(fid));
 838         if (!fid_is_sane(fid))
 839                 return 0;
 840
 841         if (osd_remote_fid(oti->oti_env, dev, fid))
 842                 return 0;
 843
 844         osd_id_gen(id, ino, OSD_OII_NOGEN);
 845         inode = osd_iget(oti, dev, id);
 846         if (IS_ERR(inode))
 847                 return PTR_ERR(inode);
 848
 849         iput(inode);
 850         osd_add_oi_cache(oti, dev, id, fid);
 851         oic->oic_fid = *fid;
 852         oic->oic_lid = *id;
 853         oic->oic_dev = dev;
 854         rc = osd_oii_insert(dev, oic, true);
 855
 856         return rc == 0 ? 1 : rc;
 857 }
 858
 859 /* When lookup item under striped directory, we need to locate the master
 860  * MDT-object of the striped directory firstly, then the client will send
 861  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 862  * and the item's name. If the system is restored from MDT file level backup,
 863  * then before the OI scrub completely built the OI files, the OI mappings of
 864  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 865  * not a problem for the master MDT-object. Because when locate the master
 866  * MDT-object, we will do name based lookup (for the striped directory itself)
 867  * firstly, during such process we can setup the correct OI mapping for the
 868  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 869  * the client will not trigger name based lookup on the MDT to locate the slave
 870  * MDT-object before locating item under the striped directory, then when
 871  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 872  * is invalid and does not know what the right OI mapping is, then the MDT has
 873  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 874  * the OI file, related OI mapping is unknown yet, please try again later. And
 875  * then client will re-try the RPC again and again until related OI mapping has
 876  * been updated. That is quite inefficient.
 877  *
 878  * To resolve above trouble, we will handle it as the following two cases:
 879  *
 880  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 881  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 882  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 883  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 884  *    directly. Please check osd_fid_lookup().
 885  *
 886  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 887  *    Under such case, during lookup the master MDT-object, we will lookup the
 888  *    slave MDT-object via readdir against the master MDT-object, because the
 889  *    slave MDT-objects information are stored as sub-directories with the name
 890  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 891  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 892  *    the correct OI mapping for the slave MDT-object. */
 893 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 894                          struct inode *inode, struct osd_idmap_cache *oic)
 895 {
 896         struct lu_buf *buf = &oti->oti_big_buf;
 897         struct dentry *dentry = &oti->oti_obj_dentry;
 898         struct file *filp = &oti->oti_file;
 899         const struct file_operations *fops;
 900         struct lmv_mds_md_v1 *lmv1;
 901         struct osd_check_lmv_buf oclb = {
 902 #ifdef HAVE_DIR_CONTEXT
 903                 .ctx.actor = osd_stripe_dir_filldir,
 904 #endif
 905                 .oclb_info = oti,
 906                 .oclb_dev = dev,
 907                 .oclb_oic = oic
 908         };
 909         int rc = 0;
 910         ENTRY;
 911
 912 again:
 913         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 914                              buf->lb_len);
 915         if (rc == -ERANGE) {
 916                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 917                 if (rc > 0) {
 918                         lu_buf_realloc(buf, rc);
 919                         if (buf->lb_buf == NULL)
 920                                 GOTO(out, rc = -ENOMEM);
 921
 922                         goto again;
 923                 }
 924         }
 925
 926         if (unlikely(rc == 0 || rc == -ENODATA))
 927                 GOTO(out, rc = 0);
 928
 929         if (rc < 0)
 930                 GOTO(out, rc);
 931
 932         if (unlikely(buf->lb_buf == NULL)) {
 933                 lu_buf_realloc(buf, rc);
 934                 if (buf->lb_buf == NULL)
 935                         GOTO(out, rc = -ENOMEM);
 936
 937                 goto again;
 938         }
 939
 940         lmv1 = buf->lb_buf;
 941         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 942                 GOTO(out, rc = 0);
 943
 944         fops = inode->i_fop;
 945         dentry->d_inode = inode;
 946         dentry->d_sb = inode->i_sb;
 947         filp->f_pos = 0;
 948         filp->f_path.dentry = dentry;
 949         filp->f_mode = FMODE_64BITHASH;
 950         filp->f_mapping = inode->i_mapping;
 951         filp->f_op = fops;
 952         filp->private_data = NULL;
 953         set_file_inode(filp, inode);
 954
 955 #ifdef HAVE_DIR_CONTEXT
 956         oclb.ctx.pos = filp->f_pos;
 957         rc = fops->iterate(filp, &oclb.ctx);
 958         filp->f_pos = oclb.ctx.pos;
 959 #else
 960         rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
 961 #endif
 962         fops->release(inode, filp);
 963
 964 out:
 965         if (rc < 0)
 966                 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
 967                        DFID": rc = %d\n",
 968                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 969                        inode->i_ino, inode->i_generation,
 970                        PFID(&oic->oic_fid), rc);
 971         else
 972                 rc = 0;
 973
 974         RETURN(rc);
 975 }
 976
 977 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 978                           const struct lu_fid *fid,
 979                           const struct lu_object_conf *conf)
 980 {
 981         struct osd_thread_info *info;
 982         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
 983         struct osd_device *dev;
 984         struct osd_idmap_cache *oic;
 985         struct osd_inode_id *id;
 986         struct inode *inode = NULL;
 987         struct osd_scrub *scrub;
 988         struct scrub_file *sf;
 989         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
 990         __u32 saved_ino;
 991         __u32 saved_gen;
 992         int result = 0;
 993         int rc1 = 0;
 994         bool remote = false;
 995         bool trusted = true;
 996         bool updated = false;
 997         ENTRY;
 998
 999         LINVRNT(osd_invariant(obj));
1000         LASSERT(obj->oo_inode == NULL);
1001         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1002
1003         dev = osd_dev(ldev);
1004         scrub = &dev->od_scrub;
1005         sf = &scrub->os_file;
1006         info = osd_oti_get(env);
1007         LASSERT(info);
1008         oic = &info->oti_cache;
1009
1010         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1011                 RETURN(-ENOENT);
1012
1013         /* For the object is created as locking anchor, or for the object to
1014          * be created on disk. No need to osd_oi_lookup() at here because FID
1015          * shouldn't never be re-used, if it's really a duplicate FID from
1016          * unexpected reason, we should be able to detect it later by calling
1017          * do_create->osd_oi_insert(). */
1018         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
1019                 GOTO(out, result = 0);
1020
1021         /* Search order: 1. per-thread cache. */
1022         if (lu_fid_eq(fid, &oic->oic_fid) &&
1023             likely(oic->oic_dev == dev)) {
1024                 id = &oic->oic_lid;
1025                 goto iget;
1026         }
1027
1028         id = &info->oti_id;
1029         if (!list_empty(&scrub->os_inconsistent_items)) {
1030                 /* Search order: 2. OI scrub pending list. */
1031                 result = osd_oii_lookup(dev, fid, id);
1032                 if (result == 0)
1033                         goto iget;
1034         }
1035
1036         /* The OI mapping in the OI file can be updated by the OI scrub
1037          * when we locate the inode via FID. So it may be not trustable. */
1038         trusted = false;
1039
1040         /* Search order: 3. OI files. */
1041         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1042         if (result == -ENOENT) {
1043                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1044                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1045                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1046                                       sf->sf_oi_bitmap))
1047                         GOTO(out, result = 0);
1048
1049                 goto trigger;
1050         }
1051
1052         if (result != 0)
1053                 GOTO(out, result);
1054
1055 iget:
1056         inode = osd_iget_check(info, dev, fid, id, trusted);
1057         if (IS_ERR(inode)) {
1058                 result = PTR_ERR(inode);
1059                 if (result == -ENOENT || result == -ESTALE)
1060                         GOTO(out, result = 0);
1061
1062                 if (result == -EREMCHG) {
1063
1064 trigger:
1065                         /* We still have chance to get the valid inode: for the
1066                          * object which is referenced by remote name entry, the
1067                          * object on the local MDT will be linked under the dir
1068                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1069                          *
1070                          * We do not know whether the object for the given FID
1071                          * is referenced by some remote name entry or not, and
1072                          * especially for DNE II, a multiple-linked object may
1073                          * have many name entries reside on many MDTs.
1074                          *
1075                          * To simplify the operation, OSD will not distinguish
1076                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1077                          * only happened for the RPC from other MDT during the
1078                          * OI scrub, or for the client side RPC with FID only,
1079                          * such as FID to path, or from old connected client. */
1080                         if (!remote &&
1081                             !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1082                                 rc1 = osd_lookup_in_remote_parent(info, dev,
1083                                                                   fid, id);
1084                                 if (rc1 == 0) {
1085                                         remote = true;
1086                                         trusted = true;
1087                                         flags |= SS_AUTO_PARTIAL;
1088                                         flags &= ~SS_AUTO_FULL;
1089                                         goto iget;
1090                                 }
1091                         }
1092
1093                         if (thread_is_running(&scrub->os_thread)) {
1094                                 if (scrub->os_partial_scan &&
1095                                     !scrub->os_in_join) {
1096                                         goto join;
1097                                 } else {
1098                                         if (inode != NULL && !IS_ERR(inode)) {
1099                                                 LASSERT(remote);
1100
1101                                                 osd_add_oi_cache(info, dev, id,
1102                                                                  fid);
1103                                                 osd_oii_insert(dev, oic, true);
1104                                         } else {
1105                                                 result = -EINPROGRESS;
1106                                         }
1107                                 }
1108                         } else if (!dev->od_noscrub) {
1109
1110 join:
1111                                 rc1 = osd_scrub_start(dev, flags);
1112                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1113                                               "for the "DFID" with flags 0x%x,"
1114                                               " rc = %d\n", osd_name(dev),
1115                                               PFID(fid), flags, rc1);
1116                                 if (rc1 == 0 || rc1 == -EALREADY) {
1117                                         if (inode != NULL && !IS_ERR(inode)) {
1118                                                 LASSERT(remote);
1119
1120                                                 osd_add_oi_cache(info, dev, id,
1121                                                                  fid);
1122                                                 osd_oii_insert(dev, oic, true);
1123                                         } else {
1124                                                 result = -EINPROGRESS;
1125                                         }
1126                                 } else {
1127                                         result = -EREMCHG;
1128                                 }
1129                         } else {
1130                                 result = -EREMCHG;
1131                         }
1132                 }
1133
1134                 if (inode == NULL || IS_ERR(inode))
1135                         GOTO(out, result);
1136         } else if (remote) {
1137                 goto trigger;
1138         }
1139
1140         obj->oo_inode = inode;
1141         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1142
1143         result = osd_check_lma(env, obj);
1144         if (result == 0)
1145                 goto found;
1146
1147         LASSERTF(id->oii_ino == inode->i_ino &&
1148                  id->oii_gen == inode->i_generation,
1149                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1150                  PFID(fid), id->oii_ino, id->oii_gen,
1151                  inode->i_ino, inode->i_generation);
1152
1153         saved_ino = inode->i_ino;
1154         saved_gen = inode->i_generation;
1155
1156         if (unlikely(result == -ENODATA)) {
1157                 /* If the OI scrub updated the OI mapping by race, it
1158                  * must be valid. Trust the inode that has no LMA EA. */
1159                 if (updated)
1160                         goto found;
1161
1162                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1163                 if (result == 0) {
1164                         /* The OI mapping is still there, the inode is still
1165                          * valid. It is just becaues the inode has no LMA EA. */
1166                         if (saved_ino == id->oii_ino &&
1167                             saved_gen == id->oii_gen)
1168                                 goto found;
1169
1170                         /* It is the OI scrub updated the OI mapping by race.
1171                          * The new OI mapping must be valid. */
1172                         iput(inode);
1173                         inode = NULL;
1174                         obj->oo_inode = NULL;
1175                         trusted = true;
1176                         updated = true;
1177                         goto iget;
1178                 }
1179
1180                 /* "result == -ENOENT" means that the OI mappinghas been
1181                  * removed by race, so the inode belongs to other object.
1182                  *
1183                  * Others error can be returned  directly. */
1184                 if (result == -ENOENT) {
1185                         LASSERT(trusted);
1186
1187                         result = 0;
1188                 }
1189         }
1190
1191         iput(inode);
1192         inode = NULL;
1193         obj->oo_inode = NULL;
1194
1195         if (result != -EREMCHG)
1196                 GOTO(out, result);
1197
1198         LASSERT(!updated);
1199
1200         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1201         /* "result == -ENOENT" means the cached OI mapping has been removed
1202          * from the OI file by race, above inode belongs to other object. */
1203         if (result == -ENOENT) {
1204                 LASSERT(trusted);
1205
1206                 GOTO(out, result = 0);
1207         }
1208
1209         if (result != 0)
1210                 GOTO(out, result);
1211
1212         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen)
1213                 goto trigger;
1214
1215         /* It is the OI scrub updated the OI mapping by race.
1216          * The new OI mapping must be valid. */
1217         trusted = true;
1218         updated = true;
1219         goto iget;
1220
1221 found:
1222         obj->oo_compat_dot_created = 1;
1223         obj->oo_compat_dotdot_created = 1;
1224
1225         if (S_ISDIR(inode->i_mode) &&
1226             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1227                 osd_check_lmv(info, dev, inode, oic);
1228
1229         result = osd_attach_jinode(inode);
1230         if (result) {
1231                 obj->oo_inode = NULL;
1232                 iput(inode);
1233                 GOTO(out, result);
1234         }
1235
1236         if (!ldiskfs_pdo)
1237                 GOTO(out, result = 0);
1238
1239         LASSERT(obj->oo_hl_head == NULL);
1240         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1241         if (obj->oo_hl_head == NULL) {
1242                 obj->oo_inode = NULL;
1243                 iput(inode);
1244                 GOTO(out, result = -ENOMEM);
1245         }
1246         GOTO(out, result = 0);
1247
1248 out:
1249         if (result != 0 && trusted)
1250                 fid_zero(&oic->oic_fid);
1251
1252         LINVRNT(osd_invariant(obj));
1253         return result;
1254 }
1255
1256 /*
1257  * Concurrency: shouldn't matter.
1258  */
1259 static void osd_object_init0(struct osd_object *obj)
1260 {
1261         LASSERT(obj->oo_inode != NULL);
1262         obj->oo_dt.do_body_ops = &osd_body_ops;
1263         obj->oo_dt.do_lu.lo_header->loh_attr |=
1264                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1265 }
1266
1267 /*
1268  * Concurrency: no concurrent access is possible that early in object
1269  * life-cycle.
1270  */
1271 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1272                            const struct lu_object_conf *conf)
1273 {
1274         struct osd_object *obj = osd_obj(l);
1275         int result;
1276
1277         LINVRNT(osd_invariant(obj));
1278
1279         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1280                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1281                 l->lo_header->loh_attr |= LOHA_EXISTS;
1282                 return 0;
1283         }
1284
1285         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1286         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1287         if (result == 0 && obj->oo_inode != NULL) {
1288                 struct osd_thread_info *oti = osd_oti_get(env);
1289                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1290
1291                 osd_object_init0(obj);
1292                 result = osd_get_lma(oti, obj->oo_inode,
1293                                      &oti->oti_obj_dentry, loa);
1294                 if (!result) {
1295                         /* Convert LMAI flags to lustre LMA flags
1296                          * and cache it to oo_lma_flags */
1297                         obj->oo_lma_flags =
1298                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1299                 } else if (result == -ENODATA) {
1300                         result = 0;
1301                 }
1302         }
1303
1304         LINVRNT(osd_invariant(obj));
1305         return result;
1306 }
1307
1308 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1309  * The left part is for value, binary mode. */
1310 struct osd_xattr_entry {
1311         struct list_head        oxe_list;
1312         size_t                  oxe_len;
1313         size_t                  oxe_namelen;
1314         bool                    oxe_exist;
1315         struct rcu_head         oxe_rcu;
1316         char                    oxe_buf[0];
1317 };
1318
1319 static int osd_oxc_get(struct osd_object *obj, const char *name,
1320                        struct lu_buf *buf)
1321 {
1322         struct osd_xattr_entry *tmp;
1323         struct osd_xattr_entry *oxe = NULL;
1324         size_t namelen = strlen(name);
1325         int rc;
1326         ENTRY;
1327
1328         rcu_read_lock();
1329         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1330                 if (namelen == tmp->oxe_namelen &&
1331                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1332                         oxe = tmp;
1333                         break;
1334                 }
1335         }
1336
1337         if (oxe == NULL)
1338                 GOTO(out, rc = -ENOENT);
1339
1340         if (!oxe->oxe_exist)
1341                 GOTO(out, rc = -ENODATA);
1342
1343         /* vallen */
1344         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1345         LASSERT(rc > 0);
1346
1347         if (buf->lb_buf == NULL)
1348                 GOTO(out, rc);
1349
1350         if (buf->lb_len < rc)
1351                 GOTO(out, rc = -ERANGE);
1352
1353         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1354         EXIT;
1355 out:
1356         rcu_read_unlock();
1357
1358         return rc;
1359 }
1360
1361 static void osd_oxc_free(struct rcu_head *head)
1362 {
1363         struct osd_xattr_entry *oxe;
1364
1365         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1366         OBD_FREE(oxe, oxe->oxe_len);
1367 }
1368
1369 static void osd_oxc_add(struct osd_object *obj, const char *name,
1370                         const char *buf, int buflen)
1371 {
1372         struct osd_xattr_entry *oxe;
1373         struct osd_xattr_entry *old = NULL;
1374         struct osd_xattr_entry *tmp;
1375         size_t namelen = strlen(name);
1376         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1377
1378         OBD_ALLOC(oxe, len);
1379         if (oxe == NULL)
1380                 return;
1381
1382         INIT_LIST_HEAD(&oxe->oxe_list);
1383         oxe->oxe_len = len;
1384         oxe->oxe_namelen = namelen;
1385         memcpy(oxe->oxe_buf, name, namelen);
1386         if (buflen > 0) {
1387                 LASSERT(buf != NULL);
1388                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1389                 oxe->oxe_exist = true;
1390         } else {
1391                 oxe->oxe_exist = false;
1392         }
1393
1394         /* this should be rarely called, just remove old and add new */
1395         spin_lock(&obj->oo_guard);
1396         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1397                 if (namelen == tmp->oxe_namelen &&
1398                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1399                         old = tmp;
1400                         break;
1401                 }
1402         }
1403         if (old != NULL) {
1404                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1405                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1406         } else {
1407                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1408         }
1409         spin_unlock(&obj->oo_guard);
1410 }
1411
1412 static void osd_oxc_del(struct osd_object *obj, const char *name)
1413 {
1414         struct osd_xattr_entry *oxe;
1415         size_t namelen = strlen(name);
1416
1417         spin_lock(&obj->oo_guard);
1418         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1419                 if (namelen == oxe->oxe_namelen &&
1420                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1421                         list_del_rcu(&oxe->oxe_list);
1422                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1423                         break;
1424                 }
1425         }
1426         spin_unlock(&obj->oo_guard);
1427 }
1428
1429 static void osd_oxc_fini(struct osd_object *obj)
1430 {
1431         struct osd_xattr_entry *oxe, *next;
1432
1433         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1434                 list_del(&oxe->oxe_list);
1435                 OBD_FREE(oxe, oxe->oxe_len);
1436         }
1437 }
1438
1439 /*
1440  * Concurrency: no concurrent access is possible that late in object
1441  * life-cycle.
1442  */
1443 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1444 {
1445         struct osd_object *obj = osd_obj(l);
1446
1447         LINVRNT(osd_invariant(obj));
1448
1449         osd_oxc_fini(obj);
1450         dt_object_fini(&obj->oo_dt);
1451         if (obj->oo_hl_head != NULL)
1452                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1453         OBD_FREE_PTR(obj);
1454 }
1455
1456 /*
1457  * Concurrency: no concurrent access is possible that late in object
1458  * life-cycle.
1459  */
1460 static void osd_index_fini(struct osd_object *o)
1461 {
1462         struct iam_container *bag;
1463
1464         if (o->oo_dir != NULL) {
1465                 bag = &o->oo_dir->od_container;
1466                 if (o->oo_inode != NULL) {
1467                         if (bag->ic_object == o->oo_inode)
1468                                 iam_container_fini(bag);
1469                 }
1470                 OBD_FREE_PTR(o->oo_dir);
1471                 o->oo_dir = NULL;
1472         }
1473 }
1474
1475 /*
1476  * Concurrency: no concurrent access is possible that late in object
1477  * life-cycle (for all existing callers, that is. New callers have to provide
1478  * their own locking.)
1479  */
1480 static int osd_inode_unlinked(const struct inode *inode)
1481 {
1482         return inode->i_nlink == 0;
1483 }
1484
1485 enum {
1486         OSD_TXN_OI_DELETE_CREDITS    = 20,
1487         OSD_TXN_INODE_DELETE_CREDITS = 20
1488 };
1489
1490 /*
1491  * Journal
1492  */
1493
1494 #if OSD_THANDLE_STATS
1495 /**
1496  * Set time when the handle is allocated
1497  */
1498 static void osd_th_alloced(struct osd_thandle *oth)
1499 {
1500         oth->oth_alloced = ktime_get();
1501 }
1502
1503 /**
1504  * Set time when the handle started
1505  */
1506 static void osd_th_started(struct osd_thandle *oth)
1507 {
1508         oth->oth_started = ktime_get();
1509 }
1510
1511 /**
1512  * Check whether the we deal with this handle for too long.
1513  */
1514 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1515                                 ktime_t alloced, ktime_t started,
1516                                 ktime_t closed)
1517 {
1518         ktime_t now = ktime_get();
1519
1520         LASSERT(dev != NULL);
1521
1522         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1523                             ktime_us_delta(started, alloced));
1524         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1525                             ktime_us_delta(closed, started));
1526         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1527                             ktime_us_delta(now, closed));
1528
1529         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1530                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1531                       oth, now, alloced, started, closed);
1532                 libcfs_debug_dumpstack(NULL);
1533         }
1534 }
1535
1536 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1537 {                                                                       \
1538         ktime_t __closed = ktime_get();                                 \
1539         ktime_t __alloced = oth->oth_alloced;                           \
1540         ktime_t __started = oth->oth_started;                           \
1541                                                                         \
1542         expr;                                                           \
1543         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1544 }
1545
1546 #else /* OSD_THANDLE_STATS */
1547
1548 #define osd_th_alloced(h)                  do {} while(0)
1549 #define osd_th_started(h)                  do {} while(0)
1550 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1551
1552 #endif /* OSD_THANDLE_STATS */
1553
1554 /*
1555  * Concurrency: doesn't access mutable data.
1556  */
1557 static int osd_param_is_not_sane(const struct osd_device *dev,
1558                                  const struct thandle *th)
1559 {
1560         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1561
1562         return oh->ot_credits > osd_transaction_size(dev);
1563 }
1564
1565 /*
1566  * Concurrency: shouldn't matter.
1567  */
1568 static void osd_trans_commit_cb(struct super_block *sb,
1569                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1570 {
1571         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1572         struct thandle     *th  = &oh->ot_super;
1573         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1574         struct dt_txn_commit_cb *dcb, *tmp;
1575
1576         LASSERT(oh->ot_handle == NULL);
1577
1578         if (error)
1579                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1580
1581         dt_txn_hook_commit(th);
1582
1583         /* call per-transaction callbacks if any */
1584         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1585                                  dcb_linkage) {
1586                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1587                          "commit callback entry: magic=%x name='%s'\n",
1588                          dcb->dcb_magic, dcb->dcb_name);
1589                 list_del_init(&dcb->dcb_linkage);
1590                 dcb->dcb_func(NULL, th, dcb, error);
1591         }
1592
1593         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1594         lu_device_put(lud);
1595         th->th_dev = NULL;
1596
1597         lu_context_exit(&th->th_ctx);
1598         lu_context_fini(&th->th_ctx);
1599         OBD_FREE_PTR(oh);
1600 }
1601
1602 #ifndef HAVE_SB_START_WRITE
1603 # define sb_start_write(sb) do {} while (0)
1604 # define sb_end_write(sb) do {} while (0)
1605 #endif
1606
1607 static struct thandle *osd_trans_create(const struct lu_env *env,
1608                                         struct dt_device *d)
1609 {
1610         struct osd_thread_info  *oti = osd_oti_get(env);
1611         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1612         struct osd_thandle      *oh;
1613         struct thandle          *th;
1614         ENTRY;
1615
1616         if (d->dd_rdonly) {
1617                 CERROR("%s: someone try to start transaction under "
1618                        "readonly mode, should be disabled.\n",
1619                        osd_name(osd_dt_dev(d)));
1620                 dump_stack();
1621                 RETURN(ERR_PTR(-EROFS));
1622         }
1623
1624         /* on pending IO in this thread should left from prev. request */
1625         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1626
1627         sb_start_write(osd_sb(osd_dt_dev(d)));
1628
1629         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1630         if (oh != NULL) {
1631                 oh->ot_quota_trans = &oti->oti_quota_trans;
1632                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1633                 th = &oh->ot_super;
1634                 th->th_dev = d;
1635                 th->th_result = 0;
1636                 th->th_tags = LCT_TX_HANDLE;
1637                 oh->ot_credits = 0;
1638                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1639                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1640                 osd_th_alloced(oh);
1641
1642                 memset(oti->oti_declare_ops, 0,
1643                        sizeof(oti->oti_declare_ops));
1644                 memset(oti->oti_declare_ops_cred, 0,
1645                        sizeof(oti->oti_declare_ops_cred));
1646                 memset(oti->oti_declare_ops_used, 0,
1647                        sizeof(oti->oti_declare_ops_used));
1648         } else {
1649                 sb_end_write(osd_sb(osd_dt_dev(d)));
1650                 th = ERR_PTR(-ENOMEM);
1651         }
1652         RETURN(th);
1653 }
1654
1655 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1656 {
1657         struct osd_thread_info  *oti = osd_oti_get(env);
1658         struct osd_thandle      *oh;
1659
1660         oh = container_of0(th, struct osd_thandle, ot_super);
1661         LASSERT(oh != NULL);
1662
1663         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1664               oti->oti_declare_ops[OSD_OT_CREATE],
1665               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1666               oti->oti_declare_ops_used[OSD_OT_CREATE],
1667               oti->oti_declare_ops[OSD_OT_DESTROY],
1668               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1669               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1670         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1671               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1672               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1673               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1674               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1675               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1676               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1677         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1678               oti->oti_declare_ops[OSD_OT_WRITE],
1679               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1680               oti->oti_declare_ops_used[OSD_OT_WRITE],
1681               oti->oti_declare_ops[OSD_OT_PUNCH],
1682               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1683               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1684               oti->oti_declare_ops[OSD_OT_QUOTA],
1685               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1686               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1687         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1688               oti->oti_declare_ops[OSD_OT_INSERT],
1689               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1690               oti->oti_declare_ops_used[OSD_OT_INSERT],
1691               oti->oti_declare_ops[OSD_OT_DELETE],
1692               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1693               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1694         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1695               oti->oti_declare_ops[OSD_OT_REF_ADD],
1696               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1697               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1698               oti->oti_declare_ops[OSD_OT_REF_DEL],
1699               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1700               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1701 }
1702
1703 /*
1704  * Concurrency: shouldn't matter.
1705  */
1706 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1707                            struct thandle *th)
1708 {
1709         struct osd_thread_info *oti = osd_oti_get(env);
1710         struct osd_device  *dev = osd_dt_dev(d);
1711         handle_t           *jh;
1712         struct osd_thandle *oh;
1713         int rc;
1714
1715         ENTRY;
1716
1717         LASSERT(current->journal_info == NULL);
1718
1719         oh = container_of0(th, struct osd_thandle, ot_super);
1720         LASSERT(oh != NULL);
1721         LASSERT(oh->ot_handle == NULL);
1722
1723         rc = dt_txn_hook_start(env, d, th);
1724         if (rc != 0)
1725                 GOTO(out, rc);
1726
1727         if (unlikely(osd_param_is_not_sane(dev, th))) {
1728                 static unsigned long last_printed;
1729                 static int last_credits;
1730
1731                 /* don't make noise on a tiny testing systems
1732                  * actual credits misuse will be caught anyway */
1733                 if (last_credits != oh->ot_credits &&
1734                     time_after(jiffies, last_printed +
1735                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1736                     osd_transaction_size(dev) > 512) {
1737                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1738                               oh->ot_credits, osd_transaction_size(dev));
1739                         osd_trans_dump_creds(env, th);
1740                         libcfs_debug_dumpstack(NULL);
1741                         last_credits = oh->ot_credits;
1742                         last_printed = jiffies;
1743                 }
1744                 /* XXX Limit the credits to 'max_transaction_buffers', and
1745                  *     let the underlying filesystem to catch the error if
1746                  *     we really need so many credits.
1747                  *
1748                  *     This should be removed when we can calculate the
1749                  *     credits precisely. */
1750                 oh->ot_credits = osd_transaction_size(dev);
1751         } else if (ldiskfs_track_declares_assert != 0) {
1752                 /* reserve few credits to prevent an assertion in JBD
1753                  * our debugging mechanism will be able to detected
1754                  * overuse. this can help to debug single-update
1755                  * transactions */
1756                 oh->ot_credits += 10;
1757                 if (unlikely(osd_param_is_not_sane(dev, th)))
1758                         oh->ot_credits = osd_transaction_size(dev);
1759         }
1760
1761         /*
1762          * XXX temporary stuff. Some abstraction layer should
1763          * be used.
1764          */
1765         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1766         osd_th_started(oh);
1767         if (!IS_ERR(jh)) {
1768                 oh->ot_handle = jh;
1769                 LASSERT(oti->oti_txns == 0);
1770                 lu_context_init(&th->th_ctx, th->th_tags);
1771                 lu_context_enter(&th->th_ctx);
1772
1773                 lu_device_get(&d->dd_lu_dev);
1774                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1775                               "osd-tx", th);
1776                 oti->oti_txns++;
1777                 rc = 0;
1778         } else {
1779                 rc = PTR_ERR(jh);
1780         }
1781 out:
1782         RETURN(rc);
1783 }
1784
1785 static int osd_seq_exists(const struct lu_env *env,
1786                           struct osd_device *osd, u64 seq)
1787 {
1788         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1789         struct seq_server_site  *ss = osd_seq_site(osd);
1790         int                     rc;
1791         ENTRY;
1792
1793         LASSERT(ss != NULL);
1794         LASSERT(ss->ss_server_fld != NULL);
1795
1796         rc = osd_fld_lookup(env, osd, seq, range);
1797         if (rc != 0) {
1798                 if (rc != -ENOENT)
1799                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1800                                osd_name(osd), seq, rc);
1801                 RETURN(0);
1802         }
1803
1804         RETURN(ss->ss_node_id == range->lsr_index);
1805 }
1806
1807 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1808 {
1809         struct dt_txn_commit_cb *dcb;
1810         struct dt_txn_commit_cb *tmp;
1811
1812         /* call per-transaction stop callbacks if any */
1813         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1814                                  dcb_linkage) {
1815                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1816                          "commit callback entry: magic=%x name='%s'\n",
1817                          dcb->dcb_magic, dcb->dcb_name);
1818                 list_del_init(&dcb->dcb_linkage);
1819                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1820         }
1821 }
1822
1823 /*
1824  * Concurrency: shouldn't matter.
1825  */
1826 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1827                           struct thandle *th)
1828 {
1829         int                     rc = 0, remove_agents = 0;
1830         struct osd_thandle     *oh;
1831         struct osd_thread_info *oti = osd_oti_get(env);
1832         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1833         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1834         struct qsd_instance    *qsd = osd->od_quota_slave;
1835         struct lquota_trans    *qtrans;
1836         ENTRY;
1837
1838         oh = container_of0(th, struct osd_thandle, ot_super);
1839
1840         /* reset OI cache for safety */
1841         oti->oti_ins_cache_used = 0;
1842
1843         remove_agents = oh->ot_remove_agents;
1844
1845         qtrans = oh->ot_quota_trans;
1846         oh->ot_quota_trans = NULL;
1847
1848         if (oh->ot_handle != NULL) {
1849                 int rc2;
1850                 handle_t *hdl = oh->ot_handle;
1851
1852                 /*
1853                  * add commit callback
1854                  * notice we don't do this in osd_trans_start()
1855                  * as underlying transaction can change during truncate
1856                  */
1857                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1858                                          &oh->ot_jcb);
1859
1860                 LASSERT(oti->oti_txns == 1);
1861                 oti->oti_txns--;
1862
1863                 rc = dt_txn_hook_stop(env, th);
1864                 if (rc != 0)
1865                         CERROR("%s: failed in transaction hook: rc = %d\n",
1866                                osd_name(osd), rc);
1867
1868                 osd_trans_stop_cb(oh, rc);
1869                 /* hook functions might modify th_sync */
1870                 hdl->h_sync = th->th_sync;
1871
1872                 oh->ot_handle = NULL;
1873                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1874                 if (rc2 != 0)
1875                         CERROR("%s: failed to stop transaction: rc = %d\n",
1876                                osd_name(osd), rc2);
1877                 if (!rc)
1878                         rc = rc2;
1879         } else {
1880                 osd_trans_stop_cb(oh, th->th_result);
1881                 OBD_FREE_PTR(oh);
1882         }
1883
1884         /* inform the quota slave device that the transaction is stopping */
1885         qsd_op_end(env, qsd, qtrans);
1886
1887         /* as we want IO to journal and data IO be concurrent, we don't block
1888          * awaiting data IO completion in osd_do_bio(), instead we wait here
1889          * once transaction is submitted to the journal. all reqular requests
1890          * don't do direct IO (except read/write), thus this wait_event becomes
1891          * no-op for them.
1892          *
1893          * IMPORTANT: we have to wait till any IO submited by the thread is
1894          * completed otherwise iobuf may be corrupted by different request
1895          */
1896         wait_event(iobuf->dr_wait,
1897                        atomic_read(&iobuf->dr_numreqs) == 0);
1898         osd_fini_iobuf(osd, iobuf);
1899         if (!rc)
1900                 rc = iobuf->dr_error;
1901
1902         if (unlikely(remove_agents != 0))
1903                 osd_process_scheduled_agent_removals(env, osd);
1904
1905         sb_end_write(osd_sb(osd));
1906
1907         RETURN(rc);
1908 }
1909
1910 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1911 {
1912         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1913                                                ot_super);
1914
1915         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1916         LASSERT(&dcb->dcb_func != NULL);
1917         if (dcb->dcb_flags & DCB_TRANS_STOP)
1918                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1919         else
1920                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1921
1922         return 0;
1923 }
1924
1925 /*
1926  * Called just before object is freed. Releases all resources except for
1927  * object itself (that is released by osd_object_free()).
1928  *
1929  * Concurrency: no concurrent access is possible that late in object
1930  * life-cycle.
1931  */
1932 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1933 {
1934         struct osd_object *obj   = osd_obj(l);
1935         struct inode      *inode = obj->oo_inode;
1936
1937         LINVRNT(osd_invariant(obj));
1938
1939         /*
1940          * If object is unlinked remove fid->ino mapping from object index.
1941          */
1942
1943         osd_index_fini(obj);
1944         if (inode != NULL) {
1945                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1946                 qid_t                    uid = i_uid_read(inode);
1947                 qid_t                    gid = i_gid_read(inode);
1948
1949                 iput(inode);
1950                 obj->oo_inode = NULL;
1951
1952                 if (qsd != NULL) {
1953                         struct osd_thread_info  *info = osd_oti_get(env);
1954                         struct lquota_id_info   *qi = &info->oti_qi;
1955
1956                         /* Release granted quota to master if necessary */
1957                         qi->lqi_id.qid_uid = uid;
1958                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1959
1960                         qi->lqi_id.qid_uid = gid;
1961                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1962
1963                         qi->lqi_id.qid_uid = i_projid_read(inode);
1964                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1965                 }
1966         }
1967 }
1968
1969 /*
1970  * Concurrency: ->loo_object_release() is called under site spin-lock.
1971  */
1972 static void osd_object_release(const struct lu_env *env,
1973                                struct lu_object *l)
1974 {
1975         struct osd_object *o = osd_obj(l);
1976         /* nobody should be releasing a non-destroyed object with nlink=0
1977          * the API allows this, but ldiskfs doesn't like and then report
1978          * this inode as deleted */
1979         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1980                 LBUG();
1981 }
1982
1983 /*
1984  * Concurrency: shouldn't matter.
1985  */
1986 static int osd_object_print(const struct lu_env *env, void *cookie,
1987                             lu_printer_t p, const struct lu_object *l)
1988 {
1989         struct osd_object *o = osd_obj(l);
1990         struct iam_descr  *d;
1991
1992         if (o->oo_dir != NULL)
1993                 d = o->oo_dir->od_container.ic_descr;
1994         else
1995                 d = NULL;
1996         return (*p)(env, cookie,
1997                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1998                     o, o->oo_inode,
1999                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2000                     o->oo_inode ? o->oo_inode->i_generation : 0,
2001                     d ? d->id_ops->id_name : "plain");
2002 }
2003
2004 /*
2005  * Concurrency: shouldn't matter.
2006  */
2007 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2008                struct obd_statfs *sfs)
2009 {
2010         struct osd_device       *osd = osd_dt_dev(d);
2011         struct super_block      *sb = osd_sb(osd);
2012         struct kstatfs          *ksfs;
2013         __u64                    reserved;
2014         int                      result = 0;
2015
2016         if (unlikely(osd->od_mnt == NULL))
2017                 return -EINPROGRESS;
2018
2019         /* osd_lproc.c call this without env, allocate ksfs for that case */
2020         if (unlikely(env == NULL)) {
2021                 OBD_ALLOC_PTR(ksfs);
2022                 if (ksfs == NULL)
2023                         return -ENOMEM;
2024         } else {
2025                 ksfs = &osd_oti_get(env)->oti_ksfs;
2026         }
2027
2028         result = sb->s_op->statfs(sb->s_root, ksfs);
2029         if (result)
2030                 goto out;
2031
2032         statfs_pack(sfs, ksfs);
2033         if (unlikely(sb->s_flags & MS_RDONLY))
2034                 sfs->os_state |= OS_STATE_READONLY;
2035         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2036                                          LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2037                 sfs->os_maxbytes = sb->s_maxbytes;
2038         else
2039                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2040
2041         /*
2042          * Reserve some space so to avoid fragmenting the filesystem too much.
2043          * Fragmentation not only impacts performance, but can also increase
2044          * metadata overhead significantly, causing grant calculation to be
2045          * wrong.
2046          *
2047          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2048          */
2049         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2050         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2051         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2052                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2053
2054         sfs->os_blocks -= reserved;
2055         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2056         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2057
2058 out:
2059         if (unlikely(env == NULL))
2060                 OBD_FREE_PTR(ksfs);
2061         return result;
2062 }
2063
2064 /**
2065  * Estimate space needed for file creations. We assume the largest filename
2066  * which is 2^64 - 1, hence a filename of 20 chars.
2067  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2068  */
2069 #ifdef __LDISKFS_DIR_REC_LEN
2070 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2071 #else
2072 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2073 #endif
2074
2075 /*
2076  * Concurrency: doesn't access mutable data.
2077  */
2078 static void osd_conf_get(const struct lu_env *env,
2079                          const struct dt_device *dev,
2080                          struct dt_device_param *param)
2081 {
2082         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2083         int                ea_overhead;
2084
2085         /*
2086          * XXX should be taken from not-yet-existing fs abstraction layer.
2087          */
2088         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2089         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2090         param->ddp_symlink_max  = sb->s_blocksize;
2091         param->ddp_mount_type     = LDD_MT_LDISKFS;
2092         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2093                 param->ddp_maxbytes = sb->s_maxbytes;
2094         else
2095                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2096         /* inode are statically allocated, so per-inode space consumption
2097          * is the space consumed by the directory entry */
2098         param->ddp_inodespace     = PER_OBJ_USAGE;
2099         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2100          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2101          * maximum length to not under count the actual number of extents
2102          * needed for writing a file. */
2103         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2104         /* worst-case extent insertion metadata overhead */
2105         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2106         param->ddp_mntopts      = 0;
2107         if (test_opt(sb, XATTR_USER))
2108                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2109         if (test_opt(sb, POSIX_ACL))
2110                 param->ddp_mntopts |= MNTOPT_ACL;
2111
2112         /* LOD might calculate the max stripe count based on max_ea_size,
2113          * so we need take account in the overhead as well,
2114          * xattr_header + magic + xattr_entry_head */
2115         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2116                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2117
2118 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2119         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2120                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2121                                                                 ea_overhead;
2122         else
2123 #endif
2124                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2125 }
2126
2127 /*
2128  * Concurrency: shouldn't matter.
2129  */
2130 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2131 {
2132         int rc;
2133
2134         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2135
2136         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2137
2138         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2139                LUSTRE_OSD_LDISKFS_NAME, rc);
2140
2141         return rc;
2142 }
2143
2144 /**
2145  * Start commit for OSD device.
2146  *
2147  * An implementation of dt_commit_async method for OSD device.
2148  * Asychronously starts underlayng fs sync and thereby a transaction
2149  * commit.
2150  *
2151  * \param env environment
2152  * \param d dt device
2153  *
2154  * \see dt_device_operations
2155  */
2156 static int osd_commit_async(const struct lu_env *env,
2157                             struct dt_device *d)
2158 {
2159         struct super_block *s = osd_sb(osd_dt_dev(d));
2160         ENTRY;
2161
2162         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2163         RETURN(s->s_op->sync_fs(s, 0));
2164 }
2165
2166 /*
2167  * Concurrency: shouldn't matter.
2168  */
2169
2170 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2171 {
2172         struct super_block *sb = osd_sb(osd_dt_dev(d));
2173         struct block_device *dev = sb->s_bdev;
2174 #ifdef HAVE_DEV_SET_RDONLY
2175         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2176         int rc = 0;
2177 #else
2178         int rc = -EOPNOTSUPP;
2179 #endif
2180         ENTRY;
2181
2182 #ifdef HAVE_DEV_SET_RDONLY
2183         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2184
2185         if (sb->s_op->freeze_fs) {
2186                 rc = sb->s_op->freeze_fs(sb);
2187                 if (rc)
2188                         goto out;
2189         }
2190
2191         if (jdev && (jdev != dev)) {
2192                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2193                        (long)jdev);
2194                 dev_set_rdonly(jdev);
2195         }
2196         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2197         dev_set_rdonly(dev);
2198
2199         if (sb->s_op->unfreeze_fs)
2200                 sb->s_op->unfreeze_fs(sb);
2201
2202 out:
2203 #endif
2204         if (rc)
2205                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2206                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2207
2208         RETURN(rc);
2209 }
2210
2211 /**
2212  * Note: we do not count into QUOTA here.
2213  * If we mount with --data_journal we may need more.
2214  */
2215 const int osd_dto_credits_noquota[DTO_NR] = {
2216         /**
2217          * Insert.
2218          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2219          * SINGLEDATA_TRANS_BLOCKS(8)
2220          * XXX Note: maybe iam need more, since iam have more level than
2221          *           EXT3 htree.
2222          */
2223         [DTO_INDEX_INSERT]  = 16,
2224         /**
2225          * Delete
2226          * just modify a single entry, probably merge few within a block
2227          */
2228         [DTO_INDEX_DELETE]  = 1,
2229         /**
2230          * Used for OI scrub
2231          */
2232         [DTO_INDEX_UPDATE]  = 16,
2233         /**
2234          * 4(inode, inode bits, groups, GDT)
2235          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2236          */
2237         [DTO_OBJECT_CREATE] = 4,
2238         /**
2239          * 4(inode, inode bits, groups, GDT)
2240          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2241          */
2242         [DTO_OBJECT_DELETE] = 4,
2243         /**
2244          * Attr set credits (inode)
2245          */
2246         [DTO_ATTR_SET_BASE] = 1,
2247         /**
2248          * Xattr set. The same as xattr of EXT3.
2249          * DATA_TRANS_BLOCKS(14)
2250          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2251          * are also counted in. Do not know why?
2252          */
2253         [DTO_XATTR_SET]     = 14,
2254         /**
2255          * credits for inode change during write.
2256          */
2257         [DTO_WRITE_BASE]    = 3,
2258         /**
2259          * credits for single block write.
2260          */
2261         [DTO_WRITE_BLOCK]   = 14,
2262         /**
2263          * Attr set credits for chown.
2264          * This is extra credits for setattr, and it is null without quota
2265          */
2266         [DTO_ATTR_SET_CHOWN] = 0
2267 };
2268
2269 static const struct dt_device_operations osd_dt_ops = {
2270         .dt_root_get       = osd_root_get,
2271         .dt_statfs         = osd_statfs,
2272         .dt_trans_create   = osd_trans_create,
2273         .dt_trans_start    = osd_trans_start,
2274         .dt_trans_stop     = osd_trans_stop,
2275         .dt_trans_cb_add   = osd_trans_cb_add,
2276         .dt_conf_get       = osd_conf_get,
2277         .dt_sync           = osd_sync,
2278         .dt_ro             = osd_ro,
2279         .dt_commit_async   = osd_commit_async,
2280 };
2281
2282 static void osd_object_read_lock(const struct lu_env *env,
2283                                  struct dt_object *dt, unsigned role)
2284 {
2285         struct osd_object *obj = osd_dt_obj(dt);
2286         struct osd_thread_info *oti = osd_oti_get(env);
2287
2288         LINVRNT(osd_invariant(obj));
2289
2290         LASSERT(obj->oo_owner != env);
2291         down_read_nested(&obj->oo_sem, role);
2292
2293         LASSERT(obj->oo_owner == NULL);
2294         oti->oti_r_locks++;
2295 }
2296
2297 static void osd_object_write_lock(const struct lu_env *env,
2298                                   struct dt_object *dt, unsigned role)
2299 {
2300         struct osd_object *obj = osd_dt_obj(dt);
2301         struct osd_thread_info *oti = osd_oti_get(env);
2302
2303         LINVRNT(osd_invariant(obj));
2304
2305         LASSERT(obj->oo_owner != env);
2306         down_write_nested(&obj->oo_sem, role);
2307
2308         LASSERT(obj->oo_owner == NULL);
2309         obj->oo_owner = env;
2310         oti->oti_w_locks++;
2311 }
2312
2313 static void osd_object_read_unlock(const struct lu_env *env,
2314                                    struct dt_object *dt)
2315 {
2316         struct osd_object *obj = osd_dt_obj(dt);
2317         struct osd_thread_info *oti = osd_oti_get(env);
2318
2319         LINVRNT(osd_invariant(obj));
2320
2321         LASSERT(oti->oti_r_locks > 0);
2322         oti->oti_r_locks--;
2323         up_read(&obj->oo_sem);
2324 }
2325
2326 static void osd_object_write_unlock(const struct lu_env *env,
2327                                     struct dt_object *dt)
2328 {
2329         struct osd_object *obj = osd_dt_obj(dt);
2330         struct osd_thread_info *oti = osd_oti_get(env);
2331
2332         LINVRNT(osd_invariant(obj));
2333
2334         LASSERT(obj->oo_owner == env);
2335         LASSERT(oti->oti_w_locks > 0);
2336         oti->oti_w_locks--;
2337         obj->oo_owner = NULL;
2338         up_write(&obj->oo_sem);
2339 }
2340
2341 static int osd_object_write_locked(const struct lu_env *env,
2342                                    struct dt_object *dt)
2343 {
2344         struct osd_object *obj = osd_dt_obj(dt);
2345
2346         LINVRNT(osd_invariant(obj));
2347
2348         return obj->oo_owner == env;
2349 }
2350
2351 static struct timespec *osd_inode_time(const struct lu_env *env,
2352                                        struct inode *inode, __u64 seconds)
2353 {
2354         struct osd_thread_info  *oti = osd_oti_get(env);
2355         struct timespec         *t   = &oti->oti_time;
2356
2357         t->tv_sec = seconds;
2358         t->tv_nsec = 0;
2359         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2360         return t;
2361 }
2362
2363 static void osd_inode_getattr(const struct lu_env *env,
2364                               struct inode *inode, struct lu_attr *attr)
2365 {
2366         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2367                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2368                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2369                            LA_BLKSIZE | LA_TYPE;
2370
2371         attr->la_atime   = LTIME_S(inode->i_atime);
2372         attr->la_mtime   = LTIME_S(inode->i_mtime);
2373         attr->la_ctime   = LTIME_S(inode->i_ctime);
2374         attr->la_mode    = inode->i_mode;
2375         attr->la_size    = i_size_read(inode);
2376         attr->la_blocks  = inode->i_blocks;
2377         attr->la_uid     = i_uid_read(inode);
2378         attr->la_gid     = i_gid_read(inode);
2379         attr->la_projid  = i_projid_read(inode);
2380         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2381         attr->la_nlink   = inode->i_nlink;
2382         attr->la_rdev    = inode->i_rdev;
2383         attr->la_blksize = 1 << inode->i_blkbits;
2384         attr->la_blkbits = inode->i_blkbits;
2385         /*
2386          * Ext4 did not transfer inherit flags from raw inode
2387          * to inode flags, and ext4 internally test raw inode
2388          * @i_flags directly. Instead of patching ext4, we do it here.
2389          */
2390         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2391                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2392 }
2393
2394 static int osd_attr_get(const struct lu_env *env,
2395                         struct dt_object *dt,
2396                         struct lu_attr *attr)
2397 {
2398         struct osd_object *obj = osd_dt_obj(dt);
2399
2400         if (unlikely(!dt_object_exists(dt)))
2401                 return -ENOENT;
2402         if (unlikely(obj->oo_destroyed))
2403                 return -ENOENT;
2404
2405         LASSERT(!dt_object_remote(dt));
2406         LINVRNT(osd_invariant(obj));
2407
2408         spin_lock(&obj->oo_guard);
2409         osd_inode_getattr(env, obj->oo_inode, attr);
2410         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2411                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2412         spin_unlock(&obj->oo_guard);
2413
2414         return 0;
2415 }
2416
2417 static int osd_declare_attr_qid(const struct lu_env *env,
2418                                 struct osd_object *obj,
2419                                 struct osd_thandle *oh, long long bspace,
2420                                 qid_t old_id, qid_t new_id, bool enforce,
2421                                 unsigned type)
2422 {
2423         int rc;
2424         struct osd_thread_info *info = osd_oti_get(env);
2425         struct lquota_id_info  *qi = &info->oti_qi;
2426
2427         qi->lqi_type = type;
2428         /* inode accounting */
2429         qi->lqi_is_blk = false;
2430
2431         /* one more inode for the new id ... */
2432         qi->lqi_id.qid_uid = new_id;
2433         qi->lqi_space      = 1;
2434         /* Reserve credits for the new id */
2435         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2436         if (rc == -EDQUOT || rc == -EINPROGRESS)
2437                 rc = 0;
2438         if (rc)
2439                 RETURN(rc);
2440
2441         /* and one less inode for the current id */
2442         qi->lqi_id.qid_uid = old_id;
2443         qi->lqi_space      = -1;
2444         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2445         if (rc == -EDQUOT || rc == -EINPROGRESS)
2446                 rc = 0;
2447         if (rc)
2448                 RETURN(rc);
2449
2450         /* block accounting */
2451         qi->lqi_is_blk = true;
2452
2453         /* more blocks for the new id ... */
2454         qi->lqi_id.qid_uid = new_id;
2455         qi->lqi_space      = bspace;
2456         /*
2457          * Credits for the new uid has been reserved, re-use "obj"
2458          * to save credit reservation.
2459          */
2460         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2461         if (rc == -EDQUOT || rc == -EINPROGRESS)
2462                 rc = 0;
2463         if (rc)
2464                 RETURN(rc);
2465
2466         /* and finally less blocks for the current uid */
2467         qi->lqi_id.qid_uid = old_id;
2468         qi->lqi_space      = -bspace;
2469         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2470         if (rc == -EDQUOT || rc == -EINPROGRESS)
2471                 rc = 0;
2472
2473         RETURN(rc);
2474 }
2475
2476 static int osd_declare_attr_set(const struct lu_env *env,
2477                                 struct dt_object *dt,
2478                                 const struct lu_attr *attr,
2479                                 struct thandle *handle)
2480 {
2481         struct osd_thandle     *oh;
2482         struct osd_object      *obj;
2483         qid_t                   uid;
2484         qid_t                   gid;
2485         long long               bspace;
2486         int                     rc = 0;
2487         bool                    enforce;
2488         ENTRY;
2489
2490         LASSERT(dt != NULL);
2491         LASSERT(handle != NULL);
2492
2493         obj = osd_dt_obj(dt);
2494         LASSERT(osd_invariant(obj));
2495
2496         oh = container_of0(handle, struct osd_thandle, ot_super);
2497         LASSERT(oh->ot_handle == NULL);
2498
2499         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2500                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2501
2502         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2503                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2504
2505         if (attr == NULL || obj->oo_inode == NULL)
2506                 RETURN(rc);
2507
2508         bspace   = obj->oo_inode->i_blocks;
2509         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2510         bspace   = toqb(bspace);
2511
2512         /* Changing ownership is always preformed by super user, it should not
2513          * fail with EDQUOT.
2514          *
2515          * We still need to call the osd_declare_qid() to calculate the journal
2516          * credits for updating quota accounting files and to trigger quota
2517          * space adjustment once the operation is completed.*/
2518         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2519                 /* USERQUOTA */
2520                 uid = i_uid_read(obj->oo_inode);
2521                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2522                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2523                                           attr->la_uid, enforce, USRQUOTA);
2524                 if (rc)
2525                         RETURN(rc);
2526
2527                 gid = i_gid_read(obj->oo_inode);
2528                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2529                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2530                                           i_gid_read(obj->oo_inode),
2531                                           attr->la_gid, enforce, GRPQUOTA);
2532                 if (rc)
2533                         RETURN(rc);
2534
2535         }
2536 #ifdef HAVE_PROJECT_QUOTA
2537         if (attr->la_valid & LA_PROJID) {
2538                 __u32 projid = i_projid_read(obj->oo_inode);
2539                 enforce = (attr->la_valid & LA_PROJID) &&
2540                                         (attr->la_projid != projid);
2541                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2542                                           (qid_t)projid, (qid_t)attr->la_projid,
2543                                           enforce, PRJQUOTA);
2544                 if (rc)
2545                         RETURN(rc);
2546         }
2547 #endif
2548         RETURN(rc);
2549 }
2550
2551 static int osd_inode_setattr(const struct lu_env *env,
2552                              struct inode *inode, const struct lu_attr *attr)
2553 {
2554         __u64 bits = attr->la_valid;
2555
2556         /* Only allow set size for regular file */
2557         if (!S_ISREG(inode->i_mode))
2558                 bits &= ~(LA_SIZE | LA_BLOCKS);
2559
2560         if (bits == 0)
2561                 return 0;
2562
2563         if (bits & LA_ATIME)
2564                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2565         if (bits & LA_CTIME)
2566                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2567         if (bits & LA_MTIME)
2568                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2569         if (bits & LA_SIZE) {
2570                 spin_lock(&inode->i_lock);
2571                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2572                 i_size_write(inode, attr->la_size);
2573                 spin_unlock(&inode->i_lock);
2574         }
2575
2576         /* OSD should not change "i_blocks" which is used by quota.
2577          * "i_blocks" should be changed by ldiskfs only. */
2578         if (bits & LA_MODE)
2579                 inode->i_mode = (inode->i_mode & S_IFMT) |
2580                                 (attr->la_mode & ~S_IFMT);
2581         if (bits & LA_UID)
2582                 i_uid_write(inode, attr->la_uid);
2583         if (bits & LA_GID)
2584                 i_gid_write(inode, attr->la_gid);
2585         if (bits & LA_PROJID)
2586                 i_projid_write(inode, attr->la_projid);
2587         if (bits & LA_NLINK)
2588                 set_nlink(inode, attr->la_nlink);
2589         if (bits & LA_RDEV)
2590                 inode->i_rdev = attr->la_rdev;
2591
2592         if (bits & LA_FLAGS) {
2593                 /* always keep S_NOCMTIME */
2594                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2595                                  S_NOCMTIME;
2596                 /*
2597                  * Ext4 did not transfer inherit flags from
2598                  * @inode->i_flags to raw inode i_flags when writing
2599                  * flags, we do it explictly here.
2600                  */
2601                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2602                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2603         }
2604         return 0;
2605 }
2606
2607 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2608 {
2609         int rc;
2610
2611         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2612             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2613                 struct iattr    iattr;
2614
2615                 ll_vfs_dq_init(inode);
2616                 iattr.ia_valid = 0;
2617                 if (attr->la_valid & LA_UID)
2618                         iattr.ia_valid |= ATTR_UID;
2619                 if (attr->la_valid & LA_GID)
2620                         iattr.ia_valid |= ATTR_GID;
2621                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2622                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2623
2624                 rc = ll_vfs_dq_transfer(inode, &iattr);
2625                 if (rc) {
2626                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2627                                "enforcement enabled on the ldiskfs "
2628                                "filesystem?\n", inode->i_sb->s_id, rc);
2629                         return rc;
2630                 }
2631         }
2632
2633 #ifdef HAVE_PROJECT_QUOTA
2634         /* Handle project id transfer here properly */
2635         if (attr->la_valid & LA_PROJID &&
2636             attr->la_projid != i_projid_read(inode)) {
2637                 rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid);
2638                 if (rc) {
2639                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2640                                "enforcement enabled on the ldiskfs "
2641                                "filesystem?\n", inode->i_sb->s_id, rc);
2642                         return rc;
2643                 }
2644         }
2645 #endif
2646         return 0;
2647 }
2648
2649 static int osd_attr_set(const struct lu_env *env,
2650                         struct dt_object *dt,
2651                         const struct lu_attr *attr,
2652                         struct thandle *handle)
2653 {
2654         struct osd_object *obj = osd_dt_obj(dt);
2655         struct inode      *inode;
2656         int rc;
2657
2658         if (!dt_object_exists(dt))
2659                 return -ENOENT;
2660
2661         LASSERT(handle != NULL);
2662         LASSERT(!dt_object_remote(dt));
2663         LASSERT(osd_invariant(obj));
2664
2665         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2666
2667         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2668                 struct osd_thread_info  *oti  = osd_oti_get(env);
2669                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2670                 struct lu_fid           *fid1 = &oti->oti_fid;
2671                 struct osd_inode_id     *id   = &oti->oti_id;
2672                 struct iam_path_descr   *ipd;
2673                 struct iam_container    *bag;
2674                 struct osd_thandle      *oh;
2675                 int                      rc;
2676
2677                 fid_cpu_to_be(fid1, fid0);
2678                 memset(id, 1, sizeof(*id));
2679                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2680                                   fid0)->oi_dir.od_container;
2681                 ipd = osd_idx_ipd_get(env, bag);
2682                 if (unlikely(ipd == NULL))
2683                         RETURN(-ENOMEM);
2684
2685                 oh = container_of0(handle, struct osd_thandle, ot_super);
2686                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2687                                 (const struct iam_rec *)id, ipd);
2688                 osd_ipd_put(env, bag, ipd);
2689                 return(rc > 0 ? 0 : rc);
2690         }
2691
2692         inode = obj->oo_inode;
2693
2694         rc = osd_quota_transfer(inode, attr);
2695         if (rc)
2696                 return rc;
2697
2698         spin_lock(&obj->oo_guard);
2699         rc = osd_inode_setattr(env, inode, attr);
2700         spin_unlock(&obj->oo_guard);
2701         if (rc != 0)
2702                 GOTO(out, rc);
2703
2704         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2705
2706         if (!(attr->la_valid & LA_FLAGS))
2707                 GOTO(out, rc);
2708
2709         /* Let's check if there are extra flags need to be set into LMA */
2710         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2711                 struct osd_thread_info *info = osd_oti_get(env);
2712                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2713
2714                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2715                                  &info->oti_ost_attrs);
2716                 if (rc)
2717                         GOTO(out, rc);
2718
2719                 lma->lma_incompat |=
2720                         lustre_to_lma_flags(attr->la_flags);
2721                 lustre_lma_swab(lma);
2722                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2723                                      lma, sizeof(*lma), XATTR_REPLACE);
2724                 if (rc != 0) {
2725                         struct osd_device *osd = osd_obj2dev(obj);
2726
2727                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2728                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2729                               lma->lma_incompat, rc);
2730                 } else {
2731                         obj->oo_lma_flags =
2732                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2733                 }
2734                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2735         }
2736 out:
2737         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2738
2739         return rc;
2740 }
2741
2742 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2743                                            struct osd_object *obj,
2744                                            const char *name, const int namelen)
2745 {
2746         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2747 }
2748
2749 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2750                       umode_t mode, struct dt_allocation_hint *hint,
2751                       struct thandle *th)
2752 {
2753         int result;
2754         struct osd_device  *osd = osd_obj2dev(obj);
2755         struct osd_thandle *oth;
2756         struct dt_object   *parent = NULL;
2757         struct inode       *inode;
2758
2759         LINVRNT(osd_invariant(obj));
2760         LASSERT(obj->oo_inode == NULL);
2761         LASSERT(obj->oo_hl_head == NULL);
2762
2763         if (S_ISDIR(mode) && ldiskfs_pdo) {
2764                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2765                 if (obj->oo_hl_head == NULL)
2766                         return -ENOMEM;
2767         }
2768
2769         oth = container_of(th, struct osd_thandle, ot_super);
2770         LASSERT(oth->ot_handle->h_transaction != NULL);
2771
2772         if (hint != NULL && hint->dah_parent != NULL &&
2773             !dt_object_remote(hint->dah_parent))
2774                 parent = hint->dah_parent;
2775
2776         inode = ldiskfs_create_inode(oth->ot_handle,
2777                                      parent ? osd_dt_obj(parent)->oo_inode :
2778                                               osd_sb(osd)->s_root->d_inode,
2779                                      mode);
2780         if (!IS_ERR(inode)) {
2781                 /* Do not update file c/mtime in ldiskfs. */
2782                 inode->i_flags |= S_NOCMTIME;
2783
2784                 /* For new created object, it must be consistent,
2785                  * and it is unnecessary to scrub against it. */
2786                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2787
2788                 obj->oo_inode = inode;
2789                 result = 0;
2790         } else {
2791                 if (obj->oo_hl_head != NULL) {
2792                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2793                         obj->oo_hl_head = NULL;
2794                 }
2795                 result = PTR_ERR(inode);
2796         }
2797         LINVRNT(osd_invariant(obj));
2798         return result;
2799 }
2800
2801 enum {
2802         OSD_NAME_LEN = 255
2803 };
2804
2805 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2806                      struct lu_attr *attr,
2807                      struct dt_allocation_hint *hint,
2808                      struct dt_object_format *dof,
2809                      struct thandle *th)
2810 {
2811         int result;
2812         struct osd_thandle *oth;
2813         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2814
2815         LASSERT(S_ISDIR(attr->la_mode));
2816
2817         oth = container_of(th, struct osd_thandle, ot_super);
2818         LASSERT(oth->ot_handle->h_transaction != NULL);
2819         result = osd_mkfile(info, obj, mode, hint, th);
2820
2821         return result;
2822 }
2823
2824 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2825                         struct lu_attr *attr,
2826                         struct dt_allocation_hint *hint,
2827                         struct dt_object_format *dof,
2828                         struct thandle *th)
2829 {
2830         int result;
2831         struct osd_thandle *oth;
2832         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2833
2834         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2835
2836         LASSERT(S_ISREG(attr->la_mode));
2837
2838         oth = container_of(th, struct osd_thandle, ot_super);
2839         LASSERT(oth->ot_handle->h_transaction != NULL);
2840
2841         result = osd_mkfile(info, obj, mode, hint, th);
2842         if (result == 0) {
2843                 LASSERT(obj->oo_inode != NULL);
2844                 if (feat->dif_flags & DT_IND_VARKEY)
2845                         result = iam_lvar_create(obj->oo_inode,
2846                                                  feat->dif_keysize_max,
2847                                                  feat->dif_ptrsize,
2848                                                  feat->dif_recsize_max,
2849                                                  oth->ot_handle);
2850                 else
2851                         result = iam_lfix_create(obj->oo_inode,
2852                                                  feat->dif_keysize_max,
2853                                                  feat->dif_ptrsize,
2854                                                  feat->dif_recsize_max,
2855                                                  oth->ot_handle);
2856
2857         }
2858         return result;
2859 }
2860
2861 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2862                      struct lu_attr *attr,
2863                      struct dt_allocation_hint *hint,
2864                      struct dt_object_format *dof,
2865                      struct thandle *th)
2866 {
2867         LASSERT(S_ISREG(attr->la_mode));
2868         return osd_mkfile(info, obj, (attr->la_mode &
2869                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2870 }
2871
2872 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2873                      struct lu_attr *attr,
2874                      struct dt_allocation_hint *hint,
2875                      struct dt_object_format *dof,
2876                      struct thandle *th)
2877 {
2878         LASSERT(S_ISLNK(attr->la_mode));
2879         return osd_mkfile(info, obj, (attr->la_mode &
2880                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2881 }
2882
2883 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2884                      struct lu_attr *attr,
2885                      struct dt_allocation_hint *hint,
2886                      struct dt_object_format *dof,
2887                      struct thandle *th)
2888 {
2889         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2890         int result;
2891
2892         LINVRNT(osd_invariant(obj));
2893         LASSERT(obj->oo_inode == NULL);
2894         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2895                 S_ISFIFO(mode) || S_ISSOCK(mode));
2896
2897         result = osd_mkfile(info, obj, mode, hint, th);
2898         if (result == 0) {
2899                 LASSERT(obj->oo_inode != NULL);
2900                 /*
2901                  * This inode should be marked dirty for i_rdev.  Currently
2902                  * that is done in the osd_attr_init().
2903                  */
2904                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2905                                    attr->la_rdev);
2906         }
2907         LINVRNT(osd_invariant(obj));
2908         return result;
2909 }
2910
2911 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2912                               struct lu_attr *,
2913                               struct dt_allocation_hint *hint,
2914                               struct dt_object_format *dof,
2915                               struct thandle *);
2916
2917 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2918 {
2919         osd_obj_type_f result;
2920
2921         switch (type) {
2922         case DFT_DIR:
2923                 result = osd_mkdir;
2924                 break;
2925         case DFT_REGULAR:
2926                 result = osd_mkreg;
2927                 break;
2928         case DFT_SYM:
2929                 result = osd_mksym;
2930                 break;
2931         case DFT_NODE:
2932                 result = osd_mknod;
2933                 break;
2934         case DFT_INDEX:
2935                 result = osd_mk_index;
2936                 break;
2937
2938         default:
2939                 LBUG();
2940                 break;
2941         }
2942         return result;
2943 }
2944
2945
2946 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2947                         struct dt_object *parent, struct dt_object *child,
2948                         umode_t child_mode)
2949 {
2950         LASSERT(ah);
2951
2952         ah->dah_parent = parent;
2953         ah->dah_mode = child_mode;
2954
2955         if (parent != NULL && !dt_object_remote(parent)) {
2956                 /* will help to find FID->ino at dt_insert("..") */
2957                 struct osd_object *pobj = osd_dt_obj(parent);
2958                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2959         }
2960 }
2961
2962 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2963                           struct lu_attr *attr, struct dt_object_format *dof)
2964 {
2965         struct inode   *inode = obj->oo_inode;
2966         __u64           valid = attr->la_valid;
2967         int             result;
2968
2969         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2970
2971         if (dof->dof_type != DFT_NODE)
2972                 attr->la_valid &= ~LA_RDEV;
2973         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2974                 attr->la_valid &= ~LA_ATIME;
2975         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2976                 attr->la_valid &= ~LA_CTIME;
2977         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2978                 attr->la_valid &= ~LA_MTIME;
2979
2980         result = osd_quota_transfer(inode, attr);
2981         if (result)
2982                 return;
2983
2984         if (attr->la_valid != 0) {
2985                 result = osd_inode_setattr(info->oti_env, inode, attr);
2986                 /*
2987                  * The osd_inode_setattr() should always succeed here.  The
2988                  * only error that could be returned is EDQUOT when we are
2989                  * trying to change the UID or GID of the inode. However, this
2990                  * should not happen since quota enforcement is no longer
2991                  * enabled on ldiskfs (lquota takes care of it).
2992                  */
2993                 LASSERTF(result == 0, "%d\n", result);
2994                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2995         }
2996
2997         attr->la_valid = valid;
2998 }
2999
3000 /**
3001  * Helper function for osd_object_create()
3002  *
3003  * \retval 0, on success
3004  */
3005 static int __osd_object_create(struct osd_thread_info *info,
3006                                struct osd_object *obj, struct lu_attr *attr,
3007                                struct dt_allocation_hint *hint,
3008                                struct dt_object_format *dof,
3009                                struct thandle *th)
3010 {
3011         int     result;
3012         __u32   umask;
3013
3014         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3015
3016         /* we drop umask so that permissions we pass are not affected */
3017         umask = current->fs->umask;
3018         current->fs->umask = 0;
3019
3020         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3021                                                   th);
3022         if (likely(obj->oo_inode != NULL)) {
3023                 LASSERT(obj->oo_inode->i_state & I_NEW);
3024
3025                 /* Unlock the inode before attr initialization to avoid
3026                  * unnecessary dqget operations. LU-6378 */
3027                 unlock_new_inode(obj->oo_inode);
3028         }
3029
3030         if (likely(result == 0)) {
3031                 osd_attr_init(info, obj, attr, dof);
3032                 osd_object_init0(obj);
3033         }
3034
3035         /* restore previous umask value */
3036         current->fs->umask = umask;
3037
3038         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3039
3040         return result;
3041 }
3042
3043 /**
3044  * Helper function for osd_object_create()
3045  *
3046  * \retval 0, on success
3047  */
3048 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3049                            const struct lu_fid *fid, struct thandle *th)
3050 {
3051         struct osd_thread_info *info = osd_oti_get(env);
3052         struct osd_inode_id    *id   = &info->oti_id;
3053         struct osd_device      *osd  = osd_obj2dev(obj);
3054         struct osd_thandle     *oh;
3055         int                     rc;
3056
3057         LASSERT(obj->oo_inode != NULL);
3058
3059         oh = container_of0(th, struct osd_thandle, ot_super);
3060         LASSERT(oh->ot_handle);
3061         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3062
3063         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3064         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3065                            OI_CHECK_FLD, NULL);
3066         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3067
3068         return rc;
3069 }
3070
3071 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3072                    u64 seq, struct lu_seq_range *range)
3073 {
3074         struct seq_server_site  *ss = osd_seq_site(osd);
3075
3076         if (fid_seq_is_idif(seq)) {
3077                 fld_range_set_ost(range);
3078                 range->lsr_index = idif_ost_idx(seq);
3079                 return 0;
3080         }
3081
3082         if (!fid_seq_in_fldb(seq)) {
3083                 fld_range_set_mdt(range);
3084                 if (ss != NULL)
3085                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3086                          * at all */
3087                         range->lsr_index = ss->ss_node_id;
3088                 return 0;
3089         }
3090
3091         LASSERT(ss != NULL);
3092         fld_range_set_any(range);
3093         /* OSD will only do local fld lookup */
3094         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3095 }
3096
3097 /*
3098  * Concurrency: no external locking is necessary.
3099  */
3100 static int osd_declare_object_create(const struct lu_env *env,
3101                                      struct dt_object *dt,
3102                                      struct lu_attr *attr,
3103                                      struct dt_allocation_hint *hint,
3104                                      struct dt_object_format *dof,
3105                                      struct thandle *handle)
3106 {
3107         struct osd_thandle      *oh;
3108         int                      rc;
3109         ENTRY;
3110
3111         LASSERT(handle != NULL);
3112
3113         oh = container_of0(handle, struct osd_thandle, ot_super);
3114         LASSERT(oh->ot_handle == NULL);
3115
3116         /* EA object consumes more credits than regular object: osd_mk_index
3117          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3118          * leaf_node, could involves the block, block bitmap, groups, GDT
3119          * change for each block, so add 4 * 2 credits in that case. */
3120         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3121                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3122                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3123         /* Reuse idle OI block may cause additional one OI block
3124          * to be changed. */
3125         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3126                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3127
3128         if (!attr)
3129                 RETURN(0);
3130
3131         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3132                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3133                                    NULL, OSD_QID_INODE);
3134         if (rc != 0)
3135                 RETURN(rc);
3136
3137         /* will help to find FID->ino mapping at dt_insert() */
3138         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3139                                    osd_dt_obj(dt));
3140
3141         RETURN(rc);
3142 }
3143
3144 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3145                              struct lu_attr *attr,
3146                              struct dt_allocation_hint *hint,
3147                              struct dt_object_format *dof, struct thandle *th)
3148 {
3149         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3150         struct osd_object       *obj    = osd_dt_obj(dt);
3151         struct osd_thread_info  *info   = osd_oti_get(env);
3152         int result;
3153         ENTRY;
3154
3155         if (dt_object_exists(dt))
3156                 return -EEXIST;
3157
3158         LINVRNT(osd_invariant(obj));
3159         LASSERT(!dt_object_remote(dt));
3160         LASSERT(osd_write_locked(env, obj));
3161         LASSERT(th != NULL);
3162
3163         if (unlikely(fid_is_acct(fid)))
3164                 /* Quota files can't be created from the kernel any more,
3165                  * 'tune2fs -O quota' will take care of creating them */
3166                 RETURN(-EPERM);
3167
3168         result = __osd_object_create(info, obj, attr, hint, dof, th);
3169         if (result == 0) {
3170                 result = __osd_oi_insert(env, obj, fid, th);
3171                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3172                         obj->oo_dt.do_body_ops = &osd_body_ops;
3173         }
3174         LASSERT(ergo(result == 0,
3175                 dt_object_exists(dt) && !dt_object_remote(dt)));
3176
3177         LASSERT(osd_invariant(obj));
3178         RETURN(result);
3179 }
3180
3181 /**
3182  * Called to destroy on-disk representation of the object
3183  *
3184  * Concurrency: must be locked
3185  */
3186 static int osd_declare_object_destroy(const struct lu_env *env,
3187                                       struct dt_object *dt,
3188                                       struct thandle *th)
3189 {
3190         struct osd_object  *obj = osd_dt_obj(dt);
3191         struct inode       *inode = obj->oo_inode;
3192         struct osd_thandle *oh;
3193         int                 rc;
3194         ENTRY;
3195
3196         if (inode == NULL)
3197                 RETURN(-ENOENT);
3198
3199         oh = container_of0(th, struct osd_thandle, ot_super);
3200         LASSERT(oh->ot_handle == NULL);
3201
3202         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3203                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3204         /* Recycle idle OI leaf may cause additional three OI blocks
3205          * to be changed. */
3206         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3207                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3208                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3209         /* one less inode */
3210         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3211                                    i_projid_read(inode), -1, oh, obj, NULL,
3212                                    OSD_QID_INODE);
3213         if (rc)
3214                 RETURN(rc);
3215         /* data to be truncated */
3216         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3217                                    i_projid_read(inode), 0, oh, obj, NULL,
3218                                    OSD_QID_BLK);
3219         if (rc)
3220                 RETURN(rc);
3221
3222         /* will help to find FID->ino when this object is being
3223          * added to PENDING/ */
3224         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3225
3226         RETURN(rc);
3227 }
3228
3229 static int osd_object_destroy(const struct lu_env *env,
3230                               struct dt_object *dt,
3231                               struct thandle *th)
3232 {
3233         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
3234         struct osd_object      *obj = osd_dt_obj(dt);
3235         struct inode           *inode = obj->oo_inode;
3236         struct osd_device      *osd = osd_obj2dev(obj);
3237         struct osd_thandle     *oh;
3238         int                     result;
3239         ENTRY;
3240
3241         oh = container_of0(th, struct osd_thandle, ot_super);
3242         LASSERT(oh->ot_handle);
3243         LASSERT(inode);
3244         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3245
3246         if (unlikely(fid_is_acct(fid)))
3247                 RETURN(-EPERM);
3248
3249         if (S_ISDIR(inode->i_mode)) {
3250                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3251                         inode->i_nlink == 2);
3252                 /* it will check/delete the inode from remote parent,
3253                  * how to optimize it? unlink performance impaction XXX */
3254                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3255                 if (result != 0)
3256                         CERROR("%s: delete inode "DFID": rc = %d\n",
3257                                osd_name(osd), PFID(fid), result);
3258
3259                 spin_lock(&obj->oo_guard);
3260                 clear_nlink(inode);
3261                 spin_unlock(&obj->oo_guard);
3262                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3263         }
3264
3265         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3266
3267         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3268
3269         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3270                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3271                                        oh->ot_handle, OI_CHECK_FLD);
3272
3273         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3274         /* XXX: add to ext3 orphan list */
3275         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3276
3277         /* not needed in the cache anymore */
3278         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3279         obj->oo_destroyed = 1;
3280
3281         RETURN(0);
3282 }
3283
3284 /**
3285  * Put the fid into lustre_mdt_attrs, and then place the structure
3286  * inode's ea. This fid should not be altered during the life time
3287  * of the inode.
3288  *
3289  * \retval +ve, on success
3290  * \retval -ve, on error
3291  *
3292  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3293  */
3294 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3295                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3296 {
3297         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3298         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3299         int rc;
3300         ENTRY;
3301
3302         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3303                 RETURN(0);
3304
3305         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3306                 rc = -ENOMEM;
3307
3308         lustre_loa_init(loa, fid, compat, incompat);
3309         lustre_loa_swab(loa, false);
3310
3311         /* For the OST device with 256 bytes inode size by default,
3312          * the PFID EA will be stored together with LMA EA to avoid
3313          * performance trouble. Otherwise the PFID EA can be stored
3314          * independently. LU-8998 */
3315         if ((compat & LMAC_FID_ON_OST) &&
3316             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3317                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3318                                      sizeof(*loa), XATTR_CREATE);
3319         else
3320                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3321                                      sizeof(*lma), XATTR_CREATE);
3322         /* LMA may already exist, but we need to check that all the
3323          * desired compat/incompat flags have been added. */
3324         if (unlikely(rc == -EEXIST)) {
3325                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3326                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3327                 if (rc < 0)
3328                         RETURN(rc);
3329
3330                 if (rc < sizeof(*lma))
3331                         RETURN(-EINVAL);
3332
3333                 lustre_loa_swab(loa, true);
3334                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3335                     ((compat == 0 && incompat == 0) ||
3336                      (!(~lma->lma_compat & compat) &&
3337                       !(~lma->lma_incompat & incompat))))
3338                         RETURN(0);
3339
3340                 lma->lma_self_fid = *fid;
3341                 lma->lma_compat |= compat;
3342                 lma->lma_incompat |= incompat;
3343                 if (rc == sizeof(*lma)) {
3344                         lustre_lma_swab(lma);
3345                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3346                                              sizeof(*lma), XATTR_REPLACE);
3347                 } else {
3348                         lustre_loa_swab(loa, false);
3349                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3350                                              sizeof(*loa), XATTR_REPLACE);
3351                 }
3352         }
3353
3354         RETURN(rc);
3355 }
3356
3357 /**
3358  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3359  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3360  * To have compatilibility with 1.8 ldiskfs driver we need to have
3361  * magic number at start of fid data.
3362  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3363  * its inmemory API.
3364  */
3365 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3366                                          const struct lu_fid *fid)
3367 {
3368         if (!fid_is_namespace_visible(fid) ||
3369             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3370                 param->edp_magic = 0;
3371                 return;
3372         }
3373
3374         param->edp_magic = LDISKFS_LUFID_MAGIC;
3375         param->edp_len =  sizeof(struct lu_fid) + 1;
3376         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3377 }
3378
3379 /**
3380  * Try to read the fid from inode ea into dt_rec.
3381  *
3382  * \param fid object fid.
3383  *
3384  * \retval 0 on success
3385  */
3386 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3387                           __u32 ino, struct lu_fid *fid,
3388                           struct osd_inode_id *id)
3389 {
3390         struct osd_thread_info *info  = osd_oti_get(env);
3391         struct inode           *inode;
3392         ENTRY;
3393
3394         osd_id_gen(id, ino, OSD_OII_NOGEN);
3395         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3396         if (IS_ERR(inode))
3397                 RETURN(PTR_ERR(inode));
3398
3399         iput(inode);
3400         RETURN(0);
3401 }
3402
3403 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3404                                         struct inode *dir,
3405                                         struct inode *parent_dir,
3406                                         const struct lu_fid *dot_fid,
3407                                         const struct lu_fid *dot_dot_fid,
3408                                         struct osd_thandle *oth)
3409 {
3410         struct ldiskfs_dentry_param *dot_ldp;
3411         struct ldiskfs_dentry_param *dot_dot_ldp;
3412         __u32 saved_nlink = dir->i_nlink;
3413         int rc;
3414
3415         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3416         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3417
3418         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3419         dot_ldp->edp_magic = 0;
3420
3421         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3422                                     dir, dot_ldp, dot_dot_ldp);
3423         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3424          * the subseqent ref_add() will increase the dir->i_nlink
3425          * as 3. That is incorrect for new created directory.
3426          *
3427          * It looks like hack, because we want to make the OSD API
3428          * to be order-independent for new created directory object
3429          * between dt_insert(..) and ref_add() operations.
3430          *
3431          * Here, we only restore the in-RAM dir-inode's nlink attr,
3432          * becuase if the nlink attr is not 2, then there will be
3433          * ref_add() called following the dt_insert(..), such call
3434          * will make both the in-RAM and on-disk dir-inode's nlink
3435          * attr to be set as 2. LU-7447 */
3436         set_nlink(dir, saved_nlink);
3437         return rc;
3438 }
3439
3440 /**
3441  * Create an local agent inode for remote entry
3442  */
3443 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3444                                                   struct osd_device *osd,
3445                                                   struct osd_object *pobj,
3446                                                   const struct lu_fid *fid,
3447                                                   __u32 type,
3448                                                   struct thandle *th)
3449 {
3450         struct osd_thread_info  *info = osd_oti_get(env);
3451         struct inode            *local;
3452         struct osd_thandle      *oh;
3453         int                     rc;
3454         ENTRY;
3455
3456         LASSERT(th);
3457         oh = container_of(th, struct osd_thandle, ot_super);
3458         LASSERT(oh->ot_handle->h_transaction != NULL);
3459
3460         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3461         if (IS_ERR(local)) {
3462                 CERROR("%s: create local error %d\n", osd_name(osd),
3463                        (int)PTR_ERR(local));
3464                 RETURN(local);
3465         }
3466
3467         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3468          * correct gid on remote file, not agent here */
3469         local->i_gid = current_fsgid();
3470         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3471         unlock_new_inode(local);
3472
3473         /* Agent inode should not have project ID */
3474 #ifdef  HAVE_PROJECT_QUOTA
3475         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL) {
3476                 rc = __ldiskfs_ioctl_setproject(local, 0);
3477                 if (rc) {
3478                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3479                                "quota enforcement enabled on the ldiskfs "
3480                                "filesystem?\n", local->i_sb->s_id, rc);
3481                         RETURN(ERR_PTR(rc));
3482                 }
3483         }
3484 #endif
3485         /* Set special LMA flag for local agent inode */
3486         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3487         if (rc != 0) {
3488                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3489                        osd_name(osd), PFID(fid), rc);
3490                 RETURN(ERR_PTR(rc));
3491         }
3492
3493         if (!S_ISDIR(type))
3494                 RETURN(local);
3495
3496         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3497                                          lu_object_fid(&pobj->oo_dt.do_lu),
3498                                          fid, oh);
3499         if (rc != 0) {
3500                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3501                         osd_name(osd), PFID(fid), rc);
3502                 RETURN(ERR_PTR(rc));
3503         }
3504
3505         RETURN(local);
3506 }
3507
3508 /**
3509  * when direntry is deleted, we have to take care of possible agent inode
3510  * referenced by that. unfortunately we can't do this at that point:
3511  * iget() within a running transaction leads to deadlock and we better do
3512  * not call that every delete declaration to save performance. so we put
3513  * a potention agent inode on a list and process that once the transaction
3514  * is over. Notice it's not any worse in terms of real orphans as regular
3515  * object destroy doesn't put inodes on the on-disk orphan list. this should
3516  * be addressed separately
3517  */
3518 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3519                                             struct osd_thandle *oh,
3520                                             __u32 ino)
3521 {
3522         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3523         struct osd_obj_orphan *oor;
3524
3525         OBD_ALLOC_PTR(oor);
3526         if (oor == NULL)
3527                 return -ENOMEM;
3528
3529         oor->oor_ino = ino;
3530         oor->oor_env = (struct lu_env *)env;
3531         spin_lock(&osd->od_osfs_lock);
3532         list_add(&oor->oor_list, &osd->od_orphan_list);
3533         spin_unlock(&osd->od_osfs_lock);
3534
3535         oh->ot_remove_agents = 1;
3536
3537         return 0;
3538
3539 }
3540
3541 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3542                                                 struct osd_device *osd)
3543 {
3544         struct osd_thread_info *info = osd_oti_get(env);
3545         struct osd_obj_orphan *oor, *tmp;
3546         struct osd_inode_id id;
3547         struct list_head list;
3548         struct inode *inode;
3549         struct lu_fid fid;
3550         handle_t *jh;
3551         __u32 ino;
3552
3553         INIT_LIST_HEAD(&list);
3554
3555         spin_lock(&osd->od_osfs_lock);
3556         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3557                 if (oor->oor_env == env) {
3558                         list_del(&oor->oor_list);
3559                         list_add(&oor->oor_list, &list);
3560                 }
3561         }
3562         spin_unlock(&osd->od_osfs_lock);
3563
3564         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3565
3566                 ino = oor->oor_ino;
3567
3568                 list_del(&oor->oor_list);
3569                 OBD_FREE_PTR(oor);
3570
3571                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3572                 inode = osd_iget_fid(info, osd, &id, &fid);
3573                 if (IS_ERR(inode))
3574                         continue;
3575
3576                 if (!osd_remote_fid(env, osd, &fid)) {
3577                         iput(inode);
3578                         continue;
3579                 }
3580
3581                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3582                 clear_nlink(inode);
3583                 mark_inode_dirty(inode);
3584                 ldiskfs_journal_stop(jh);
3585                 iput(inode);
3586         }
3587
3588         return 0;
3589 }
3590
3591 /**
3592  * OSD layer object create function for interoperability mode (b11826).
3593  * This is mostly similar to osd_object_create(). Only difference being, fid is
3594  * inserted into inode ea here.
3595  *
3596  * \retval   0, on success
3597  * \retval -ve, on error
3598  */
3599 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3600                                 struct lu_attr *attr,
3601                                 struct dt_allocation_hint *hint,
3602                                 struct dt_object_format *dof,
3603                                 struct thandle *th)
3604 {
3605         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3606         struct osd_object       *obj    = osd_dt_obj(dt);
3607         struct osd_thread_info  *info   = osd_oti_get(env);
3608         int                      result, on_ost = 0;
3609
3610         ENTRY;
3611
3612         if (dt_object_exists(dt))
3613                 RETURN(-EEXIST);
3614
3615         LASSERT(osd_invariant(obj));
3616         LASSERT(!dt_object_remote(dt));
3617         LASSERT(osd_write_locked(env, obj));
3618         LASSERT(th != NULL);
3619
3620         if (unlikely(fid_is_acct(fid)))
3621                 /* Quota files can't be created from the kernel any more,
3622                  * 'tune2fs -O quota' will take care of creating them */
3623                 RETURN(-EPERM);
3624
3625         result = __osd_object_create(info, obj, attr, hint, dof, th);
3626         if (result == 0) {
3627                 if (fid_is_idif(fid) &&
3628                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3629                         struct lu_fid *tfid = &info->oti_fid;
3630                         struct ost_id *oi   = &info->oti_ostid;
3631
3632                         fid_to_ostid(fid, oi);
3633                         ostid_to_fid(tfid, oi, 0);
3634                         on_ost = 1;
3635                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3636                                                 LMAC_FID_ON_OST, 0);
3637                 } else {
3638                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3639                                                fid, OI_CHECK_FLD);
3640                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3641                                                 on_ost ? LMAC_FID_ON_OST : 0,
3642                                                 0);
3643                 }
3644                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3645                         obj->oo_dt.do_body_ops = &osd_body_ops;
3646         }
3647
3648         if (result == 0)
3649                 result = __osd_oi_insert(env, obj, fid, th);
3650
3651         /* a small optimization - dt_insert() isn't usually applied
3652          * to OST objects, so we don't need to cache OI mapping for
3653          * OST objects */
3654         if (result == 0 && on_ost == 0) {
3655                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3656                 result = osd_idc_find_and_init(env, osd, obj);
3657                 LASSERT(result == 0);
3658         }
3659
3660         LASSERT(ergo(result == 0,
3661                      dt_object_exists(dt) && !dt_object_remote(dt)));
3662         LINVRNT(osd_invariant(obj));
3663         RETURN(result);
3664 }
3665
3666 static int osd_declare_object_ref_add(const struct lu_env *env,
3667                                       struct dt_object *dt,
3668                                       struct thandle *handle)
3669 {
3670         struct osd_thandle       *oh;
3671
3672         /* it's possible that object doesn't exist yet */
3673         LASSERT(handle != NULL);
3674
3675         oh = container_of0(handle, struct osd_thandle, ot_super);
3676         LASSERT(oh->ot_handle == NULL);
3677
3678         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3679                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3680
3681         return 0;
3682 }
3683
3684 /*
3685  * Concurrency: @dt is write locked.
3686  */
3687 static int osd_object_ref_add(const struct lu_env *env,
3688                               struct dt_object *dt, struct thandle *th)
3689 {
3690         struct osd_object  *obj = osd_dt_obj(dt);
3691         struct inode       *inode = obj->oo_inode;
3692         struct osd_thandle *oh;
3693         int                 rc = 0;
3694
3695         if (!dt_object_exists(dt) || obj->oo_destroyed)
3696                 return -ENOENT;
3697
3698         LINVRNT(osd_invariant(obj));
3699         LASSERT(!dt_object_remote(dt));
3700         LASSERT(osd_write_locked(env, obj));
3701         LASSERT(th != NULL);
3702
3703         oh = container_of0(th, struct osd_thandle, ot_super);
3704         LASSERT(oh->ot_handle != NULL);
3705
3706         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3707
3708         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3709                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3710         /*
3711          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3712          * (65000) subdirectories by storing "1" in i_nlink if the link count
3713          * would otherwise overflow. Directory tranversal tools understand
3714          * that (st_nlink == 1) indicates that the filesystem dose not track
3715          * hard links count on the directory, and will not abort subdirectory
3716          * scanning early once (st_nlink - 2) subdirs have been found.
3717          *
3718          * This also has to properly handle the case of inodes with nlink == 0
3719          * in case they are being linked into the PENDING directory
3720          */
3721         spin_lock(&obj->oo_guard);
3722         if (unlikely(inode->i_nlink == 0))
3723                 /* inc_nlink from 0 may cause WARN_ON */
3724                 set_nlink(inode, 1);
3725         else {
3726                 ldiskfs_inc_count(oh->ot_handle, inode);
3727                 if (!S_ISDIR(inode->i_mode))
3728                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3729         }
3730         spin_unlock(&obj->oo_guard);
3731
3732         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3733         LINVRNT(osd_invariant(obj));
3734
3735         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3736
3737         return rc;
3738 }
3739
3740 static int osd_declare_object_ref_del(const struct lu_env *env,
3741                                       struct dt_object *dt,
3742                                       struct thandle *handle)
3743 {
3744         struct osd_thandle *oh;
3745
3746         if (!dt_object_exists(dt))
3747                 return -ENOENT;
3748
3749         LASSERT(!dt_object_remote(dt));
3750         LASSERT(handle != NULL);
3751
3752         oh = container_of0(handle, struct osd_thandle, ot_super);
3753         LASSERT(oh->ot_handle == NULL);
3754
3755         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3756                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3757
3758         return 0;
3759 }
3760
3761 /*
3762  * Concurrency: @dt is write locked.
3763  */
3764 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3765                               struct thandle *th)
3766 {
3767         struct osd_object       *obj = osd_dt_obj(dt);
3768         struct inode            *inode = obj->oo_inode;
3769         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3770         struct osd_thandle      *oh;
3771
3772         if (!dt_object_exists(dt))
3773                 return -ENOENT;
3774
3775         LINVRNT(osd_invariant(obj));
3776         LASSERT(!dt_object_remote(dt));
3777         LASSERT(osd_write_locked(env, obj));
3778         LASSERT(th != NULL);
3779
3780         oh = container_of0(th, struct osd_thandle, ot_super);
3781         LASSERT(oh->ot_handle != NULL);
3782
3783         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3784
3785         spin_lock(&obj->oo_guard);
3786         /* That can be result of upgrade from old Lustre version and
3787          * applied only to local files.  Just skip this ref_del call.
3788          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3789         if (inode->i_nlink == 0) {
3790                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3791                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3792                              ", maybe an upgraded file? (LU-3915)\n",
3793                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3794                 spin_unlock(&obj->oo_guard);
3795                 return 0;
3796         }
3797
3798         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3799                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3800
3801         ldiskfs_dec_count(oh->ot_handle, inode);
3802         spin_unlock(&obj->oo_guard);
3803
3804         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3805         LINVRNT(osd_invariant(obj));
3806
3807         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3808
3809         return 0;
3810 }
3811
3812 /*
3813  * Concurrency: @dt is read locked.
3814  */
3815 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3816                          struct lu_buf *buf, const char *name)
3817 {
3818         struct osd_object      *obj    = osd_dt_obj(dt);
3819         struct inode           *inode  = obj->oo_inode;
3820         struct osd_thread_info *info   = osd_oti_get(env);
3821         struct dentry          *dentry = &info->oti_obj_dentry;
3822         bool                    cache_xattr = false;
3823         int                     rc;
3824
3825         LASSERT(buf);
3826
3827         /* version get is not real XATTR but uses xattr API */
3828         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3829                 dt_obj_version_t *ver = buf->lb_buf;
3830
3831                 /* for version we are just using xattr API but change inode
3832                  * field instead */
3833                 if (buf->lb_len == 0)
3834                         return sizeof(dt_obj_version_t);
3835
3836                 if (buf->lb_len < sizeof(dt_obj_version_t))
3837                         return -ERANGE;
3838
3839                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3840                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3841
3842                 *ver = LDISKFS_I(inode)->i_fs_version;
3843
3844                 return sizeof(dt_obj_version_t);
3845         }
3846
3847         if (!dt_object_exists(dt))
3848                 return -ENOENT;
3849
3850         LASSERT(!dt_object_remote(dt));
3851         LASSERT(inode->i_op != NULL);
3852         LASSERT(inode->i_op->getxattr != NULL);
3853
3854         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3855             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3856                 cache_xattr = true;
3857
3858         if (cache_xattr) {
3859                 rc = osd_oxc_get(obj, name, buf);
3860                 if (rc != -ENOENT)
3861                         return rc;
3862         }
3863
3864         rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3865         if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3866                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3867                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3868                 struct filter_fid *ff;
3869                 struct ost_layout *ol;
3870
3871                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3872
3873                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3874                 if (rc)
3875                         return rc;
3876
3877                 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3878                         rc = -ENODATA;
3879                         goto cache;
3880                 }
3881
3882                 rc = sizeof(*ff);
3883                 if (buf->lb_len == 0 || !buf->lb_buf)
3884                         return rc;
3885
3886                 if (buf->lb_len < rc)
3887                         return -ERANGE;
3888
3889                 ff = buf->lb_buf;
3890                 ol = &ff->ff_layout;
3891                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3892                                                   PFID_STRIPE_IDX_BITS);
3893                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3894                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3895                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3896                 if (lma->lma_compat & LMAC_COMP_INFO) {
3897                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3898                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3899                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3900                 } else {
3901                         ol->ol_comp_start = 0;
3902                         ol->ol_comp_end = 0;
3903                         ol->ol_comp_id = 0;
3904                 }
3905         }
3906
3907 cache:
3908         if (cache_xattr) {
3909                 if (rc == -ENOENT || rc == -ENODATA)
3910                         osd_oxc_add(obj, name, NULL, 0);
3911                 else if (rc > 0 && buf->lb_buf != NULL)
3912                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3913         }
3914
3915         return rc;
3916 }
3917
3918 static int osd_declare_xattr_set(const struct lu_env *env,
3919                                  struct dt_object *dt,
3920                                  const struct lu_buf *buf, const char *name,
3921                                  int fl, struct thandle *handle)
3922 {
3923         struct osd_thandle *oh;
3924         int credits = 0;
3925         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3926
3927         LASSERT(handle != NULL);
3928
3929         oh = container_of0(handle, struct osd_thandle, ot_super);
3930         LASSERT(oh->ot_handle == NULL);
3931
3932         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3933                 /* For non-upgrading case, the LMA is set first and
3934                  * usually fit inode. But for upgrade case, the LMA
3935                  * may be in another separated EA block. */
3936                 if (dt_object_exists(dt)) {
3937                         if (fl == LU_XATTR_REPLACE)
3938                                 credits = 1;
3939                         else
3940                                 goto upgrade;
3941                 }
3942         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3943                 credits = 1;
3944         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3945                 /* We may need to delete the old PFID EA. */
3946                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3947                 if (fl == LU_XATTR_REPLACE)
3948                         credits += 1;
3949                 else
3950                         goto upgrade;
3951         } else {
3952
3953 upgrade:
3954                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3955
3956                 if (buf != NULL) {
3957                         ssize_t buflen;
3958
3959                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3960                                 /* learn xattr size from osd_xattr_get if
3961                                    attribute has not been read yet */
3962                                 buflen = __osd_xattr_get(
3963                                     osd_dt_obj(dt)->oo_inode,
3964                                     &osd_oti_get(env)->oti_obj_dentry,
3965                                     name, NULL, 0);
3966                                 if (buflen < 0)
3967                                         buflen = 0;
3968                         } else {
3969                                 buflen = buf->lb_len;
3970                         }
3971
3972                         if (buflen > sb->s_blocksize) {
3973                                 credits += osd_calc_bkmap_credits(
3974                                     sb, NULL, 0, -1,
3975                                     (buflen + sb->s_blocksize - 1) >>
3976                                     sb->s_blocksize_bits);
3977                         }
3978                 }
3979                 /*
3980                  * xattr set may involve inode quota change, reserve credits for
3981                  * dquot_initialize()
3982                  */
3983                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3984         }
3985
3986         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3987
3988         return 0;
3989 }
3990
3991 /*
3992  * Concurrency: @dt is write locked.
3993  */
3994 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3995                          const struct lu_buf *buf, const char *name, int fl,
3996                          struct thandle *handle)
3997 {
3998         struct osd_object *obj = osd_dt_obj(dt);
3999         struct inode *inode = obj->oo_inode;
4000         struct osd_thread_info *info = osd_oti_get(env);
4001         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4002         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4003         int fs_flags = 0;
4004         int len;
4005         int rc;
4006         ENTRY;
4007
4008         LASSERT(handle);
4009         LASSERT(buf);
4010
4011         /* version set is not real XATTR */
4012         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4013                 dt_obj_version_t *version = buf->lb_buf;
4014
4015                 /* for version we are just using xattr API but change inode
4016                  * field instead */
4017                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4018
4019                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4020                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4021
4022                 LDISKFS_I(inode)->i_fs_version = *version;
4023                 /* Version is set after all inode operations are finished,
4024                  * so we should mark it dirty here */
4025                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4026
4027                 RETURN(0);
4028         }
4029
4030         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4031                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4032
4033         len = buf->lb_len;
4034         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4035         if (fl & LU_XATTR_REPLACE)
4036                 fs_flags |= XATTR_REPLACE;
4037
4038         if (fl & LU_XATTR_CREATE)
4039                 fs_flags |= XATTR_CREATE;
4040
4041         /* For the OST device with 256 bytes inode size by default,
4042          * the PFID EA will be stored together with LMA EA to avoid
4043          * performance trouble. Otherwise the PFID EA can be stored
4044          * independently. LU-8998 */
4045         if (strcmp(name, XATTR_NAME_FID) == 0 &&
4046             LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
4047                 struct dentry *dentry = &info->oti_obj_dentry;
4048                 struct filter_fid *ff;
4049                 struct ost_layout *ol;
4050                 int fl;
4051
4052                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4053
4054                 ff = buf->lb_buf;
4055                 ol = &ff->ff_layout;
4056                 /* Old client does not send stripe information, store
4057                  * the PFID EA on disk directly. */
4058                 if (buf->lb_len == sizeof(struct lu_fid) ||
4059                     ol->ol_stripe_size == 0) {
4060                         len = sizeof(struct lu_fid);
4061                         goto set;
4062                 }
4063
4064                 if (buf->lb_len != sizeof(*ff))
4065                         RETURN(-EINVAL);
4066
4067                 rc = osd_get_lma(info, inode, dentry, loa);
4068                 if (unlikely(rc == -ENODATA)) {
4069                         /* Usually for upgarding from old device */
4070                         lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4071                                         LMAC_FID_ON_OST, 0);
4072                         fl = XATTR_CREATE;
4073                 } else if (rc) {
4074                         RETURN(rc);
4075                 } else {
4076                         fl = XATTR_REPLACE;
4077                 }
4078
4079                 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4080                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4081                                              PFID_STRIPE_IDX_BITS;
4082                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4083                 lma->lma_compat |= LMAC_STRIPE_INFO;
4084                 if (ol->ol_comp_id != 0) {
4085                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4086                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4087                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4088                         lma->lma_compat |= LMAC_COMP_INFO;
4089                 }
4090
4091                 lustre_loa_swab(loa, false);
4092
4093                 /* Remove old PFID EA entry firstly. */
4094                 ll_vfs_dq_init(inode);
4095                 rc = inode->i_op->removexattr(dentry, name);
4096                 if (rc && rc != -ENODATA)
4097                         RETURN(rc);
4098
4099                 /* Store the PFID EA inside the LMA EA. */
4100                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4101                                      sizeof(*loa), fl);
4102
4103                 RETURN(rc);
4104         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4105                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4106                 if (rc)
4107                         RETURN(rc);
4108
4109                 lma->lma_incompat |= LMAI_STRIPED;
4110                 lustre_lma_swab(lma);
4111                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4112                                      sizeof(*lma), XATTR_REPLACE);
4113                 if (rc != 0)
4114                         RETURN(rc);
4115         }
4116
4117 set:
4118         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4119         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4120
4121         if (rc == 0 &&
4122             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4123              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4124                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4125
4126         return rc;
4127 }
4128
4129 /*
4130  * Concurrency: @dt is read locked.
4131  */
4132 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4133                           const struct lu_buf *buf)
4134 {
4135         struct osd_object      *obj    = osd_dt_obj(dt);
4136         struct inode           *inode  = obj->oo_inode;
4137         struct osd_thread_info *info   = osd_oti_get(env);
4138         struct dentry          *dentry = &info->oti_obj_dentry;
4139
4140         if (!dt_object_exists(dt))
4141                 return -ENOENT;
4142
4143         LASSERT(!dt_object_remote(dt));
4144         LASSERT(inode->i_op != NULL);
4145         LASSERT(inode->i_op->listxattr != NULL);
4146
4147         dentry->d_inode = inode;
4148         dentry->d_sb = inode->i_sb;
4149         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4150 }
4151
4152 static int osd_declare_xattr_del(const struct lu_env *env,
4153                                  struct dt_object *dt, const char *name,
4154                                  struct thandle *handle)
4155 {
4156         struct osd_thandle *oh;
4157         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4158
4159         LASSERT(!dt_object_remote(dt));
4160         LASSERT(handle != NULL);
4161
4162         oh = container_of0(handle, struct osd_thandle, ot_super);
4163         LASSERT(oh->ot_handle == NULL);
4164
4165         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4166                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4167         /*
4168          * xattr del may involve inode quota change, reserve credits for
4169          * dquot_initialize()
4170          */
4171         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4172
4173         return 0;
4174 }
4175
4176 /*
4177  * Concurrency: @dt is write locked.
4178  */
4179 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4180                          const char *name, struct thandle *handle)
4181 {
4182         struct osd_object      *obj    = osd_dt_obj(dt);
4183         struct inode           *inode  = obj->oo_inode;
4184         struct osd_thread_info *info   = osd_oti_get(env);
4185         struct dentry          *dentry = &info->oti_obj_dentry;
4186         int                     rc;
4187
4188         if (!dt_object_exists(dt))
4189                 return -ENOENT;
4190
4191         LASSERT(!dt_object_remote(dt));
4192         LASSERT(inode->i_op != NULL);
4193         LASSERT(inode->i_op->removexattr != NULL);
4194         LASSERT(handle != NULL);
4195
4196         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4197
4198         ll_vfs_dq_init(inode);
4199         dentry->d_inode = inode;
4200         dentry->d_sb = inode->i_sb;
4201         rc = inode->i_op->removexattr(dentry, name);
4202         if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4203                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4204
4205                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4206
4207                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4208                                  &info->oti_ost_attrs);
4209                 if (!rc) {
4210                         if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4211                                 rc = -ENODATA;
4212                                 goto out;
4213                         }
4214
4215                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4216                         lustre_lma_swab(lma);
4217                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4218                                              sizeof(*lma), XATTR_REPLACE);
4219                 }
4220         }
4221
4222 out:
4223         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4224
4225         if (rc == 0 &&
4226             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4227              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4228                 osd_oxc_del(obj, name);
4229
4230         return rc;
4231 }
4232
4233 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4234                            __u64 start, __u64 end)
4235 {
4236         struct osd_object       *obj    = osd_dt_obj(dt);
4237         struct inode            *inode  = obj->oo_inode;
4238         struct osd_thread_info  *info   = osd_oti_get(env);
4239         struct dentry           *dentry = &info->oti_obj_dentry;
4240         struct file             *file   = &info->oti_file;
4241         int                     rc;
4242
4243         ENTRY;
4244
4245         dentry->d_inode = inode;
4246         dentry->d_sb = inode->i_sb;
4247         file->f_path.dentry = dentry;
4248         file->f_mapping = inode->i_mapping;
4249         file->f_op = inode->i_fop;
4250         set_file_inode(file, inode);
4251
4252         rc = ll_vfs_fsync_range(file, start, end, 0);
4253
4254         RETURN(rc);
4255 }
4256
4257 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4258 {
4259         return 0;
4260 }
4261
4262 /*
4263  * Index operations.
4264  */
4265
4266 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4267                            const struct dt_index_features *feat)
4268 {
4269         struct iam_descr *descr;
4270
4271         if (osd_object_is_root(o))
4272                 return feat == &dt_directory_features;
4273
4274         LASSERT(o->oo_dir != NULL);
4275
4276         descr = o->oo_dir->od_container.ic_descr;
4277         if (feat == &dt_directory_features) {
4278                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4279                         return 1;
4280                 else
4281                         return 0;
4282         } else {
4283                 return
4284                         feat->dif_keysize_min <= descr->id_key_size &&
4285                         descr->id_key_size <= feat->dif_keysize_max &&
4286                         feat->dif_recsize_min <= descr->id_rec_size &&
4287                         descr->id_rec_size <= feat->dif_recsize_max &&
4288                         !(feat->dif_flags & (DT_IND_VARKEY |
4289                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4290                         ergo(feat->dif_flags & DT_IND_UPDATE,
4291                              1 /* XXX check that object (and file system) is
4292                                 * writable */);
4293         }
4294 }
4295
4296 static int osd_iam_container_init(const struct lu_env *env,
4297                                   struct osd_object *obj,
4298                                   struct osd_directory *dir)
4299 {
4300         struct iam_container *bag = &dir->od_container;
4301         int result;
4302
4303         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4304         if (result != 0)
4305                 return result;
4306
4307         result = iam_container_setup(bag);
4308         if (result == 0)
4309                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4310         else
4311                 iam_container_fini(bag);
4312
4313         return result;
4314 }
4315
4316
4317 /*
4318  * Concurrency: no external locking is necessary.
4319  */
4320 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4321                          const struct dt_index_features *feat)
4322 {
4323         int                      result;
4324         int                      skip_iam = 0;
4325         struct osd_object       *obj = osd_dt_obj(dt);
4326
4327         LINVRNT(osd_invariant(obj));
4328
4329         if (osd_object_is_root(obj)) {
4330                 dt->do_index_ops = &osd_index_ea_ops;
4331                 result = 0;
4332         } else if (feat == &dt_directory_features) {
4333                 dt->do_index_ops = &osd_index_ea_ops;
4334                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4335                         result = 0;
4336                 else
4337                         result = -ENOTDIR;
4338                 skip_iam = 1;
4339         } else if (unlikely(feat == &dt_otable_features)) {
4340                 dt->do_index_ops = &osd_otable_ops;
4341                 return 0;
4342         } else if (unlikely(feat == &dt_acct_features)) {
4343                 dt->do_index_ops = &osd_acct_index_ops;
4344                 result = 0;
4345                 skip_iam = 1;
4346         } else if (!osd_has_index(obj)) {
4347                 struct osd_directory *dir;
4348
4349                 OBD_ALLOC_PTR(dir);
4350                 if (dir != NULL) {
4351
4352                         spin_lock(&obj->oo_guard);
4353                         if (obj->oo_dir == NULL)
4354                                 obj->oo_dir = dir;
4355                         else
4356                                 /*
4357                                  * Concurrent thread allocated container data.
4358                                  */
4359                                 OBD_FREE_PTR(dir);
4360                         spin_unlock(&obj->oo_guard);
4361                         /*
4362                          * Now, that we have container data, serialize its
4363                          * initialization.
4364                          */
4365                         down_write(&obj->oo_ext_idx_sem);
4366                         /*
4367                          * recheck under lock.
4368                          */
4369                         if (!osd_has_index(obj))
4370                                 result = osd_iam_container_init(env, obj,
4371                                                                 obj->oo_dir);
4372                         else
4373                                 result = 0;
4374                         up_write(&obj->oo_ext_idx_sem);
4375                 } else {
4376                         result = -ENOMEM;
4377                 }
4378         } else {
4379                 result = 0;
4380         }
4381
4382         if (result == 0 && skip_iam == 0) {
4383                 if (!osd_iam_index_probe(env, obj, feat))
4384                         result = -ENOTDIR;
4385         }
4386         LINVRNT(osd_invariant(obj));
4387
4388         return result;
4389 }
4390
4391 static int osd_otable_it_attr_get(const struct lu_env *env,
4392                                  struct dt_object *dt,
4393                                  struct lu_attr *attr)
4394 {
4395         attr->la_valid = 0;
4396         return 0;
4397 }
4398
4399 static const struct dt_object_operations osd_obj_ops = {
4400         .do_read_lock         = osd_object_read_lock,
4401         .do_write_lock        = osd_object_write_lock,
4402         .do_read_unlock       = osd_object_read_unlock,
4403         .do_write_unlock      = osd_object_write_unlock,
4404         .do_write_locked      = osd_object_write_locked,
4405         .do_attr_get          = osd_attr_get,
4406         .do_declare_attr_set  = osd_declare_attr_set,
4407         .do_attr_set          = osd_attr_set,
4408         .do_ah_init           = osd_ah_init,
4409         .do_declare_create    = osd_declare_object_create,
4410         .do_create            = osd_object_create,
4411         .do_declare_destroy   = osd_declare_object_destroy,
4412         .do_destroy           = osd_object_destroy,
4413         .do_index_try         = osd_index_try,
4414         .do_declare_ref_add   = osd_declare_object_ref_add,
4415         .do_ref_add           = osd_object_ref_add,
4416         .do_declare_ref_del   = osd_declare_object_ref_del,
4417         .do_ref_del           = osd_object_ref_del,
4418         .do_xattr_get         = osd_xattr_get,
4419         .do_declare_xattr_set = osd_declare_xattr_set,
4420         .do_xattr_set         = osd_xattr_set,
4421         .do_declare_xattr_del = osd_declare_xattr_del,
4422         .do_xattr_del         = osd_xattr_del,
4423         .do_xattr_list        = osd_xattr_list,
4424         .do_object_sync       = osd_object_sync,
4425         .do_invalidate        = osd_invalidate,
4426 };
4427
4428 /**
4429  * dt_object_operations for interoperability mode
4430  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4431  */
4432 static const struct dt_object_operations osd_obj_ea_ops = {
4433         .do_read_lock         = osd_object_read_lock,
4434         .do_write_lock        = osd_object_write_lock,
4435         .do_read_unlock       = osd_object_read_unlock,
4436         .do_write_unlock      = osd_object_write_unlock,
4437         .do_write_locked      = osd_object_write_locked,
4438         .do_attr_get          = osd_attr_get,
4439         .do_declare_attr_set  = osd_declare_attr_set,
4440         .do_attr_set          = osd_attr_set,
4441         .do_ah_init           = osd_ah_init,
4442         .do_declare_create    = osd_declare_object_create,
4443         .do_create            = osd_object_ea_create,
4444         .do_declare_destroy   = osd_declare_object_destroy,
4445         .do_destroy           = osd_object_destroy,
4446         .do_index_try         = osd_index_try,
4447         .do_declare_ref_add   = osd_declare_object_ref_add,
4448         .do_ref_add           = osd_object_ref_add,
4449         .do_declare_ref_del   = osd_declare_object_ref_del,
4450         .do_ref_del           = osd_object_ref_del,
4451         .do_xattr_get         = osd_xattr_get,
4452         .do_declare_xattr_set = osd_declare_xattr_set,
4453         .do_xattr_set         = osd_xattr_set,
4454         .do_declare_xattr_del = osd_declare_xattr_del,
4455         .do_xattr_del         = osd_xattr_del,
4456         .do_xattr_list        = osd_xattr_list,
4457         .do_object_sync       = osd_object_sync,
4458         .do_invalidate        = osd_invalidate,
4459 };
4460
4461 static const struct dt_object_operations osd_obj_otable_it_ops = {
4462         .do_attr_get    = osd_otable_it_attr_get,
4463         .do_index_try   = osd_index_try,
4464 };
4465
4466 static int osd_index_declare_iam_delete(const struct lu_env *env,
4467                                         struct dt_object *dt,
4468                                         const struct dt_key *key,
4469                                         struct thandle *handle)
4470 {
4471         struct osd_thandle    *oh;
4472
4473         oh = container_of0(handle, struct osd_thandle, ot_super);
4474         LASSERT(oh->ot_handle == NULL);
4475
4476         /* Recycle  may cause additional three blocks to be changed. */
4477         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4478                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4479
4480         return 0;
4481 }
4482
4483 /**
4484  *      delete a (key, value) pair from index \a dt specified by \a key
4485  *
4486  *      \param  dt      osd index object
4487  *      \param  key     key for index
4488  *      \param  rec     record reference
4489  *      \param  handle  transaction handler
4490  *
4491  *      \retval  0  success
4492  *      \retval -ve   failure
4493  */
4494 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4495                                 const struct dt_key *key,
4496                                 struct thandle *handle)
4497 {
4498         struct osd_thread_info *oti = osd_oti_get(env);
4499         struct osd_object      *obj = osd_dt_obj(dt);
4500         struct osd_thandle     *oh;
4501         struct iam_path_descr  *ipd;
4502         struct iam_container   *bag = &obj->oo_dir->od_container;
4503         int                     rc;
4504         ENTRY;
4505
4506         if (!dt_object_exists(dt))
4507                 RETURN(-ENOENT);
4508
4509         LINVRNT(osd_invariant(obj));
4510         LASSERT(!dt_object_remote(dt));
4511         LASSERT(bag->ic_object == obj->oo_inode);
4512         LASSERT(handle != NULL);
4513
4514         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4515
4516         ipd = osd_idx_ipd_get(env, bag);
4517         if (unlikely(ipd == NULL))
4518                 RETURN(-ENOMEM);
4519
4520         oh = container_of0(handle, struct osd_thandle, ot_super);
4521         LASSERT(oh->ot_handle != NULL);
4522         LASSERT(oh->ot_handle->h_transaction != NULL);
4523
4524         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4525                 /* swab quota uid/gid provided by caller */
4526                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4527                 key = (const struct dt_key *)&oti->oti_quota_id;
4528         }
4529
4530         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4531         osd_ipd_put(env, bag, ipd);
4532         LINVRNT(osd_invariant(obj));
4533         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4534         RETURN(rc);
4535 }
4536
4537 static int osd_index_declare_ea_delete(const struct lu_env *env,
4538                                        struct dt_object *dt,
4539                                        const struct dt_key *key,
4540                                        struct thandle *handle)
4541 {
4542         struct osd_thandle *oh;
4543         struct inode       *inode;
4544         int                 rc, credits;
4545         ENTRY;
4546
4547         LASSERT(!dt_object_remote(dt));
4548         LASSERT(handle != NULL);
4549
4550         oh = container_of0(handle, struct osd_thandle, ot_super);
4551         LASSERT(oh->ot_handle == NULL);
4552
4553         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4554         if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4555                 /* '..' to a remote object has a local representative */
4556                 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4557                 /* to reset LMAI_REMOTE_PARENT */
4558                 credits += 1;
4559         }
4560         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4561
4562         inode = osd_dt_obj(dt)->oo_inode;
4563         if (inode == NULL)
4564                 RETURN(-ENOENT);
4565
4566         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4567                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4568                                    NULL, OSD_QID_BLK);
4569         RETURN(rc);
4570 }
4571
4572 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4573                                           struct dt_rec *fid)
4574 {
4575         struct osd_fid_pack *rec;
4576         int                  rc = -ENODATA;
4577
4578         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4579                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4580                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4581                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4582                         rc = -EINVAL;
4583         }
4584         return rc;
4585 }
4586
4587 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4588                           const struct lu_fid *fid)
4589 {
4590         struct seq_server_site  *ss = osd_seq_site(osd);
4591         ENTRY;
4592
4593         /* FID seqs not in FLDB, must be local seq */
4594         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4595                 RETURN(0);
4596
4597         /* If FLD is not being initialized yet, it only happens during the
4598          * initialization, likely during mgs initialization, and we assume
4599          * this is local FID. */
4600         if (ss == NULL || ss->ss_server_fld == NULL)
4601                 RETURN(0);
4602
4603         /* Only check the local FLDB here */
4604         if (osd_seq_exists(env, osd, fid_seq(fid)))
4605                 RETURN(0);
4606
4607         RETURN(1);
4608 }
4609
4610 /**
4611  * Index delete function for interoperability mode (b11826).
4612  * It will remove the directory entry added by osd_index_ea_insert().
4613  * This entry is needed to maintain name->fid mapping.
4614  *
4615  * \param key,  key i.e. file entry to be deleted
4616  *
4617  * \retval   0, on success
4618  * \retval -ve, on error
4619  */
4620 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4621                                const struct dt_key *key, struct thandle *handle)
4622 {
4623         struct osd_object          *obj = osd_dt_obj(dt);
4624         struct inode               *dir = obj->oo_inode;
4625         struct dentry              *dentry;
4626         struct osd_thandle         *oh;
4627         struct ldiskfs_dir_entry_2 *de = NULL;
4628         struct buffer_head         *bh;
4629         struct htree_lock          *hlock = NULL;
4630         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
4631         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4632         int                        rc;
4633         ENTRY;
4634
4635         if (!dt_object_exists(dt))
4636                 RETURN(-ENOENT);
4637
4638         LINVRNT(osd_invariant(obj));
4639         LASSERT(!dt_object_remote(dt));
4640         LASSERT(handle != NULL);
4641
4642         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4643
4644         oh = container_of(handle, struct osd_thandle, ot_super);
4645         LASSERT(oh->ot_handle != NULL);
4646         LASSERT(oh->ot_handle->h_transaction != NULL);
4647
4648         ll_vfs_dq_init(dir);
4649         dentry = osd_child_dentry_get(env, obj,
4650                                       (char *)key, strlen((char *)key));
4651
4652         if (obj->oo_hl_head != NULL) {
4653                 hlock = osd_oti_get(env)->oti_hlock;
4654                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4655                                    dir, LDISKFS_HLOCK_DEL);
4656         } else {
4657                 down_write(&obj->oo_ext_idx_sem);
4658         }
4659
4660         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4661         if (!IS_ERR(bh)) {
4662                 /* If this is not the ".." entry, it might be a remote DNE
4663                  * entry and  we need to check if the FID is for a remote
4664                  * MDT.  If the FID is  not in the directory entry (e.g.
4665                  * upgraded 1.8 filesystem without dirdata enabled) then
4666                  * we need to get the FID from the LMA. For a remote directory
4667                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4668                  * case.
4669                  *
4670                  * Delete the entry before the agent inode in order to
4671                  * simplify error handling.  At worst an error after deleting
4672                  * the entry first might leak the agent inode afterward. The
4673                  * reverse would need filesystem abort in case of error deleting
4674                  * the entry after the agent had been removed, or leave a
4675                  * dangling entry pointing at a random inode. */
4676                 if (strcmp((char *)key, dotdot) != 0) {
4677                         LASSERT(de != NULL);
4678                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4679                         if (rc == -ENODATA) {
4680                                 /* can't get FID, postpone to the end of the
4681                                  * transaction when iget() is safe */
4682                                 osd_schedule_agent_inode_removal(env, oh,
4683                                                 le32_to_cpu(de->inode));
4684                         } else if (rc == 0 &&
4685                                    unlikely(osd_remote_fid(env, osd, fid))) {
4686                                 osd_schedule_agent_inode_removal(env, oh,
4687                                                 le32_to_cpu(de->inode));
4688                         }
4689                 }
4690                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4691                 brelse(bh);
4692         } else {
4693                 rc = PTR_ERR(bh);
4694         }
4695         if (hlock != NULL)
4696                 ldiskfs_htree_unlock(hlock);
4697         else
4698                 up_write(&obj->oo_ext_idx_sem);
4699
4700         if (rc != 0)
4701                 GOTO(out, rc);
4702
4703         /* For inode on the remote MDT, .. will point to
4704          * /Agent directory, Check whether it needs to delete
4705          * from agent directory */
4706         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4707                 int ret;
4708
4709                 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4710                                                     obj, oh);
4711                 if (ret != 0)
4712                         /* Sigh, the entry has been deleted, and
4713                          * it is not easy to revert it back, so
4714                          * let's keep this error private, and let
4715                          * LFSCK fix it. XXX */
4716                         CERROR("%s: delete remote parent "DFID": rc = %d\n",
4717                                osd_name(osd), PFID(fid), ret);
4718         }
4719 out:
4720         LASSERT(osd_invariant(obj));
4721         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4722         RETURN(rc);
4723 }
4724
4725 /**
4726  *      Lookup index for \a key and copy record to \a rec.
4727  *
4728  *      \param  dt      osd index object
4729  *      \param  key     key for index
4730  *      \param  rec     record reference
4731  *
4732  *      \retval  +ve  success : exact mach
4733  *      \retval  0    return record with key not greater than \a key
4734  *      \retval -ve   failure
4735  */
4736 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4737                                 struct dt_rec *rec, const struct dt_key *key)
4738 {
4739         struct osd_object      *obj = osd_dt_obj(dt);
4740         struct iam_path_descr  *ipd;
4741         struct iam_container   *bag = &obj->oo_dir->od_container;
4742         struct osd_thread_info *oti = osd_oti_get(env);
4743         struct iam_iterator    *it = &oti->oti_idx_it;
4744         struct iam_rec         *iam_rec;
4745         int                     rc;
4746         ENTRY;
4747
4748         if (!dt_object_exists(dt))
4749                 RETURN(-ENOENT);
4750
4751         LASSERT(osd_invariant(obj));
4752         LASSERT(!dt_object_remote(dt));
4753         LASSERT(bag->ic_object == obj->oo_inode);
4754
4755         ipd = osd_idx_ipd_get(env, bag);
4756         if (IS_ERR(ipd))
4757                 RETURN(-ENOMEM);
4758
4759         /* got ipd now we can start iterator. */
4760         iam_it_init(it, bag, 0, ipd);
4761
4762         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4763                 /* swab quota uid/gid provided by caller */
4764                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4765                 key = (const struct dt_key *)&oti->oti_quota_id;
4766         }
4767
4768         rc = iam_it_get(it, (struct iam_key *)key);
4769         if (rc >= 0) {
4770                 if (S_ISDIR(obj->oo_inode->i_mode))
4771                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4772                 else
4773                         iam_rec = (struct iam_rec *) rec;
4774
4775                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4776
4777                 if (S_ISDIR(obj->oo_inode->i_mode))
4778                         osd_fid_unpack((struct lu_fid *) rec,
4779                                        (struct osd_fid_pack *)iam_rec);
4780                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4781                         osd_quota_unpack(obj, rec);
4782         }
4783
4784         iam_it_put(it);
4785         iam_it_fini(it);
4786         osd_ipd_put(env, bag, ipd);
4787
4788         LINVRNT(osd_invariant(obj));
4789
4790         RETURN(rc);
4791 }
4792
4793 static int osd_index_declare_iam_insert(const struct lu_env *env,
4794                                         struct dt_object *dt,
4795                                         const struct dt_rec *rec,
4796                                         const struct dt_key *key,
4797                                         struct thandle *handle)
4798 {
4799         struct osd_thandle *oh;
4800
4801         LASSERT(handle != NULL);
4802
4803         oh = container_of0(handle, struct osd_thandle, ot_super);
4804         LASSERT(oh->ot_handle == NULL);
4805
4806         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4807                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4808
4809         return 0;
4810 }
4811
4812 /**
4813  *      Inserts (key, value) pair in \a dt index object.
4814  *
4815  *      \param  dt      osd index object
4816  *      \param  key     key for index
4817  *      \param  rec     record reference
4818  *      \param  th      transaction handler
4819  *
4820  *      \retval  0  success
4821  *      \retval -ve failure
4822  */
4823 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4824                                 const struct dt_rec *rec,
4825                                 const struct dt_key *key, struct thandle *th,
4826                                 int ignore_quota)
4827 {
4828         struct osd_object     *obj = osd_dt_obj(dt);
4829         struct iam_path_descr *ipd;
4830         struct osd_thandle    *oh;
4831         struct iam_container  *bag;
4832         struct osd_thread_info *oti = osd_oti_get(env);
4833         struct iam_rec         *iam_rec;
4834         int                     rc;
4835         ENTRY;
4836
4837         if (!dt_object_exists(dt))
4838                 RETURN(-ENOENT);
4839
4840         LINVRNT(osd_invariant(obj));
4841         LASSERT(!dt_object_remote(dt));
4842
4843         bag = &obj->oo_dir->od_container;
4844         LASSERT(bag->ic_object == obj->oo_inode);
4845         LASSERT(th != NULL);
4846
4847         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4848
4849         ipd = osd_idx_ipd_get(env, bag);
4850         if (unlikely(ipd == NULL))
4851                 RETURN(-ENOMEM);
4852
4853         oh = container_of0(th, struct osd_thandle, ot_super);
4854         LASSERT(oh->ot_handle != NULL);
4855         LASSERT(oh->ot_handle->h_transaction != NULL);
4856         if (S_ISDIR(obj->oo_inode->i_mode)) {
4857                 iam_rec = (struct iam_rec *)oti->oti_ldp;
4858                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4859         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4860                 /* pack quota uid/gid */
4861                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4862                 key = (const struct dt_key *)&oti->oti_quota_id;
4863                 /* pack quota record */
4864                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4865                 iam_rec = (struct iam_rec *)rec;
4866         } else {
4867                 iam_rec = (struct iam_rec *)rec;
4868         }
4869
4870         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4871                         iam_rec, ipd);
4872         osd_ipd_put(env, bag, ipd);
4873         LINVRNT(osd_invariant(obj));
4874         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4875         RETURN(rc);
4876 }
4877
4878 /**
4879  * Calls ldiskfs_add_entry() to add directory entry
4880  * into the directory. This is required for
4881  * interoperability mode (b11826)
4882  *
4883  * \retval   0, on success
4884  * \retval -ve, on error
4885  */
4886 static int __osd_ea_add_rec(struct osd_thread_info *info,
4887                             struct osd_object *pobj, struct inode  *cinode,
4888                             const char *name, const struct lu_fid *fid,
4889                             struct htree_lock *hlock, struct thandle *th)
4890 {
4891         struct ldiskfs_dentry_param *ldp;
4892         struct dentry               *child;
4893         struct osd_thandle          *oth;
4894         int                          rc;
4895
4896         oth = container_of(th, struct osd_thandle, ot_super);
4897         LASSERT(oth->ot_handle != NULL);
4898         LASSERT(oth->ot_handle->h_transaction != NULL);
4899         LASSERT(pobj->oo_inode);
4900
4901         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4902         if (unlikely(pobj->oo_inode ==
4903                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4904                 ldp->edp_magic = 0;
4905         else
4906                 osd_get_ldiskfs_dirent_param(ldp, fid);
4907         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4908         child->d_fsdata = (void *)ldp;
4909         ll_vfs_dq_init(pobj->oo_inode);
4910         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4911                                    child, cinode, hlock);
4912         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4913                 struct ldiskfs_dir_entry_2      *de;
4914                 struct buffer_head              *bh;
4915                 int                              rc1;
4916
4917                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4918                                             NULL, hlock);
4919                 if (!IS_ERR(bh)) {
4920                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4921                                                                bh);
4922                         if (rc1 == 0) {
4923                                 if (S_ISDIR(cinode->i_mode))
4924                                         de->file_type = LDISKFS_DIRENT_LUFID |
4925                                                         LDISKFS_FT_REG_FILE;
4926                                 else
4927                                         de->file_type = LDISKFS_DIRENT_LUFID |
4928                                                         LDISKFS_FT_DIR;
4929                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4930                                                               NULL, bh);
4931                         }
4932                         brelse(bh);
4933                 }
4934         }
4935
4936         RETURN(rc);
4937 }
4938
4939 /**
4940  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4941  * into the directory.Also sets flags into osd object to
4942  * indicate dot and dotdot are created. This is required for
4943  * interoperability mode (b11826)
4944  *
4945  * \param dir   directory for dot and dotdot fixup.
4946  * \param obj   child object for linking
4947  *
4948  * \retval   0, on success
4949  * \retval -ve, on error
4950  */
4951 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4952                               struct osd_object *dir,
4953                               struct inode *parent_dir, const char *name,
4954                               const struct lu_fid *dot_fid,
4955                               const struct lu_fid *dot_dot_fid,
4956                               struct thandle *th)
4957 {
4958         struct inode                *inode = dir->oo_inode;
4959         struct osd_thandle          *oth;
4960         int result = 0;
4961
4962         oth = container_of(th, struct osd_thandle, ot_super);
4963         LASSERT(oth->ot_handle->h_transaction != NULL);
4964         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4965
4966         if (strcmp(name, dot) == 0) {
4967                 if (dir->oo_compat_dot_created) {
4968                         result = -EEXIST;
4969                 } else {
4970                         LASSERT(inode->i_ino == parent_dir->i_ino);
4971                         dir->oo_compat_dot_created = 1;
4972                         result = 0;
4973                 }
4974         } else if (strcmp(name, dotdot) == 0) {
4975                 if (!dir->oo_compat_dot_created)
4976                         return -EINVAL;
4977                 /* in case of rename, dotdot is already created */
4978                 if (dir->oo_compat_dotdot_created) {
4979                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4980                                                 dot_dot_fid, NULL, th);
4981                 }
4982
4983                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4984                         struct lu_fid tfid = *dot_dot_fid;
4985
4986                         tfid.f_oid--;
4987                         result = osd_add_dot_dotdot_internal(info,
4988                                         dir->oo_inode, parent_dir, dot_fid,
4989                                         &tfid, oth);
4990                 } else {
4991                         result = osd_add_dot_dotdot_internal(info,
4992                                         dir->oo_inode, parent_dir, dot_fid,
4993                                         dot_dot_fid, oth);
4994                 }
4995
4996                 if (result == 0)
4997                         dir->oo_compat_dotdot_created = 1;
4998         }
4999
5000         return result;
5001 }
5002
5003
5004 /**
5005  * It will call the appropriate osd_add* function and return the
5006  * value, return by respective functions.
5007  */
5008 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5009                           struct inode *cinode, const char *name,
5010                           const struct lu_fid *fid, struct thandle *th)
5011 {
5012         struct osd_thread_info *info   = osd_oti_get(env);
5013         struct htree_lock      *hlock;
5014         int                     rc;
5015
5016         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5017
5018         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
5019                                                    name[2] =='\0'))) {
5020                 if (hlock != NULL) {
5021                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5022                                            pobj->oo_inode, 0);
5023                 } else {
5024                         down_write(&pobj->oo_ext_idx_sem);
5025                 }
5026
5027                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5028                                         lu_object_fid(&pobj->oo_dt.do_lu),
5029                                         fid, th);
5030         } else {
5031                 if (hlock != NULL) {
5032                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5033                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5034                 } else {
5035                         down_write(&pobj->oo_ext_idx_sem);
5036                 }
5037
5038                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5039                         struct lu_fid *tfid = &info->oti_fid;
5040
5041                         *tfid = *fid;
5042                         tfid->f_ver = ~0;
5043                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5044                                               tfid, hlock, th);
5045                 } else {
5046                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5047                                               hlock, th);
5048                 }
5049         }
5050         if (hlock != NULL)
5051                 ldiskfs_htree_unlock(hlock);
5052         else
5053                 up_write(&pobj->oo_ext_idx_sem);
5054
5055         return rc;
5056 }
5057
5058 static int
5059 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5060                       struct osd_idmap_cache *oic)
5061 {
5062         struct osd_scrub *scrub = &dev->od_scrub;
5063         struct lu_fid *fid = &oic->oic_fid;
5064         struct osd_inode_id *id = &oic->oic_lid;
5065         struct inode *inode = NULL;
5066         int once  = 0;
5067         bool insert;
5068         int rc;
5069         ENTRY;
5070
5071         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5072                 RETURN(0);
5073
5074         if (scrub->os_pos_current > id->oii_ino)
5075                 RETURN(0);
5076
5077 again:
5078         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5079         if (rc == -ENOENT) {
5080                 __u32 gen = id->oii_gen;
5081
5082                 insert = true;
5083                 if (inode != NULL)
5084                         goto trigger;
5085
5086                 inode = osd_iget(oti, dev, id);
5087                 /* The inode has been removed (by race maybe). */
5088                 if (IS_ERR(inode)) {
5089                         rc = PTR_ERR(inode);
5090
5091                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5092                 }
5093
5094                 /* The OI mapping is lost. */
5095                 if (gen != OSD_OII_NOGEN)
5096                         goto trigger;
5097
5098                 iput(inode);
5099                 /* The inode may has been reused by others, we do not know,
5100                  * leave it to be handled by subsequent osd_fid_lookup(). */
5101                 RETURN(0);
5102         } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
5103                 RETURN(rc);
5104         } else {
5105                 insert = false;
5106         }
5107
5108 trigger:
5109         if (thread_is_running(&scrub->os_thread)) {
5110                 if (inode == NULL) {
5111                         inode = osd_iget(oti, dev, id);
5112                         /* The inode has been removed (by race maybe). */
5113                         if (IS_ERR(inode)) {
5114                                 rc = PTR_ERR(inode);
5115
5116                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5117                         }
5118                 }
5119
5120                 rc = osd_oii_insert(dev, oic, insert);
5121                 /* There is race condition between osd_oi_lookup and OI scrub.
5122                  * The OI scrub finished just after osd_oi_lookup() failure.
5123                  * Under such case, it is unnecessary to trigger OI scrub again,
5124                  * but try to call osd_oi_lookup() again. */
5125                 if (unlikely(rc == -EAGAIN))
5126                         goto again;
5127
5128                 if (!S_ISDIR(inode->i_mode))
5129                         rc = 0;
5130                 else
5131                         rc = osd_check_lmv(oti, dev, inode, oic);
5132
5133                 iput(inode);
5134                 RETURN(rc);
5135         }
5136
5137         if (!dev->od_noscrub && ++once == 1) {
5138                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5139                                      SS_CLEAR_FAILOUT);
5140                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5141                        "%.16s: trigger partial OI scrub for RPC inconsistency "
5142                        "checking FID "DFID": rc = %d\n",
5143                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
5144                        PFID(fid), rc);
5145                 if (rc == 0 || rc == -EALREADY)
5146                         goto again;
5147         }
5148
5149         if (inode != NULL)
5150                 iput(inode);
5151
5152         RETURN(rc);
5153 }
5154
5155 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5156                                struct osd_device *dev,
5157                                struct osd_idmap_cache *oic,
5158                                struct lu_fid *fid, __u32 ino)
5159 {
5160         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5161         struct inode *inode;
5162         int rc;
5163
5164         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5165         inode = osd_iget(oti, dev, &oic->oic_lid);
5166         if (IS_ERR(inode)) {
5167                 fid_zero(&oic->oic_fid);
5168                 return PTR_ERR(inode);
5169         }
5170
5171         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5172         iput(inode);
5173         if (rc != 0)
5174                 fid_zero(&oic->oic_fid);
5175         else
5176                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5177         return rc;
5178 }
5179
5180 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5181                       struct osd_inode_id *id, const struct lu_fid *fid)
5182 {
5183         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5184                id->oii_ino, id->oii_gen, info);
5185         info->oti_cache.oic_lid = *id;
5186         info->oti_cache.oic_fid = *fid;
5187         info->oti_cache.oic_dev = osd;
5188 }
5189
5190 /**
5191  * Get parent FID from the linkEA.
5192  *
5193  * For a directory which parent resides on remote MDT, to satisfy the
5194  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5195  * the other hand, to make the lookup(..) on the directory can return
5196  * the real parent FID, we append the real parent FID after its ".."
5197  * name entry in the /REMOTE_PARENT_DIR.
5198  *
5199  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5200  * backup. So after the restore, we cannot get the right parent FID from
5201  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5202  * we have stored the real parent FID in the directory object's linkEA,
5203  * we can parse the linkEA for the real parent FID.
5204  *
5205  * \param[in] env       pointer to the thread context
5206  * \param[in] obj       pointer to the object to be handled
5207  * \param[out]fid       pointer to the buffer to hold the parent FID
5208  *
5209  * \retval              0 for getting the real parent FID successfully
5210  * \retval              negative error number on failure
5211  */
5212 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5213                                     struct osd_object *obj,
5214                                     struct lu_fid *fid)
5215 {
5216         struct osd_thread_info  *oti    = osd_oti_get(env);
5217         struct lu_buf           *buf    = &oti->oti_big_buf;
5218         struct dentry           *dentry = &oti->oti_obj_dentry;
5219         struct inode            *inode  = obj->oo_inode;
5220         struct linkea_data       ldata  = { NULL };
5221         int                      rc;
5222         ENTRY;
5223
5224         fid_zero(fid);
5225         if (!S_ISDIR(inode->i_mode))
5226                 RETURN(-EIO);
5227
5228 again:
5229         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5230                              buf->lb_buf, buf->lb_len);
5231         if (rc == -ERANGE) {
5232                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5233                                      NULL, 0);
5234                 if (rc > 0) {
5235                         lu_buf_realloc(buf, rc);
5236                         if (buf->lb_buf == NULL)
5237                                 RETURN(-ENOMEM);
5238
5239                         goto again;
5240                 }
5241         }
5242
5243         if (unlikely(rc == 0))
5244                 RETURN(-ENODATA);
5245
5246         if (rc < 0)
5247                 RETURN(rc);
5248
5249         if (unlikely(buf->lb_buf == NULL)) {
5250                 lu_buf_realloc(buf, rc);
5251                 if (buf->lb_buf == NULL)
5252                         RETURN(-ENOMEM);
5253
5254                 goto again;
5255         }
5256
5257         ldata.ld_buf = buf;
5258         rc = linkea_init_with_rec(&ldata);
5259         if (!rc) {
5260                 linkea_first_entry(&ldata);
5261                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5262         }
5263
5264         RETURN(rc);
5265 }
5266
5267 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5268                                     struct inode *inode,
5269                                     const struct lu_fid *pfid,
5270                                     const char *name, const int namelen)
5271 {
5272         struct osd_thread_info *oti = osd_oti_get(env);
5273         struct lu_buf *buf = &oti->oti_big_buf;
5274         struct dentry *dentry = &oti->oti_obj_dentry;
5275         struct linkea_data ldata = { NULL };
5276         struct lu_name cname = { .ln_name = name,
5277                                  .ln_namelen = namelen };
5278         int rc;
5279         ENTRY;
5280
5281 again:
5282         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5283                              buf->lb_buf, buf->lb_len);
5284         if (rc == -ERANGE)
5285                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5286
5287         if (rc < 0)
5288                 RETURN(rc);
5289
5290         if (unlikely(rc == 0))
5291                 RETURN(-ENODATA);
5292
5293         if (buf->lb_len < rc) {
5294                 lu_buf_realloc(buf, rc);
5295                 if (buf->lb_buf == NULL)
5296                         RETURN(-ENOMEM);
5297
5298                 goto again;
5299         }
5300
5301         ldata.ld_buf = buf;
5302         rc = linkea_init_with_rec(&ldata);
5303         if (!rc)
5304                 rc = linkea_links_find(&ldata, &cname, pfid);
5305
5306         RETURN(rc);
5307 }
5308
5309 /**
5310  * Calls ->lookup() to find dentry. From dentry get inode and
5311  * read inode's ea to get fid. This is required for  interoperability
5312  * mode (b11826)
5313  *
5314  * \retval   0, on success
5315  * \retval -ve, on error
5316  */
5317 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5318                              struct dt_rec *rec, const struct dt_key *key)
5319 {
5320         struct inode                    *dir    = obj->oo_inode;
5321         struct dentry                   *dentry;
5322         struct ldiskfs_dir_entry_2      *de;
5323         struct buffer_head              *bh;
5324         struct lu_fid                   *fid = (struct lu_fid *) rec;
5325         struct htree_lock               *hlock = NULL;
5326         int                             ino;
5327         int                             rc;
5328         ENTRY;
5329
5330         LASSERT(dir->i_op != NULL);
5331         LASSERT(dir->i_op->lookup != NULL);
5332
5333         dentry = osd_child_dentry_get(env, obj,
5334                                       (char *)key, strlen((char *)key));
5335
5336         if (obj->oo_hl_head != NULL) {
5337                 hlock = osd_oti_get(env)->oti_hlock;
5338                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5339                                    dir, LDISKFS_HLOCK_LOOKUP);
5340         } else {
5341                 down_read(&obj->oo_ext_idx_sem);
5342         }
5343
5344         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5345         if (!IS_ERR(bh)) {
5346                 struct osd_thread_info *oti = osd_oti_get(env);
5347                 struct osd_inode_id *id = &oti->oti_id;
5348                 struct osd_idmap_cache *oic = &oti->oti_cache;
5349                 struct osd_device *dev = osd_obj2dev(obj);
5350
5351                 ino = le32_to_cpu(de->inode);
5352                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5353                         brelse(bh);
5354                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5355                         GOTO(out, rc);
5356                 }
5357
5358                 rc = osd_get_fid_from_dentry(de, rec);
5359
5360                 /* done with de, release bh */
5361                 brelse(bh);
5362                 if (rc != 0) {
5363                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
5364                                 const char *name = (const char *)key;
5365
5366                                 /* If the parent is on remote MDT, and there
5367                                  * is no FID-in-dirent, then we have to get
5368                                  * the parent FID from the linkEA.  */
5369                                 if (likely(strlen(name) == 2 &&
5370                                            name[0] == '.' && name[1] == '.'))
5371                                         rc = osd_get_pfid_from_linkea(env, obj,
5372                                                                       fid);
5373                         } else {
5374                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5375                         }
5376                 } else {
5377                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5378                 }
5379
5380                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5381                         fid_zero(&oic->oic_fid);
5382
5383                         GOTO(out, rc);
5384                 }
5385
5386                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5387                 rc = osd_consistency_check(oti, dev, oic);
5388                 if (rc != 0)
5389                         fid_zero(&oic->oic_fid);
5390         } else {
5391                 rc = PTR_ERR(bh);
5392         }
5393
5394         GOTO(out, rc);
5395
5396 out:
5397         if (hlock != NULL)
5398                 ldiskfs_htree_unlock(hlock);
5399         else
5400                 up_read(&obj->oo_ext_idx_sem);
5401         return rc;
5402 }
5403
5404 static int osd_index_declare_ea_insert(const struct lu_env *env,
5405                                        struct dt_object *dt,
5406                                        const struct dt_rec *rec,
5407                                        const struct dt_key *key,
5408                                        struct thandle *handle)
5409 {
5410         struct osd_thandle      *oh;
5411         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5412         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5413         const struct lu_fid     *fid    = rec1->rec_fid;
5414         int                      credits, rc = 0;
5415         struct osd_idmap_cache  *idc;
5416         ENTRY;
5417
5418         LASSERT(!dt_object_remote(dt));
5419         LASSERT(handle != NULL);
5420         LASSERT(fid != NULL);
5421         LASSERT(rec1->rec_type != 0);
5422
5423         oh = container_of0(handle, struct osd_thandle, ot_super);
5424         LASSERT(oh->ot_handle == NULL);
5425
5426         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5427
5428         /* we can't call iget() while a transactions is running
5429          * (this can lead to a deadlock), but we need to know
5430          * inum and object type. so we find this information at
5431          * declaration and cache in per-thread info */
5432         idc = osd_idc_find_or_init(env, osd, fid);
5433         if (IS_ERR(idc))
5434                 RETURN(PTR_ERR(idc));
5435         if (idc->oic_remote) {
5436                 /* a reference to remote inode is represented by an
5437                  * agent inode which we have to create */
5438                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5439                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5440         }
5441
5442         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5443
5444         if (osd_dt_obj(dt)->oo_inode != NULL) {
5445                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5446
5447                 /* We ignore block quota on meta pool (MDTs), so needn't
5448                  * calculate how many blocks will be consumed by this index
5449                  * insert */
5450                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5451                                            i_gid_read(inode),
5452                                            i_projid_read(inode), 0,
5453                                            oh, osd_dt_obj(dt), NULL,
5454                                            OSD_QID_BLK);
5455         }
5456
5457         RETURN(rc);
5458 }
5459
5460 /**
5461  * Index add function for interoperability mode (b11826).
5462  * It will add the directory entry.This entry is needed to
5463  * maintain name->fid mapping.
5464  *
5465  * \param key it is key i.e. file entry to be inserted
5466  * \param rec it is value of given key i.e. fid
5467  *
5468  * \retval   0, on success
5469  * \retval -ve, on error
5470  */
5471 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5472                                const struct dt_rec *rec,
5473                                const struct dt_key *key, struct thandle *th,
5474                                int ignore_quota)
5475 {
5476         struct osd_object       *obj = osd_dt_obj(dt);
5477         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5478         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5479         const struct lu_fid     *fid    = rec1->rec_fid;
5480         const char              *name = (const char *)key;
5481         struct osd_thread_info  *oti   = osd_oti_get(env);
5482         struct inode            *child_inode = NULL;
5483         struct osd_idmap_cache  *idc;
5484         int                     rc;
5485         ENTRY;
5486
5487         if (!dt_object_exists(dt))
5488                 RETURN(-ENOENT);
5489
5490         LASSERT(osd_invariant(obj));
5491         LASSERT(!dt_object_remote(dt));
5492         LASSERT(th != NULL);
5493
5494         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5495
5496         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5497
5498         idc = osd_idc_find(env, osd, fid);
5499         if (unlikely(idc == NULL)) {
5500                 /* this dt_insert() wasn't declared properly, so
5501                  * FID is missing in OI cache. we better do not
5502                  * lookup FID in FLDB/OI and don't risk to deadlock,
5503                  * but in some special cases (lfsck testing, etc)
5504                  * it's much simpler than fixing a caller */
5505                 CERROR("%s: "DFID" wasn't declared for insert\n",
5506                        osd_name(osd), PFID(fid));
5507                 dump_stack();
5508                 idc = osd_idc_find_or_init(env, osd, fid);
5509                 if (IS_ERR(idc))
5510                         RETURN(PTR_ERR(idc));
5511         }
5512
5513         if (idc->oic_remote) {
5514                 /* Insert remote entry */
5515                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5516                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
5517                         struct osd_thandle      *oh;
5518
5519                         /* If parent on remote MDT, we need put this object
5520                          * under AGENT */
5521                         oh = container_of(th, typeof(*oh), ot_super);
5522                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
5523                         if (rc != 0) {
5524                                 CERROR("%s: add "DFID" error: rc = %d\n",
5525                                        osd_name(osd),
5526                                        PFID(lu_object_fid(&dt->do_lu)), rc);
5527                                 RETURN(rc);
5528                         }
5529
5530                         child_inode = igrab(omm->omm_remote_parent->d_inode);
5531                 } else {
5532                         child_inode = osd_create_local_agent_inode(env, osd,
5533                                         obj, fid, rec1->rec_type & S_IFMT, th);
5534                         if (IS_ERR(child_inode))
5535                                 RETURN(PTR_ERR(child_inode));
5536                 }
5537         } else {
5538                 /* Insert local entry */
5539                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5540                         /* for a reason OI cache wasn't filled properly */
5541                         CERROR("%s: OIC for "DFID" isn't filled\n",
5542                                osd_name(osd), PFID(fid));
5543                         RETURN(-EINVAL);
5544                 }
5545                 child_inode = oti->oti_inode;
5546                 if (unlikely(child_inode == NULL)) {
5547                         struct ldiskfs_inode_info *lii;
5548                         OBD_ALLOC_PTR(lii);
5549                         if (lii == NULL)
5550                                 RETURN(-ENOMEM);
5551                         child_inode = oti->oti_inode = &lii->vfs_inode;
5552                 }
5553                 child_inode->i_sb = osd_sb(osd);
5554                 child_inode->i_ino = idc->oic_lid.oii_ino;
5555                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5556         }
5557
5558         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5559
5560         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5561                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5562
5563         if (child_inode && child_inode != oti->oti_inode)
5564                 iput(child_inode);
5565         LASSERT(osd_invariant(obj));
5566         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5567         RETURN(rc);
5568 }
5569
5570 /**
5571  *  Initialize osd Iterator for given osd index object.
5572  *
5573  *  \param  dt      osd index object
5574  */
5575
5576 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5577                                      struct dt_object *dt,
5578                                      __u32 unused)
5579 {
5580         struct osd_it_iam      *it;
5581         struct osd_object      *obj = osd_dt_obj(dt);
5582         struct lu_object       *lo  = &dt->do_lu;
5583         struct iam_path_descr  *ipd;
5584         struct iam_container   *bag = &obj->oo_dir->od_container;
5585
5586         if (!dt_object_exists(dt))
5587                 return ERR_PTR(-ENOENT);
5588
5589         OBD_ALLOC_PTR(it);
5590         if (it == NULL)
5591                 return ERR_PTR(-ENOMEM);
5592
5593         ipd = osd_it_ipd_get(env, bag);
5594         if (likely(ipd != NULL)) {
5595                 it->oi_obj = obj;
5596                 it->oi_ipd = ipd;
5597                 lu_object_get(lo);
5598                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5599                 return (struct dt_it *)it;
5600         } else {
5601                 OBD_FREE_PTR(it);
5602                 return ERR_PTR(-ENOMEM);
5603         }
5604 }
5605
5606 /**
5607  * free given Iterator.
5608  */
5609 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5610 {
5611         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5612         struct osd_object       *obj = it->oi_obj;
5613
5614         iam_it_fini(&it->oi_it);
5615         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5616         osd_object_put(env, obj);
5617         OBD_FREE_PTR(it);
5618 }
5619
5620 /**
5621  *  Move Iterator to record specified by \a key
5622  *
5623  *  \param  di      osd iterator
5624  *  \param  key     key for index
5625  *
5626  *  \retval +ve  di points to record with least key not larger than key
5627  *  \retval  0   di points to exact matched key
5628  *  \retval -ve  failure
5629  */
5630
5631 static int osd_it_iam_get(const struct lu_env *env,
5632                           struct dt_it *di, const struct dt_key *key)
5633 {
5634         struct osd_thread_info  *oti = osd_oti_get(env);
5635         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5636
5637         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5638                 /* swab quota uid/gid */
5639                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5640                 key = (struct dt_key *)&oti->oti_quota_id;
5641         }
5642
5643         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5644 }
5645
5646 /**
5647  *  Release Iterator
5648  *
5649  *  \param  di      osd iterator
5650  */
5651 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5652 {
5653         struct osd_it_iam *it = (struct osd_it_iam *)di;
5654
5655         iam_it_put(&it->oi_it);
5656 }
5657
5658 /**
5659  *  Move iterator by one record
5660  *
5661  *  \param  di      osd iterator
5662  *
5663  *  \retval +1   end of container reached
5664  *  \retval  0   success
5665  *  \retval -ve  failure
5666  */
5667
5668 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5669 {
5670         struct osd_it_iam *it = (struct osd_it_iam *)di;
5671
5672         return iam_it_next(&it->oi_it);
5673 }
5674
5675 /**
5676  * Return pointer to the key under iterator.
5677  */
5678
5679 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5680                                  const struct dt_it *di)
5681 {
5682         struct osd_thread_info *oti = osd_oti_get(env);
5683         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5684         struct osd_object      *obj = it->oi_obj;
5685         struct dt_key          *key;
5686
5687         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5688
5689         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5690                 /* swab quota uid/gid */
5691                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5692                 key = (struct dt_key *)&oti->oti_quota_id;
5693         }
5694
5695         return key;
5696 }
5697
5698 /**
5699  * Return size of key under iterator (in bytes)
5700  */
5701
5702 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5703 {
5704         struct osd_it_iam *it = (struct osd_it_iam *)di;
5705
5706         return iam_it_key_size(&it->oi_it);
5707 }
5708
5709 static inline void
5710 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5711 {
5712         /* check if file type is required */
5713         if (ent->lde_attrs & LUDA_TYPE) {
5714                 struct luda_type *lt;
5715                 int align = sizeof(*lt) - 1;
5716
5717                 len = (len + align) & ~align;
5718                 lt = (struct luda_type *)(ent->lde_name + len);
5719                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5720         }
5721
5722         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5723 }
5724
5725 /**
5726  * build lu direct from backend fs dirent.
5727  */
5728
5729 static inline void
5730 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5731                    char *name, __u16 namelen, __u16 type, __u32 attr)
5732 {
5733         ent->lde_attrs = attr | LUDA_FID;
5734         fid_cpu_to_le(&ent->lde_fid, fid);
5735
5736         ent->lde_hash = cpu_to_le64(offset);
5737         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5738
5739         strncpy(ent->lde_name, name, namelen);
5740         ent->lde_name[namelen] = '\0';
5741         ent->lde_namelen = cpu_to_le16(namelen);
5742
5743         /* append lustre attributes */
5744         osd_it_append_attrs(ent, namelen, type);
5745 }
5746
5747 /**
5748  * Return pointer to the record under iterator.
5749  */
5750 static int osd_it_iam_rec(const struct lu_env *env,
5751                           const struct dt_it *di,
5752                           struct dt_rec *dtrec, __u32 attr)
5753 {
5754         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5755         struct osd_thread_info *info = osd_oti_get(env);
5756         ENTRY;
5757
5758         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5759                 const struct osd_fid_pack *rec;
5760                 struct lu_fid             *fid = &info->oti_fid;
5761                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5762                 char                      *name;
5763                 int                        namelen;
5764                 __u64                      hash;
5765                 int                        rc;
5766
5767                 name = (char *)iam_it_key_get(&it->oi_it);
5768                 if (IS_ERR(name))
5769                         RETURN(PTR_ERR(name));
5770
5771                 namelen = iam_it_key_size(&it->oi_it);
5772
5773                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5774                 if (IS_ERR(rec))
5775                         RETURN(PTR_ERR(rec));
5776
5777                 rc = osd_fid_unpack(fid, rec);
5778                 if (rc)
5779                         RETURN(rc);
5780
5781                 hash = iam_it_store(&it->oi_it);
5782
5783                 /* IAM does not store object type in IAM index (dir) */
5784                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5785                                    0, LUDA_FID);
5786         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5787                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5788                            (struct iam_rec *)dtrec);
5789                 osd_quota_unpack(it->oi_obj, dtrec);
5790         } else {
5791                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5792                            (struct iam_rec *)dtrec);
5793         }
5794
5795         RETURN(0);
5796 }
5797
5798 /**
5799  * Returns cookie for current Iterator position.
5800  */
5801 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5802 {
5803         struct osd_it_iam *it = (struct osd_it_iam *)di;
5804
5805         return iam_it_store(&it->oi_it);
5806 }
5807
5808 /**
5809  * Restore iterator from cookie.
5810  *
5811  * \param  di      osd iterator
5812  * \param  hash    Iterator location cookie
5813  *
5814  * \retval +ve  di points to record with least key not larger than key.
5815  * \retval  0   di points to exact matched key
5816  * \retval -ve  failure
5817  */
5818
5819 static int osd_it_iam_load(const struct lu_env *env,
5820                            const struct dt_it *di, __u64 hash)
5821 {
5822         struct osd_it_iam *it = (struct osd_it_iam *)di;
5823
5824         return iam_it_load(&it->oi_it, hash);
5825 }
5826
5827 static const struct dt_index_operations osd_index_iam_ops = {
5828         .dio_lookup         = osd_index_iam_lookup,
5829         .dio_declare_insert = osd_index_declare_iam_insert,
5830         .dio_insert         = osd_index_iam_insert,
5831         .dio_declare_delete = osd_index_declare_iam_delete,
5832         .dio_delete         = osd_index_iam_delete,
5833         .dio_it     = {
5834                 .init     = osd_it_iam_init,
5835                 .fini     = osd_it_iam_fini,
5836                 .get      = osd_it_iam_get,
5837                 .put      = osd_it_iam_put,
5838                 .next     = osd_it_iam_next,
5839                 .key      = osd_it_iam_key,
5840                 .key_size = osd_it_iam_key_size,
5841                 .rec      = osd_it_iam_rec,
5842                 .store    = osd_it_iam_store,
5843                 .load     = osd_it_iam_load
5844         }
5845 };
5846
5847
5848 /**
5849  * Creates or initializes iterator context.
5850  *
5851  * \retval struct osd_it_ea, iterator structure on success
5852  *
5853  */
5854 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5855                                     struct dt_object *dt,
5856                                     __u32 attr)
5857 {
5858         struct osd_object       *obj  = osd_dt_obj(dt);
5859         struct osd_thread_info  *info = osd_oti_get(env);
5860         struct osd_it_ea        *oie;
5861         struct file             *file;
5862         struct lu_object        *lo   = &dt->do_lu;
5863         struct dentry           *obj_dentry;
5864         ENTRY;
5865
5866         if (!dt_object_exists(dt) || obj->oo_destroyed)
5867                 RETURN(ERR_PTR(-ENOENT));
5868
5869         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5870         if (oie == NULL)
5871                 RETURN(ERR_PTR(-ENOMEM));
5872         obj_dentry = &oie->oie_dentry;
5873
5874         obj_dentry->d_inode = obj->oo_inode;
5875         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5876         obj_dentry->d_name.hash = 0;
5877
5878         oie->oie_rd_dirent       = 0;
5879         oie->oie_it_dirent       = 0;
5880         oie->oie_dirent          = NULL;
5881         if (unlikely(!info->oti_it_ea_buf_used)) {
5882                 oie->oie_buf = info->oti_it_ea_buf;
5883                 info->oti_it_ea_buf_used = 1;
5884         } else {
5885                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5886                 if (oie->oie_buf == NULL)
5887                         RETURN(ERR_PTR(-ENOMEM));
5888         }
5889         oie->oie_obj             = obj;
5890
5891         file = &oie->oie_file;
5892
5893         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5894         if (attr & LUDA_64BITHASH)
5895                 file->f_mode    = FMODE_64BITHASH;
5896         else
5897                 file->f_mode    = FMODE_32BITHASH;
5898         file->f_path.dentry     = obj_dentry;
5899         file->f_mapping         = obj->oo_inode->i_mapping;
5900         file->f_op              = obj->oo_inode->i_fop;
5901         set_file_inode(file, obj->oo_inode);
5902
5903         lu_object_get(lo);
5904         RETURN((struct dt_it *) oie);
5905 }
5906
5907 /**
5908  * Destroy or finishes iterator context.
5909  *
5910  * \param di iterator structure to be destroyed
5911  */
5912 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5913 {
5914         struct osd_thread_info  *info = osd_oti_get(env);
5915         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
5916         struct osd_object       *obj    = oie->oie_obj;
5917         struct inode            *inode  = obj->oo_inode;
5918
5919         ENTRY;
5920         oie->oie_file.f_op->release(inode, &oie->oie_file);
5921         osd_object_put(env, obj);
5922         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5923                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5924         else
5925                 info->oti_it_ea_buf_used = 0;
5926         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5927         EXIT;
5928 }
5929
5930 /**
5931  * It position the iterator at given key, so that next lookup continues from
5932  * that key Or it is similar to dio_it->load() but based on a key,
5933  * rather than file position.
5934  *
5935  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5936  * to the beginning.
5937  *
5938  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5939  */
5940 static int osd_it_ea_get(const struct lu_env *env,
5941                          struct dt_it *di, const struct dt_key *key)
5942 {
5943         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
5944
5945         ENTRY;
5946         LASSERT(((const char *)key)[0] == '\0');
5947         it->oie_file.f_pos      = 0;
5948         it->oie_rd_dirent       = 0;
5949         it->oie_it_dirent       = 0;
5950         it->oie_dirent          = NULL;
5951
5952         RETURN(+1);
5953 }
5954
5955 /**
5956  * Does nothing
5957  */
5958 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5959 {
5960 }
5961
5962 struct osd_filldir_cbs {
5963 #ifdef HAVE_DIR_CONTEXT
5964         struct dir_context ctx;
5965 #endif
5966         struct osd_it_ea  *it;
5967 };
5968 /**
5969  * It is called internally by ->readdir(). It fills the
5970  * iterator's in-memory data structure with required
5971  * information i.e. name, namelen, rec_size etc.
5972  *
5973  * \param buf in which information to be filled in.
5974  * \param name name of the file in given dir
5975  *
5976  * \retval 0 on success
5977  * \retval 1 on buffer full
5978  */
5979 #ifdef HAVE_FILLDIR_USE_CTX
5980 static int osd_ldiskfs_filldir(struct dir_context *buf,
5981 #else
5982 static int osd_ldiskfs_filldir(void *buf,
5983 #endif
5984                                const char *name, int namelen,
5985                                loff_t offset, __u64 ino, unsigned d_type)
5986 {
5987         struct osd_it_ea        *it   =
5988                 ((struct osd_filldir_cbs *)buf)->it;
5989         struct osd_object       *obj  = it->oie_obj;
5990         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5991         struct lu_fid           *fid  = &ent->oied_fid;
5992         struct osd_fid_pack     *rec;
5993         ENTRY;
5994
5995         /* this should never happen */
5996         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5997                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5998                 RETURN(-EIO);
5999         }
6000
6001         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
6002             OSD_IT_EA_BUFSIZE)
6003                 RETURN(1);
6004
6005         /* "." is just the object itself. */
6006         if (namelen == 1 && name[0] == '.') {
6007                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6008         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6009                 rec = (struct osd_fid_pack*) (name + namelen + 1);
6010                 if (osd_fid_unpack(fid, rec) != 0)
6011                         fid_zero(fid);
6012         } else {
6013                 fid_zero(fid);
6014         }
6015         d_type &= ~LDISKFS_DIRENT_LUFID;
6016
6017         /* NOT export local root. */
6018         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6019                 ino = obj->oo_inode->i_ino;
6020                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6021         }
6022
6023         ent->oied_ino     = ino;
6024         ent->oied_off     = offset;
6025         ent->oied_namelen = namelen;
6026         ent->oied_type    = d_type;
6027
6028         memcpy(ent->oied_name, name, namelen);
6029
6030         it->oie_rd_dirent++;
6031         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6032         RETURN(0);
6033 }
6034
6035 /**
6036  * Calls ->readdir() to load a directory entry at a time
6037  * and stored it in iterator's in-memory data structure.
6038  *
6039  * \param di iterator's in memory structure
6040  *
6041  * \retval   0 on success
6042  * \retval -ve on error
6043  * \retval +1 reach the end of entry
6044  */
6045 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6046                                const struct dt_it *di)
6047 {
6048         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
6049         struct osd_object  *obj   = it->oie_obj;
6050         struct inode       *inode = obj->oo_inode;
6051         struct htree_lock  *hlock = NULL;
6052         struct file        *filp  = &it->oie_file;
6053         int                 rc = 0;
6054         struct osd_filldir_cbs buf = {
6055 #ifdef HAVE_DIR_CONTEXT
6056                 .ctx.actor = osd_ldiskfs_filldir,
6057 #endif
6058                 .it = it
6059         };
6060
6061         ENTRY;
6062         it->oie_dirent = it->oie_buf;
6063         it->oie_rd_dirent = 0;
6064
6065         if (obj->oo_hl_head != NULL) {
6066                 hlock = osd_oti_get(env)->oti_hlock;
6067                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6068                                    inode, LDISKFS_HLOCK_READDIR);
6069         } else {
6070                 down_read(&obj->oo_ext_idx_sem);
6071         }
6072
6073 #ifdef HAVE_DIR_CONTEXT
6074         buf.ctx.pos = filp->f_pos;
6075         rc = inode->i_fop->iterate(filp, &buf.ctx);
6076         filp->f_pos = buf.ctx.pos;
6077 #else
6078         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6079 #endif
6080
6081         if (hlock != NULL)
6082                 ldiskfs_htree_unlock(hlock);
6083         else
6084                 up_read(&obj->oo_ext_idx_sem);
6085
6086         if (it->oie_rd_dirent == 0) {
6087                 /*If it does not get any dirent, it means it has been reached
6088                  *to the end of the dir */
6089                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6090                 if (rc == 0)
6091                         rc = 1;
6092         } else {
6093                 it->oie_dirent = it->oie_buf;
6094                 it->oie_it_dirent = 1;
6095         }
6096
6097         RETURN(rc);
6098 }
6099
6100 /**
6101  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6102  * to load a directory entry at a time and stored it in
6103  * iterator's in-memory data structure.
6104  *
6105  * \param di iterator's in memory structure
6106  *
6107  * \retval +ve iterator reached to end
6108  * \retval   0 iterator not reached to end
6109  * \retval -ve on error
6110  */
6111 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6112 {
6113         struct osd_it_ea *it = (struct osd_it_ea *)di;
6114         int rc;
6115
6116         ENTRY;
6117
6118         if (it->oie_it_dirent < it->oie_rd_dirent) {
6119                 it->oie_dirent =
6120                         (void *) it->oie_dirent +
6121                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6122                                        it->oie_dirent->oied_namelen);
6123                 it->oie_it_dirent++;
6124                 RETURN(0);
6125         } else {
6126                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6127                         rc = +1;
6128                 else
6129                         rc = osd_ldiskfs_it_fill(env, di);
6130         }
6131
6132         RETURN(rc);
6133 }
6134
6135 /**
6136  * Returns the key at current position from iterator's in memory structure.
6137  *
6138  * \param di iterator's in memory structure
6139  *
6140  * \retval key i.e. struct dt_key on success
6141  */
6142 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6143                                     const struct dt_it *di)
6144 {
6145         struct osd_it_ea *it = (struct osd_it_ea *)di;
6146
6147         return (struct dt_key *)it->oie_dirent->oied_name;
6148 }
6149
6150 /**
6151  * Returns the key's size at current position from iterator's in memory structure.
6152  *
6153  * \param di iterator's in memory structure
6154  *
6155  * \retval key_size i.e. struct dt_key on success
6156  */
6157 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6158 {
6159         struct osd_it_ea *it = (struct osd_it_ea *)di;
6160
6161         return it->oie_dirent->oied_namelen;
6162 }
6163
6164 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6165 {
6166         if (LDISKFS_DIR_REC_LEN(de) >=
6167             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6168                 return true;
6169
6170         return false;
6171 }
6172
6173 static inline bool
6174 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6175                      unsigned blocksize, bool dotdot)
6176 {
6177         if (dotdot)
6178                 return osd_dotdot_has_space(de);
6179
6180         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6181             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6182                 return true;
6183
6184         return false;
6185 }
6186
6187 static int
6188 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6189                     handle_t *jh, struct dentry *dentry,
6190                     const struct lu_fid *fid, struct buffer_head *bh,
6191                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6192                     bool dotdot)
6193 {
6194         struct inode                *dir        = dentry->d_parent->d_inode;
6195         struct inode                *inode      = dentry->d_inode;
6196         struct osd_fid_pack         *rec;
6197         struct ldiskfs_dentry_param *ldp;
6198         int                          namelen    = dentry->d_name.len;
6199         int                          rc;
6200         struct osd_thread_info     *info        = osd_oti_get(env);
6201         ENTRY;
6202
6203         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6204                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6205                 RETURN(0);
6206
6207         /* There is enough space to hold the FID-in-dirent. */
6208         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6209                 rc = ldiskfs_journal_get_write_access(jh, bh);
6210                 if (rc != 0)
6211                         RETURN(rc);
6212
6213                 de->name[namelen] = 0;
6214                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6215                 rec->fp_len = sizeof(struct lu_fid) + 1;
6216                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6217                 de->file_type |= LDISKFS_DIRENT_LUFID;
6218                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6219
6220                 RETURN(rc);
6221         }
6222
6223         LASSERT(!dotdot);
6224
6225         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6226         if (rc != 0)
6227                 RETURN(rc);
6228
6229         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6230         osd_get_ldiskfs_dirent_param(ldp, fid);
6231         dentry->d_fsdata = (void *)ldp;
6232         ll_vfs_dq_init(dir);
6233         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6234         /* It is too bad, we cannot reinsert the name entry back.
6235          * That means we lose it! */
6236         if (rc != 0)
6237                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6238                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6239                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6240                        dir->i_ino, dir->i_generation, namelen,
6241                        dentry->d_name.name, PFID(fid), rc);
6242
6243         RETURN(rc);
6244 }
6245
6246 static int
6247 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6248                         struct osd_it_ea *it, struct lu_fid *fid,
6249                         struct osd_inode_id *id, __u32 *attr)
6250 {
6251         struct osd_thread_info     *info        = osd_oti_get(env);
6252         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6253         struct osd_device          *dev         = osd_obj2dev(obj);
6254         struct super_block         *sb          = osd_sb(dev);
6255         const char                 *devname     = osd_name(dev);
6256         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6257         struct inode               *dir         = obj->oo_inode;
6258         struct htree_lock          *hlock       = NULL;
6259         struct buffer_head         *bh          = NULL;
6260         handle_t                   *jh          = NULL;
6261         struct ldiskfs_dir_entry_2 *de;
6262         struct dentry              *dentry;
6263         struct inode               *inode;
6264         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6265         int                         credits;
6266         int                         rc;
6267         bool                        dotdot      = false;
6268         bool                        dirty       = false;
6269         ENTRY;
6270
6271         if (ent->oied_name[0] == '.') {
6272                 if (ent->oied_namelen == 1)
6273                         RETURN(0);
6274
6275                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6276                         dotdot = true;
6277         }
6278
6279         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6280         inode = osd_iget(info, dev, id);
6281         if (IS_ERR(inode)) {
6282                 rc = PTR_ERR(inode);
6283                 if (rc == -ENOENT || rc == -ESTALE) {
6284                         /* Maybe dangling name entry, or
6285                          * corrupted directory entry. */
6286                         *attr |= LUDA_UNKNOWN;
6287                         rc = 0;
6288                 } else {
6289                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6290                                "check_repair, dir = %lu/%u, name = %.*s, "
6291                                "ino = %llu, rc = %d\n",
6292                                devname, dir->i_ino, dir->i_generation,
6293                                ent->oied_namelen, ent->oied_name,
6294                                ent->oied_ino, rc);
6295                 }
6296
6297                 RETURN(rc);
6298         }
6299
6300         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6301                                            ent->oied_namelen);
6302         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6303         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6304                 lma = NULL;
6305         else if (rc != 0)
6306                 GOTO(out, rc);
6307
6308         /* We need to ensure that the name entry is still valid.
6309          * Because it may be removed or renamed by other already.
6310          *
6311          * The unlink or rename operation will start journal before PDO lock,
6312          * so to avoid deadlock, here we need to start journal handle before
6313          * related PDO lock also. But because we do not know whether there
6314          * will be something to be repaired before PDO lock, we just start
6315          * journal without conditions.
6316          *
6317          * We may need to remove the name entry firstly, then insert back.
6318          * One credit is for user quota file update.
6319          * One credit is for group quota file update.
6320          * Two credits are for dirty inode. */
6321         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6322                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6323
6324         if (dev->od_dirent_journal != 0) {
6325
6326 again:
6327                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6328                 if (IS_ERR(jh)) {
6329                         rc = PTR_ERR(jh);
6330                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6331                                "check_repair, dir = %lu/%u, credits = %d, "
6332                                "name = %.*s, ino = %llu: rc = %d\n",
6333                                devname, dir->i_ino, dir->i_generation, credits,
6334                                ent->oied_namelen, ent->oied_name,
6335                                ent->oied_ino, rc);
6336
6337                         GOTO(out_inode, rc);
6338                 }
6339
6340                 if (obj->oo_hl_head != NULL) {
6341                         hlock = osd_oti_get(env)->oti_hlock;
6342                         /* "0" means exclusive lock for the whole directory.
6343                          * We need to prevent others access such name entry
6344                          * during the delete + insert. Neither HLOCK_ADD nor
6345                          * HLOCK_DEL cannot guarantee the atomicity. */
6346                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6347                 } else {
6348                         down_write(&obj->oo_ext_idx_sem);
6349                 }
6350         } else {
6351                 if (obj->oo_hl_head != NULL) {
6352                         hlock = osd_oti_get(env)->oti_hlock;
6353                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6354                                            LDISKFS_HLOCK_LOOKUP);
6355                 } else {
6356                         down_read(&obj->oo_ext_idx_sem);
6357                 }
6358         }
6359
6360         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6361         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6362                 *attr |= LUDA_IGNORE;
6363
6364                 GOTO(out, rc = 0);
6365         }
6366
6367         /* For dotdot entry, if there is not enough space to hold the
6368          * FID-in-dirent, just keep them there. It only happens when the
6369          * device upgraded from 1.8 or restored from MDT file-level backup.
6370          * For the whole directory, only dotdot entry have no FID-in-dirent
6371          * and needs to get FID from LMA when readdir, it will not affect the
6372          * performance much. */
6373         if (dotdot && !osd_dotdot_has_space(de)) {
6374                 *attr |= LUDA_UNKNOWN;
6375
6376                 GOTO(out, rc = 0);
6377         }
6378
6379         if (lma != NULL) {
6380                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6381                         GOTO(out, rc = 0);
6382
6383                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6384                         struct lu_fid *tfid = &lma->lma_self_fid;
6385
6386                         if (likely(dotdot &&
6387                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6388                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6389                                 /* It must be REMOTE_PARENT_DIR and as the
6390                                  * 'dotdot' entry of remote directory */
6391                                 *attr |= LUDA_IGNORE;
6392                         } else {
6393                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6394                                        "parent directory, but got %.*s under "
6395                                        "dir = %lu/%u with the FID "DFID"\n",
6396                                        devname, ent->oied_namelen,
6397                                        ent->oied_name, dir->i_ino,
6398                                        dir->i_generation, PFID(tfid));
6399
6400                                 *attr |= LUDA_UNKNOWN;
6401                         }
6402
6403                         GOTO(out, rc = 0);
6404                 }
6405         }
6406
6407         if (!fid_is_zero(fid)) {
6408                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6409                                               ent->oied_namelen);
6410                 if (rc == -ENOENT ||
6411                     (rc == -ENODATA &&
6412                      !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6413                         /* linkEA does not recognize the dirent entry,
6414                          * it may because the dirent entry corruption
6415                          * and points to other's inode. */
6416                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6417                                "recognize the dirent, dir = %lu/%u, "
6418                                " name = %.*s, ino = %llu, "
6419                                DFID": rc = %d\n", devname, dir->i_ino,
6420                                dir->i_generation, ent->oied_namelen,
6421                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6422                         *attr |= LUDA_UNKNOWN;
6423
6424                         GOTO(out, rc = 0);
6425                 }
6426
6427                 if (rc && rc != -ENODATA) {
6428                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6429                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6430                                DFID": rc = %d\n", devname, dir->i_ino,
6431                                dir->i_generation, ent->oied_namelen,
6432                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6433                         *attr |= LUDA_UNKNOWN;
6434
6435                         GOTO(out, rc = 0);
6436                 }
6437         }
6438
6439         if (lma != NULL) {
6440                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6441                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6442                 if (*attr & LUDA_VERIFY_DRYRUN) {
6443                         *fid = lma->lma_self_fid;
6444                         *attr |= LUDA_REPAIR;
6445
6446                         GOTO(out, rc = 0);
6447                 }
6448
6449                 if (jh == NULL) {
6450                         brelse(bh);
6451                         dev->od_dirent_journal = 1;
6452                         if (hlock != NULL) {
6453                                 ldiskfs_htree_unlock(hlock);
6454                                 hlock = NULL;
6455                         } else {
6456                                 up_read(&obj->oo_ext_idx_sem);
6457                         }
6458
6459                         goto again;
6460                 }
6461
6462                 *fid = lma->lma_self_fid;
6463                 dirty = true;
6464                 /* Update or append the FID-in-dirent. */
6465                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6466                                          bh, de, hlock, dotdot);
6467                 if (rc == 0)
6468                         *attr |= LUDA_REPAIR;
6469                 else
6470                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6471                                "the dirent, dir = %lu/%u, name = %.*s, "
6472                                "ino = %llu, "DFID": rc = %d\n",
6473                                devname, dir->i_ino, dir->i_generation,
6474                                ent->oied_namelen, ent->oied_name,
6475                                ent->oied_ino, PFID(fid), rc);
6476         } else {
6477                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6478                 if (*attr & LUDA_VERIFY_DRYRUN) {
6479                         if (fid_is_sane(fid)) {
6480                                 *attr |= LUDA_REPAIR;
6481                         } else if (dev->od_index == 0) {
6482                                 lu_igif_build(fid, inode->i_ino,
6483                                               inode->i_generation);
6484                                 *attr |= LUDA_UPGRADE;
6485                         }
6486
6487                         GOTO(out, rc = 0);
6488                 }
6489
6490                 if (jh == NULL) {
6491                         brelse(bh);
6492                         dev->od_dirent_journal = 1;
6493                         if (hlock != NULL) {
6494                                 ldiskfs_htree_unlock(hlock);
6495                                 hlock = NULL;
6496                         } else {
6497                                 up_read(&obj->oo_ext_idx_sem);
6498                         }
6499
6500                         goto again;
6501                 }
6502
6503                 dirty = true;
6504                 if (unlikely(fid_is_sane(fid))) {
6505                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6506                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6507                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6508                         if (rc == 0)
6509                                 *attr |= LUDA_REPAIR;
6510                         else
6511                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6512                                        "update dirent, dir = %lu/%u, "
6513                                        "name = %.*s, ino = %llu, "
6514                                        DFID": rc = %d\n",
6515                                        devname, dir->i_ino, dir->i_generation,
6516                                        ent->oied_namelen, ent->oied_name,
6517                                        ent->oied_ino, PFID(fid), rc);
6518                 } else if (dev->od_index == 0) {
6519                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6520                         /* It is probably IGIF object. Only aappend the
6521                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6522                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6523                                                  bh, de, hlock, dotdot);
6524                         if (rc == 0)
6525                                 *attr |= LUDA_UPGRADE;
6526                         else
6527                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6528                                        "after the dirent, dir = %lu/%u, "
6529                                        "name = %.*s, ino = %llu, "
6530                                        DFID": rc = %d\n",
6531                                        devname, dir->i_ino, dir->i_generation,
6532                                        ent->oied_namelen, ent->oied_name,
6533                                        ent->oied_ino, PFID(fid), rc);
6534                 }
6535         }
6536
6537         GOTO(out, rc);
6538
6539 out:
6540         if (!IS_ERR(bh))
6541                 brelse(bh);
6542         if (hlock != NULL) {
6543                 ldiskfs_htree_unlock(hlock);
6544         } else {
6545                 if (dev->od_dirent_journal != 0)
6546                         up_write(&obj->oo_ext_idx_sem);
6547                 else
6548                         up_read(&obj->oo_ext_idx_sem);
6549         }
6550
6551         if (jh != NULL)
6552                 ldiskfs_journal_stop(jh);
6553
6554 out_inode:
6555         iput(inode);
6556         if (rc >= 0 && !dirty)
6557                 dev->od_dirent_journal = 0;
6558
6559         return rc;
6560 }
6561
6562 /**
6563  * Returns the value at current position from iterator's in memory structure.
6564  *
6565  * \param di struct osd_it_ea, iterator's in memory structure
6566  * \param attr attr requested for dirent.
6567  * \param lde lustre dirent
6568  *
6569  * \retval   0 no error and \param lde has correct lustre dirent.
6570  * \retval -ve on error
6571  */
6572 static inline int osd_it_ea_rec(const struct lu_env *env,
6573                                 const struct dt_it *di,
6574                                 struct dt_rec *dtrec, __u32 attr)
6575 {
6576         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6577         struct osd_object      *obj   = it->oie_obj;
6578         struct osd_device      *dev   = osd_obj2dev(obj);
6579         struct osd_thread_info *oti   = osd_oti_get(env);
6580         struct osd_inode_id    *id    = &oti->oti_id;
6581         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6582         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6583         __u32                   ino   = it->oie_dirent->oied_ino;
6584         int                     rc    = 0;
6585         ENTRY;
6586
6587         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6588
6589         if (attr & LUDA_VERIFY) {
6590                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6591                         attr |= LUDA_IGNORE;
6592                         /* If the parent is on remote MDT, and there
6593                          * is no FID-in-dirent, then we have to get
6594                          * the parent FID from the linkEA.  */
6595                         if (!fid_is_sane(fid) &&
6596                             it->oie_dirent->oied_namelen == 2 &&
6597                             it->oie_dirent->oied_name[0] == '.' &&
6598                             it->oie_dirent->oied_name[1] == '.')
6599                                 osd_get_pfid_from_linkea(env, obj, fid);
6600                 } else {
6601                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6602                                                      &attr);
6603                 }
6604
6605                 if (!fid_is_sane(fid))
6606                         attr |= LUDA_UNKNOWN;
6607         } else {
6608                 attr &= ~LU_DIRENT_ATTRS_MASK;
6609                 if (!fid_is_sane(fid)) {
6610                         bool is_dotdot = false;
6611                         if (it->oie_dirent->oied_namelen == 2 &&
6612                             it->oie_dirent->oied_name[0] == '.' &&
6613                             it->oie_dirent->oied_name[1] == '.')
6614                                 is_dotdot = true;
6615                         /* If the parent is on remote MDT, and there
6616                          * is no FID-in-dirent, then we have to get
6617                          * the parent FID from the linkEA.  */
6618                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6619                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6620                         } else {
6621                                 if (is_dotdot == false &&
6622                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6623                                         RETURN(-ENOENT);
6624
6625                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6626                         }
6627                 } else {
6628                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6629                 }
6630         }
6631
6632         /* Pack the entry anyway, at least the offset is right. */
6633         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6634                            it->oie_dirent->oied_name,
6635                            it->oie_dirent->oied_namelen,
6636                            it->oie_dirent->oied_type, attr);
6637
6638         if (rc < 0)
6639                 RETURN(rc);
6640
6641         if (osd_remote_fid(env, dev, fid))
6642                 RETURN(0);
6643
6644         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6645                 osd_add_oi_cache(oti, dev, id, fid);
6646
6647         RETURN(rc > 0 ? 0 : rc);
6648 }
6649
6650 /**
6651  * Returns the record size size at current position.
6652  *
6653  * This function will return record(lu_dirent) size in bytes.
6654  *
6655  * \param[in] env       execution environment
6656  * \param[in] di        iterator's in memory structure
6657  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6658  *                      calculate the lu_dirent size.
6659  *
6660  * \retval      record size(in bytes & in memory) of the current lu_dirent
6661  *              entry.
6662  */
6663 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6664                               __u32 attr)
6665 {
6666         struct osd_it_ea *it = (struct osd_it_ea *)di;
6667
6668         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6669 }
6670
6671 /**
6672  * Returns a cookie for current position of the iterator head, so that
6673  * user can use this cookie to load/start the iterator next time.
6674  *
6675  * \param di iterator's in memory structure
6676  *
6677  * \retval cookie for current position, on success
6678  */
6679 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6680 {
6681         struct osd_it_ea *it = (struct osd_it_ea *)di;
6682
6683         return it->oie_dirent->oied_off;
6684 }
6685
6686 /**
6687  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6688  * to load a directory entry at a time and stored it i inn,
6689  * in iterator's in-memory data structure.
6690  *
6691  * \param di struct osd_it_ea, iterator's in memory structure
6692  *
6693  * \retval +ve on success
6694  * \retval -ve on error
6695  */
6696 static int osd_it_ea_load(const struct lu_env *env,
6697                           const struct dt_it *di, __u64 hash)
6698 {
6699         struct osd_it_ea *it = (struct osd_it_ea *)di;
6700         int rc;
6701
6702         ENTRY;
6703         it->oie_file.f_pos = hash;
6704
6705         rc =  osd_ldiskfs_it_fill(env, di);
6706         if (rc > 0)
6707                 rc = -ENODATA;
6708
6709         if (rc == 0)
6710                 rc = +1;
6711
6712         RETURN(rc);
6713 }
6714
6715 /**
6716  * Index lookup function for interoperability mode (b11826).
6717  *
6718  * \param key,  key i.e. file name to be searched
6719  *
6720  * \retval +ve, on success
6721  * \retval -ve, on error
6722  */
6723 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6724                                struct dt_rec *rec, const struct dt_key *key)
6725 {
6726         struct osd_object *obj = osd_dt_obj(dt);
6727         int rc = 0;
6728
6729         ENTRY;
6730
6731         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6732         LINVRNT(osd_invariant(obj));
6733
6734         rc = osd_ea_lookup_rec(env, obj, rec, key);
6735         if (rc == 0)
6736                 rc = +1;
6737         RETURN(rc);
6738 }
6739
6740 /**
6741  * Index and Iterator operations for interoperability
6742  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6743  */
6744 static const struct dt_index_operations osd_index_ea_ops = {
6745         .dio_lookup         = osd_index_ea_lookup,
6746         .dio_declare_insert = osd_index_declare_ea_insert,
6747         .dio_insert         = osd_index_ea_insert,
6748         .dio_declare_delete = osd_index_declare_ea_delete,
6749         .dio_delete         = osd_index_ea_delete,
6750         .dio_it     = {
6751                 .init     = osd_it_ea_init,
6752                 .fini     = osd_it_ea_fini,
6753                 .get      = osd_it_ea_get,
6754                 .put      = osd_it_ea_put,
6755                 .next     = osd_it_ea_next,
6756                 .key      = osd_it_ea_key,
6757                 .key_size = osd_it_ea_key_size,
6758                 .rec      = osd_it_ea_rec,
6759                 .rec_size = osd_it_ea_rec_size,
6760                 .store    = osd_it_ea_store,
6761                 .load     = osd_it_ea_load
6762         }
6763 };
6764
6765 static void *osd_key_init(const struct lu_context *ctx,
6766                           struct lu_context_key *key)
6767 {
6768         struct osd_thread_info *info;
6769
6770         OBD_ALLOC_PTR(info);
6771         if (info == NULL)
6772                 return ERR_PTR(-ENOMEM);
6773
6774         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6775         if (info->oti_it_ea_buf == NULL)
6776                 goto out_free_info;
6777
6778         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6779
6780         info->oti_hlock = ldiskfs_htree_lock_alloc();
6781         if (info->oti_hlock == NULL)
6782                 goto out_free_ea;
6783
6784         return info;
6785
6786  out_free_ea:
6787         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6788  out_free_info:
6789         OBD_FREE_PTR(info);
6790         return ERR_PTR(-ENOMEM);
6791 }
6792
6793 static void osd_key_fini(const struct lu_context *ctx,
6794                          struct lu_context_key *key, void* data)
6795 {
6796         struct osd_thread_info *info = data;
6797         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6798         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6799
6800         if (info->oti_inode != NULL)
6801                 OBD_FREE_PTR(lli);
6802         if (info->oti_hlock != NULL)
6803                 ldiskfs_htree_lock_free(info->oti_hlock);
6804         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6805         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6806         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6807         lu_buf_free(&info->oti_big_buf);
6808         if (idc != NULL) {
6809                 LASSERT(info->oti_ins_cache_size > 0);
6810                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6811                 info->oti_ins_cache = NULL;
6812                 info->oti_ins_cache_size = 0;
6813         }
6814         OBD_FREE_PTR(info);
6815 }
6816
6817 static void osd_key_exit(const struct lu_context *ctx,
6818                          struct lu_context_key *key, void *data)
6819 {
6820         struct osd_thread_info *info = data;
6821
6822         LASSERT(info->oti_r_locks == 0);
6823         LASSERT(info->oti_w_locks == 0);
6824         LASSERT(info->oti_txns    == 0);
6825 }
6826
6827 /* type constructor/destructor: osd_type_init, osd_type_fini */
6828 LU_TYPE_INIT_FINI(osd, &osd_key);
6829
6830 struct lu_context_key osd_key = {
6831         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6832         .lct_init = osd_key_init,
6833         .lct_fini = osd_key_fini,
6834         .lct_exit = osd_key_exit
6835 };
6836
6837
6838 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6839                            const char *name, struct lu_device *next)
6840 {
6841         struct osd_device *osd = osd_dev(d);
6842
6843         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6844             >= sizeof(osd->od_svname))
6845                 return -E2BIG;
6846         return osd_procfs_init(osd, name);
6847 }
6848
6849 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6850 {
6851         struct seq_server_site  *ss = osd_seq_site(osd);
6852         int                     rc;
6853         ENTRY;
6854
6855         if (osd->od_is_ost || osd->od_cl_seq != NULL)
6856                 RETURN(0);
6857
6858         if (unlikely(ss == NULL))
6859                 RETURN(-ENODEV);
6860
6861         OBD_ALLOC_PTR(osd->od_cl_seq);
6862         if (osd->od_cl_seq == NULL)
6863                 RETURN(-ENOMEM);
6864
6865         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6866                              osd->od_svname, ss->ss_server_seq);
6867         if (rc != 0) {
6868                 OBD_FREE_PTR(osd->od_cl_seq);
6869                 osd->od_cl_seq = NULL;
6870                 RETURN(rc);
6871         }
6872
6873         if (ss->ss_node_id == 0) {
6874                 /* If the OSD on the sequence controller(MDT0), then allocate
6875                  * sequence here, otherwise allocate sequence after connected
6876                  * to MDT0 (see mdt_register_lwp_callback()). */
6877                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6878                                    &osd->od_cl_seq->lcs_space, env);
6879         }
6880
6881         RETURN(rc);
6882 }
6883
6884 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6885 {
6886         if (osd->od_cl_seq == NULL)
6887                 return;
6888
6889         seq_client_fini(osd->od_cl_seq);
6890         OBD_FREE_PTR(osd->od_cl_seq);
6891         osd->od_cl_seq = NULL;
6892 }
6893
6894 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6895 {
6896         ENTRY;
6897
6898         /* shutdown quota slave instance associated with the device */
6899         if (o->od_quota_slave != NULL) {
6900                 qsd_fini(env, o->od_quota_slave);
6901                 o->od_quota_slave = NULL;
6902         }
6903
6904         osd_fid_fini(env, o);
6905
6906         RETURN(0);
6907 }
6908
6909 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6910 {
6911         ENTRY;
6912
6913         if (o->od_mnt != NULL) {
6914                 shrink_dcache_sb(osd_sb(o));
6915                 osd_sync(env, &o->od_dt_dev);
6916
6917                 mntput(o->od_mnt);
6918                 o->od_mnt = NULL;
6919         }
6920
6921         EXIT;
6922 }
6923
6924 static int osd_mount(const struct lu_env *env,
6925                      struct osd_device *o, struct lustre_cfg *cfg)
6926 {
6927         const char              *name  = lustre_cfg_string(cfg, 0);
6928         const char              *dev  = lustre_cfg_string(cfg, 1);
6929         const char              *opts;
6930         unsigned long            page, s_flags, lmd_flags = 0;
6931         struct page             *__page;
6932         struct file_system_type *type;
6933         char                    *options = NULL;
6934         char                    *str;
6935         struct osd_thread_info  *info = osd_oti_get(env);
6936         struct lu_fid           *fid = &info->oti_fid;
6937         struct inode            *inode;
6938         int                      rc = 0, force_over_512tb = 0;
6939         ENTRY;
6940
6941         if (o->od_mnt != NULL)
6942                 RETURN(0);
6943
6944         if (strlen(dev) >= sizeof(o->od_mntdev))
6945                 RETURN(-E2BIG);
6946         strcpy(o->od_mntdev, dev);
6947
6948         str = lustre_cfg_string(cfg, 2);
6949         s_flags = simple_strtoul(str, NULL, 0);
6950         str = strstr(str, ":");
6951         if (str)
6952                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6953         opts = lustre_cfg_string(cfg, 3);
6954 #ifdef __BIG_ENDIAN
6955         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6956                 CERROR("%s: device %s extents feature is not guaranteed to "
6957                        "work on big-endian systems. Use \"bigendian_extents\" "
6958                        "mount option to override.\n", name, dev);
6959                 RETURN(-EINVAL);
6960         }
6961 #endif
6962 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6963         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6964                 CWARN("force_over_128tb option is deprecated. "
6965                       "Filesystems less than 512TB can be created without any "
6966                       "force options. Use force_over_512tb option for "
6967                       "filesystems greater than 512TB.\n");
6968         }
6969 #endif
6970 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6971         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6972                 CWARN("force_over_256tb option is deprecated. "
6973                       "Filesystems less than 512TB can be created without any "
6974                       "force options. Use force_over_512tb option for "
6975                       "filesystems greater than 512TB.\n");
6976         }
6977 #endif
6978
6979         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6980                 force_over_512tb = 1;
6981
6982         __page = alloc_page(GFP_KERNEL);
6983         if (__page == NULL)
6984                 GOTO(out, rc = -ENOMEM);
6985         page = (unsigned long)page_address(__page);
6986         options = (char *)page;
6987         *options = '\0';
6988         if (opts != NULL) {
6989                 /* strip out the options for back compatiblity */
6990                 static char *sout[] = {
6991                         "mballoc",
6992                         "iopen",
6993                         "noiopen",
6994                         "iopen_nopriv",
6995                         "extents",
6996                         "noextents",
6997                         /* strip out option we processed in osd */
6998                         "bigendian_extents",
6999                         "force_over_128tb",
7000                         "force_over_256tb",
7001                         "force_over_512tb",
7002                         NULL
7003                 };
7004                 strcat(options, opts);
7005                 for (rc = 0, str = options; sout[rc]; ) {
7006                         char *op = strstr(str, sout[rc]);
7007                         if (op == NULL) {
7008                                 rc++;
7009                                 str = options;
7010                                 continue;
7011                         }
7012                         if (op == options || *(op - 1) == ',') {
7013                                 str = op + strlen(sout[rc]);
7014                                 if (*str == ',' || *str == '\0') {
7015                                         *str == ',' ? str++ : str;
7016                                         memmove(op, str, strlen(str) + 1);
7017                                 }
7018                         }
7019                         for (str = op; *str != ',' && *str != '\0'; str++)
7020                                 ;
7021                 }
7022         } else {
7023                 strncat(options, "user_xattr,acl", 14);
7024         }
7025
7026         /* Glom up mount options */
7027         if (*options != '\0')
7028                 strcat(options, ",");
7029         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7030
7031         type = get_fs_type("ldiskfs");
7032         if (!type) {
7033                 CERROR("%s: cannot find ldiskfs module\n", name);
7034                 GOTO(out, rc = -ENODEV);
7035         }
7036
7037         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7038         module_put(type->owner);
7039
7040         if (IS_ERR(o->od_mnt)) {
7041                 rc = PTR_ERR(o->od_mnt);
7042                 o->od_mnt = NULL;
7043                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7044                 GOTO(out, rc);
7045         }
7046
7047         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7048                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7049                                  force_over_512tb == 0) {
7050                 CERROR("%s: device %s LDISKFS does not support filesystems "
7051                        "greater than 512TB and can cause data corruption. "
7052                        "Use \"force_over_512tb\" mount option to override.\n",
7053                        name, dev);
7054                 GOTO(out_mnt, rc = -EINVAL);
7055         }
7056
7057         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7058 #ifdef HAVE_DEV_SET_RDONLY
7059                 dev_set_rdonly(osd_sb(o)->s_bdev);
7060                 o->od_dt_dev.dd_rdonly = 1;
7061                 LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name);
7062 #else
7063                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7064                               name);
7065
7066                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7067 #endif
7068         } else {
7069 #ifdef HAVE_DEV_SET_RDONLY
7070                 if (dev_check_rdonly(osd_sb(o)->s_bdev)) {
7071                         CERROR("%s: underlying device %s is marked as "
7072                                "read-only. Setup failed\n", name, dev);
7073
7074                         GOTO(out_mnt, rc = -EROFS);
7075                 }
7076 #endif
7077         }
7078
7079         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7080                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7081                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7082                 GOTO(out_mnt, rc = -EINVAL);
7083         }
7084
7085 #ifdef LDISKFS_MOUNT_DIRDATA
7086         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7087                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7088                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7089         else if (!o->od_is_ost)
7090                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7091                       "enabling the dirdata feature. If you do not want to "
7092                       "downgrade to Lustre-1.x again, you can enable it via "
7093                       "'tune2fs -O dirdata device'\n", name, dev);
7094 #endif
7095         inode = osd_sb(o)->s_root->d_inode;
7096         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7097         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7098         if (rc != 0) {
7099                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7100                 GOTO(out_mnt, rc);
7101         }
7102
7103         if (lmd_flags & LMD_FLG_NOSCRUB)
7104                 o->od_noscrub = 1;
7105
7106         GOTO(out, rc = 0);
7107
7108 out_mnt:
7109         mntput(o->od_mnt);
7110         o->od_mnt = NULL;
7111
7112 out:
7113         if (__page)
7114                 __free_page(__page);
7115
7116         return rc;
7117 }
7118
7119 static struct lu_device *osd_device_fini(const struct lu_env *env,
7120                                          struct lu_device *d)
7121 {
7122         struct osd_device *o = osd_dev(d);
7123         ENTRY;
7124
7125         osd_shutdown(env, o);
7126         osd_procfs_fini(o);
7127         osd_scrub_cleanup(env, o);
7128         osd_obj_map_fini(o);
7129         osd_umount(env, o);
7130
7131         RETURN(NULL);
7132 }
7133
7134 static int osd_device_init0(const struct lu_env *env,
7135                             struct osd_device *o,
7136                             struct lustre_cfg *cfg)
7137 {
7138         struct lu_device        *l = osd2lu_dev(o);
7139         struct osd_thread_info *info;
7140         int                     rc;
7141         int                     cplen = 0;
7142
7143         /* if the module was re-loaded, env can loose its keys */
7144         rc = lu_env_refill((struct lu_env *) env);
7145         if (rc)
7146                 GOTO(out, rc);
7147         info = osd_oti_get(env);
7148         LASSERT(info);
7149
7150         l->ld_ops = &osd_lu_ops;
7151         o->od_dt_dev.dd_ops = &osd_dt_ops;
7152
7153         spin_lock_init(&o->od_osfs_lock);
7154         mutex_init(&o->od_otable_mutex);
7155         INIT_LIST_HEAD(&o->od_orphan_list);
7156
7157         o->od_read_cache = 1;
7158         o->od_writethrough_cache = 1;
7159         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7160
7161         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7162                         sizeof(o->od_svname));
7163         if (cplen >= sizeof(o->od_svname)) {
7164                 rc = -E2BIG;
7165                 GOTO(out, rc);
7166         }
7167
7168         o->od_index = -1; /* -1 means index is invalid */
7169         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7170         if (rc == LDD_F_SV_TYPE_OST)
7171                 o->od_is_ost = 1;
7172
7173         o->od_full_scrub_ratio = OFSR_DEFAULT;
7174         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7175         rc = osd_mount(env, o, cfg);
7176         if (rc != 0)
7177                 GOTO(out, rc);
7178
7179         rc = osd_obj_map_init(env, o);
7180         if (rc != 0)
7181                 GOTO(out_mnt, rc);
7182
7183         rc = lu_site_init(&o->od_site, l);
7184         if (rc != 0)
7185                 GOTO(out_compat, rc);
7186         o->od_site.ls_bottom_dev = l;
7187
7188         rc = lu_site_init_finish(&o->od_site);
7189         if (rc != 0)
7190                 GOTO(out_site, rc);
7191
7192         INIT_LIST_HEAD(&o->od_ios_list);
7193         /* setup scrub, including OI files initialization */
7194         rc = osd_scrub_setup(env, o);
7195         if (rc < 0)
7196                 GOTO(out_site, rc);
7197
7198         rc = osd_procfs_init(o, o->od_svname);
7199         if (rc != 0) {
7200                 CERROR("%s: can't initialize procfs: rc = %d\n",
7201                        o->od_svname, rc);
7202                 GOTO(out_scrub, rc);
7203         }
7204
7205         LASSERT(l->ld_site->ls_linkage.next != NULL);
7206         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7207
7208         /* initialize quota slave instance */
7209         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7210                                      o->od_proc_entry);
7211         if (IS_ERR(o->od_quota_slave)) {
7212                 rc = PTR_ERR(o->od_quota_slave);
7213                 o->od_quota_slave = NULL;
7214                 GOTO(out_procfs, rc);
7215         }
7216
7217         RETURN(0);
7218
7219 out_procfs:
7220         osd_procfs_fini(o);
7221 out_scrub:
7222         osd_scrub_cleanup(env, o);
7223 out_site:
7224         lu_site_fini(&o->od_site);
7225 out_compat:
7226         osd_obj_map_fini(o);
7227 out_mnt:
7228         osd_umount(env, o);
7229 out:
7230         return rc;
7231 }
7232
7233 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7234                                           struct lu_device_type *t,
7235                                           struct lustre_cfg *cfg)
7236 {
7237         struct osd_device *o;
7238         int                rc;
7239
7240         OBD_ALLOC_PTR(o);
7241         if (o == NULL)
7242                 return ERR_PTR(-ENOMEM);
7243
7244         rc = dt_device_init(&o->od_dt_dev, t);
7245         if (rc == 0) {
7246                 /* Because the ctx might be revived in dt_device_init,
7247                  * refill the env here */
7248                 lu_env_refill((struct lu_env *)env);
7249                 rc = osd_device_init0(env, o, cfg);
7250                 if (rc)
7251                         dt_device_fini(&o->od_dt_dev);
7252         }
7253
7254         if (unlikely(rc != 0))
7255                 OBD_FREE_PTR(o);
7256
7257         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7258 }
7259
7260 static struct lu_device *osd_device_free(const struct lu_env *env,
7261                                          struct lu_device *d)
7262 {
7263         struct osd_device *o = osd_dev(d);
7264         ENTRY;
7265
7266         /* XXX: make osd top device in order to release reference */
7267         d->ld_site->ls_top_dev = d;
7268         lu_site_purge(env, d->ld_site, -1);
7269         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7270                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7271                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7272         }
7273         lu_site_fini(&o->od_site);
7274         dt_device_fini(&o->od_dt_dev);
7275         OBD_FREE_PTR(o);
7276         RETURN(NULL);
7277 }
7278
7279 static int osd_process_config(const struct lu_env *env,
7280                               struct lu_device *d, struct lustre_cfg *cfg)
7281 {
7282         struct osd_device               *o = osd_dev(d);
7283         int                             rc;
7284         ENTRY;
7285
7286         switch (cfg->lcfg_command) {
7287         case LCFG_SETUP:
7288                 rc = osd_mount(env, o, cfg);
7289                 break;
7290         case LCFG_CLEANUP:
7291                 lu_dev_del_linkage(d->ld_site, d);
7292                 rc = osd_shutdown(env, o);
7293                 break;
7294         case LCFG_PARAM:
7295                 LASSERT(&o->od_dt_dev);
7296                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7297                                               cfg, &o->od_dt_dev);
7298                 if (rc > 0 || rc == -ENOSYS)
7299                         rc = class_process_proc_param(PARAM_OST,
7300                                                       lprocfs_osd_obd_vars,
7301                                                       cfg, &o->od_dt_dev);
7302                 break;
7303         default:
7304                 rc = -ENOSYS;
7305         }
7306
7307         RETURN(rc);
7308 }
7309
7310 static int osd_recovery_complete(const struct lu_env *env,
7311                                  struct lu_device *d)
7312 {
7313         struct osd_device       *osd = osd_dev(d);
7314         int                      rc = 0;
7315         ENTRY;
7316
7317         if (osd->od_quota_slave == NULL)
7318                 RETURN(0);
7319
7320         /* start qsd instance on recovery completion, this notifies the quota
7321          * slave code that we are about to process new requests now */
7322         rc = qsd_start(env, osd->od_quota_slave);
7323         RETURN(rc);
7324 }
7325
7326 /*
7327  * we use exports to track all osd users
7328  */
7329 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7330                            struct obd_device *obd, struct obd_uuid *cluuid,
7331                            struct obd_connect_data *data, void *localdata)
7332 {
7333         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7334         struct lustre_handle  conn;
7335         int                   rc;
7336         ENTRY;
7337
7338         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7339
7340         rc = class_connect(&conn, obd, cluuid);
7341         if (rc)
7342                 RETURN(rc);
7343
7344         *exp = class_conn2export(&conn);
7345
7346         spin_lock(&osd->od_osfs_lock);
7347         osd->od_connects++;
7348         spin_unlock(&osd->od_osfs_lock);
7349
7350         RETURN(0);
7351 }
7352
7353 /*
7354  * once last export (we don't count self-export) disappeared
7355  * osd can be released
7356  */
7357 static int osd_obd_disconnect(struct obd_export *exp)
7358 {
7359         struct obd_device *obd = exp->exp_obd;
7360         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7361         int                rc, release = 0;
7362         ENTRY;
7363
7364         /* Only disconnect the underlying layers on the final disconnect. */
7365         spin_lock(&osd->od_osfs_lock);
7366         osd->od_connects--;
7367         if (osd->od_connects == 0)
7368                 release = 1;
7369         spin_unlock(&osd->od_osfs_lock);
7370
7371         rc = class_disconnect(exp); /* bz 9811 */
7372
7373         if (rc == 0 && release)
7374                 class_manual_cleanup(obd);
7375         RETURN(rc);
7376 }
7377
7378 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7379                        struct lu_device *dev)
7380 {
7381         struct osd_device       *osd    = osd_dev(dev);
7382         struct lr_server_data   *lsd    =
7383                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7384         int                      result = 0;
7385         ENTRY;
7386
7387         if (osd->od_quota_slave != NULL) {
7388                 /* set up quota slave objects */
7389                 result = qsd_prepare(env, osd->od_quota_slave);
7390                 if (result != 0)
7391                         RETURN(result);
7392         }
7393
7394         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7395 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7396                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7397                         osd->od_index_in_idif = 1;
7398                 } else {
7399                         osd->od_index_in_idif = 0;
7400                         result = osd_register_proc_index_in_idif(osd);
7401                         if (result != 0)
7402                                 RETURN(result);
7403                 }
7404 #else
7405                 osd->od_index_in_idif = 1;
7406 #endif
7407         }
7408
7409         result = osd_fid_init(env, osd);
7410
7411         RETURN(result);
7412 }
7413
7414 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7415                          struct lu_fid *fid, struct md_op_data *op_data)
7416 {
7417         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7418
7419         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7420 }
7421
7422 static const struct lu_object_operations osd_lu_obj_ops = {
7423         .loo_object_init      = osd_object_init,
7424         .loo_object_delete    = osd_object_delete,
7425         .loo_object_release   = osd_object_release,
7426         .loo_object_free      = osd_object_free,
7427         .loo_object_print     = osd_object_print,
7428         .loo_object_invariant = osd_object_invariant
7429 };
7430
7431 const struct lu_device_operations osd_lu_ops = {
7432         .ldo_object_alloc      = osd_object_alloc,
7433         .ldo_process_config    = osd_process_config,
7434         .ldo_recovery_complete = osd_recovery_complete,
7435         .ldo_prepare           = osd_prepare,
7436 };
7437
7438 static const struct lu_device_type_operations osd_device_type_ops = {
7439         .ldto_init = osd_type_init,
7440         .ldto_fini = osd_type_fini,
7441
7442         .ldto_start = osd_type_start,
7443         .ldto_stop  = osd_type_stop,
7444
7445         .ldto_device_alloc = osd_device_alloc,
7446         .ldto_device_free  = osd_device_free,
7447
7448         .ldto_device_init    = osd_device_init,
7449         .ldto_device_fini    = osd_device_fini
7450 };
7451
7452 static struct lu_device_type osd_device_type = {
7453         .ldt_tags     = LU_DEVICE_DT,
7454         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7455         .ldt_ops      = &osd_device_type_ops,
7456         .ldt_ctx_tags = LCT_LOCAL,
7457 };
7458
7459 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7460 {
7461         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7462         struct super_block *sb = osd_sb(osd);
7463
7464         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7465 }
7466
7467 /*
7468  * lprocfs legacy support.
7469  */
7470 static struct obd_ops osd_obd_device_ops = {
7471         .o_owner = THIS_MODULE,
7472         .o_connect      = osd_obd_connect,
7473         .o_disconnect   = osd_obd_disconnect,
7474         .o_fid_alloc    = osd_fid_alloc,
7475         .o_health_check = osd_health_check,
7476 };
7477
7478 static int __init osd_init(void)
7479 {
7480         int rc;
7481
7482         LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7483 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7484         /* please, try to keep osd_thread_info smaller than a page */
7485         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7486 #endif
7487
7488         osd_oi_mod_init();
7489
7490         rc = lu_kmem_init(ldiskfs_caches);
7491         if (rc)
7492                 return rc;
7493
7494         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7495                                  lprocfs_osd_module_vars,
7496                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7497         if (rc)
7498                 lu_kmem_fini(ldiskfs_caches);
7499         return rc;
7500 }
7501
7502 static void __exit osd_exit(void)
7503 {
7504         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7505         lu_kmem_fini(ldiskfs_caches);
7506 }
7507
7508 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7509 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7510 MODULE_VERSION(LUSTRE_VERSION_STRING);
7511 MODULE_LICENSE("GPL");
7512
7513 module_init(osd_init);
7514 module_exit(osd_exit);