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
2387 static int osd_attr_get(const struct lu_env *env,
2388                         struct dt_object *dt,
2389                         struct lu_attr *attr)
2390 {
2391         struct osd_object *obj = osd_dt_obj(dt);
2392
2393         if (unlikely(!dt_object_exists(dt)))
2394                 return -ENOENT;
2395         if (unlikely(obj->oo_destroyed))
2396                 return -ENOENT;
2397
2398         LASSERT(!dt_object_remote(dt));
2399         LINVRNT(osd_invariant(obj));
2400
2401         spin_lock(&obj->oo_guard);
2402         osd_inode_getattr(env, obj->oo_inode, attr);
2403         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2404                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2405         spin_unlock(&obj->oo_guard);
2406
2407         return 0;
2408 }
2409
2410 static int osd_declare_attr_qid(const struct lu_env *env,
2411                                 struct osd_object *obj,
2412                                 struct osd_thandle *oh, long long bspace,
2413                                 qid_t old_id, qid_t new_id, bool enforce,
2414                                 unsigned type)
2415 {
2416         int rc;
2417         struct osd_thread_info *info = osd_oti_get(env);
2418         struct lquota_id_info  *qi = &info->oti_qi;
2419
2420         qi->lqi_type = type;
2421         /* inode accounting */
2422         qi->lqi_is_blk = false;
2423
2424         /* one more inode for the new id ... */
2425         qi->lqi_id.qid_uid = new_id;
2426         qi->lqi_space      = 1;
2427         /* Reserve credits for the new id */
2428         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2429         if (rc == -EDQUOT || rc == -EINPROGRESS)
2430                 rc = 0;
2431         if (rc)
2432                 RETURN(rc);
2433
2434         /* and one less inode for the current id */
2435         qi->lqi_id.qid_uid = old_id;
2436         qi->lqi_space      = -1;
2437         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2438         if (rc == -EDQUOT || rc == -EINPROGRESS)
2439                 rc = 0;
2440         if (rc)
2441                 RETURN(rc);
2442
2443         /* block accounting */
2444         qi->lqi_is_blk = true;
2445
2446         /* more blocks for the new uid ... */
2447         qi->lqi_id.qid_uid = new_id;
2448         qi->lqi_space      = bspace;
2449         /*
2450          * Credits for the new uid has been reserved, re-use "obj"
2451          * to save credit reservation.
2452          */
2453         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2454         if (rc == -EDQUOT || rc == -EINPROGRESS)
2455                 rc = 0;
2456         if (rc)
2457                 RETURN(rc);
2458
2459         /* and finally less blocks for the current uid */
2460         qi->lqi_id.qid_uid = old_id;
2461         qi->lqi_space      = -bspace;
2462         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2463         if (rc == -EDQUOT || rc == -EINPROGRESS)
2464                 rc = 0;
2465
2466         RETURN(rc);
2467 }
2468
2469 static int osd_declare_attr_set(const struct lu_env *env,
2470                                 struct dt_object *dt,
2471                                 const struct lu_attr *attr,
2472                                 struct thandle *handle)
2473 {
2474         struct osd_thandle     *oh;
2475         struct osd_object      *obj;
2476         qid_t                   uid;
2477         qid_t                   gid;
2478         long long               bspace;
2479         int                     rc = 0;
2480         bool                    enforce;
2481         ENTRY;
2482
2483         LASSERT(dt != NULL);
2484         LASSERT(handle != NULL);
2485
2486         obj = osd_dt_obj(dt);
2487         LASSERT(osd_invariant(obj));
2488
2489         oh = container_of0(handle, struct osd_thandle, ot_super);
2490         LASSERT(oh->ot_handle == NULL);
2491
2492         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2493                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2494
2495         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2496                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2497
2498         if (attr == NULL || obj->oo_inode == NULL)
2499                 RETURN(rc);
2500
2501         bspace   = obj->oo_inode->i_blocks;
2502         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2503         bspace   = toqb(bspace);
2504
2505         /* Changing ownership is always preformed by super user, it should not
2506          * fail with EDQUOT.
2507          *
2508          * We still need to call the osd_declare_qid() to calculate the journal
2509          * credits for updating quota accounting files and to trigger quota
2510          * space adjustment once the operation is completed.*/
2511         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2512                 /* USERQUOTA */
2513                 uid = i_uid_read(obj->oo_inode);
2514                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2515                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2516                                 attr->la_uid, enforce, USRQUOTA);
2517                 if (rc)
2518                         RETURN(rc);
2519
2520                 gid = i_gid_read(obj->oo_inode);
2521                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2522                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2523                                 i_gid_read(obj->oo_inode), attr->la_gid,
2524                                 enforce, GRPQUOTA);
2525                 if (rc)
2526                         RETURN(rc);
2527
2528         }
2529 #ifdef  HAVE_PROJECT_QUOTA
2530         if (attr->la_valid & LA_PROJID) {
2531                 __u32 projid = i_projid_read(obj->oo_inode);
2532                 enforce = (attr->la_valid & LA_PROJID) &&
2533                                         (attr->la_projid != projid);
2534                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2535                                 (qid_t)projid, (qid_t)attr->la_projid,
2536                                 enforce, PRJQUOTA);
2537                 if (rc)
2538                         RETURN(rc);
2539         }
2540 #endif
2541         RETURN(rc);
2542 }
2543
2544 static int osd_inode_setattr(const struct lu_env *env,
2545                              struct inode *inode, const struct lu_attr *attr)
2546 {
2547         __u64 bits = attr->la_valid;
2548
2549         /* Only allow set size for regular file */
2550         if (!S_ISREG(inode->i_mode))
2551                 bits &= ~(LA_SIZE | LA_BLOCKS);
2552
2553         if (bits == 0)
2554                 return 0;
2555
2556         if (bits & LA_ATIME)
2557                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2558         if (bits & LA_CTIME)
2559                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2560         if (bits & LA_MTIME)
2561                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2562         if (bits & LA_SIZE) {
2563                 spin_lock(&inode->i_lock);
2564                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2565                 i_size_write(inode, attr->la_size);
2566                 spin_unlock(&inode->i_lock);
2567         }
2568
2569         /* OSD should not change "i_blocks" which is used by quota.
2570          * "i_blocks" should be changed by ldiskfs only. */
2571         if (bits & LA_MODE)
2572                 inode->i_mode = (inode->i_mode & S_IFMT) |
2573                                 (attr->la_mode & ~S_IFMT);
2574         if (bits & LA_UID)
2575                 i_uid_write(inode, attr->la_uid);
2576         if (bits & LA_GID)
2577                 i_gid_write(inode, attr->la_gid);
2578         if (bits & LA_PROJID)
2579                 i_projid_write(inode, attr->la_projid);
2580         if (bits & LA_NLINK)
2581                 set_nlink(inode, attr->la_nlink);
2582         if (bits & LA_RDEV)
2583                 inode->i_rdev = attr->la_rdev;
2584
2585         if (bits & LA_FLAGS) {
2586                 /* always keep S_NOCMTIME */
2587                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2588                                  S_NOCMTIME;
2589         }
2590         return 0;
2591 }
2592
2593 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2594 {
2595         int rc;
2596
2597         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2598             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2599                 struct iattr    iattr;
2600
2601                 ll_vfs_dq_init(inode);
2602                 iattr.ia_valid = 0;
2603                 if (attr->la_valid & LA_UID)
2604                         iattr.ia_valid |= ATTR_UID;
2605                 if (attr->la_valid & LA_GID)
2606                         iattr.ia_valid |= ATTR_GID;
2607                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2608                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2609
2610                 rc = ll_vfs_dq_transfer(inode, &iattr);
2611                 if (rc) {
2612                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2613                                "enforcement enabled on the ldiskfs "
2614                                "filesystem?\n", inode->i_sb->s_id, rc);
2615                         return rc;
2616                 }
2617         }
2618
2619 #ifdef  HAVE_PROJECT_QUOTA
2620         /* Handle project id Transfer here properly */
2621         if (attr->la_valid & LA_PROJID && attr->la_projid !=
2622                                                 i_projid_read(inode)) {
2623                 rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid);
2624                 if (rc) {
2625                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2626                                "enforcement enabled on the ldiskfs "
2627                                "filesystem?\n", inode->i_sb->s_id, rc);
2628                         return rc;
2629                 }
2630         }
2631 #endif
2632         return 0;
2633 }
2634
2635 static int osd_attr_set(const struct lu_env *env,
2636                         struct dt_object *dt,
2637                         const struct lu_attr *attr,
2638                         struct thandle *handle)
2639 {
2640         struct osd_object *obj = osd_dt_obj(dt);
2641         struct inode      *inode;
2642         int rc;
2643
2644         if (!dt_object_exists(dt))
2645                 return -ENOENT;
2646
2647         LASSERT(handle != NULL);
2648         LASSERT(!dt_object_remote(dt));
2649         LASSERT(osd_invariant(obj));
2650
2651         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2652
2653         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2654                 struct osd_thread_info  *oti  = osd_oti_get(env);
2655                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2656                 struct lu_fid           *fid1 = &oti->oti_fid;
2657                 struct osd_inode_id     *id   = &oti->oti_id;
2658                 struct iam_path_descr   *ipd;
2659                 struct iam_container    *bag;
2660                 struct osd_thandle      *oh;
2661                 int                      rc;
2662
2663                 fid_cpu_to_be(fid1, fid0);
2664                 memset(id, 1, sizeof(*id));
2665                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2666                                   fid0)->oi_dir.od_container;
2667                 ipd = osd_idx_ipd_get(env, bag);
2668                 if (unlikely(ipd == NULL))
2669                         RETURN(-ENOMEM);
2670
2671                 oh = container_of0(handle, struct osd_thandle, ot_super);
2672                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2673                                 (const struct iam_rec *)id, ipd);
2674                 osd_ipd_put(env, bag, ipd);
2675                 return(rc > 0 ? 0 : rc);
2676         }
2677
2678         inode = obj->oo_inode;
2679
2680         rc = osd_quota_transfer(inode, attr);
2681         if (rc)
2682                 return rc;
2683
2684         spin_lock(&obj->oo_guard);
2685         rc = osd_inode_setattr(env, inode, attr);
2686         spin_unlock(&obj->oo_guard);
2687         if (rc != 0)
2688                 GOTO(out, rc);
2689
2690         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2691
2692         if (!(attr->la_valid & LA_FLAGS))
2693                 GOTO(out, rc);
2694
2695         /* Let's check if there are extra flags need to be set into LMA */
2696         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2697                 struct osd_thread_info *info = osd_oti_get(env);
2698                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2699
2700                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2701                                  &info->oti_ost_attrs);
2702                 if (rc)
2703                         GOTO(out, rc);
2704
2705                 lma->lma_incompat |=
2706                         lustre_to_lma_flags(attr->la_flags);
2707                 lustre_lma_swab(lma);
2708                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2709                                      lma, sizeof(*lma), XATTR_REPLACE);
2710                 if (rc != 0) {
2711                         struct osd_device *osd = osd_obj2dev(obj);
2712
2713                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2714                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2715                               lma->lma_incompat, rc);
2716                 } else {
2717                         obj->oo_lma_flags =
2718                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2719                 }
2720                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2721         }
2722 out:
2723         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2724
2725         return rc;
2726 }
2727
2728 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2729                                            struct osd_object *obj,
2730                                            const char *name, const int namelen)
2731 {
2732         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2733 }
2734
2735 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2736                       umode_t mode, struct dt_allocation_hint *hint,
2737                       struct thandle *th)
2738 {
2739         int result;
2740         struct osd_device  *osd = osd_obj2dev(obj);
2741         struct osd_thandle *oth;
2742         struct dt_object   *parent = NULL;
2743         struct inode       *inode;
2744
2745         LINVRNT(osd_invariant(obj));
2746         LASSERT(obj->oo_inode == NULL);
2747         LASSERT(obj->oo_hl_head == NULL);
2748
2749         if (S_ISDIR(mode) && ldiskfs_pdo) {
2750                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2751                 if (obj->oo_hl_head == NULL)
2752                         return -ENOMEM;
2753         }
2754
2755         oth = container_of(th, struct osd_thandle, ot_super);
2756         LASSERT(oth->ot_handle->h_transaction != NULL);
2757
2758         if (hint != NULL && hint->dah_parent != NULL &&
2759             !dt_object_remote(hint->dah_parent))
2760                 parent = hint->dah_parent;
2761
2762         inode = ldiskfs_create_inode(oth->ot_handle,
2763                                      parent ? osd_dt_obj(parent)->oo_inode :
2764                                               osd_sb(osd)->s_root->d_inode,
2765                                      mode);
2766         if (!IS_ERR(inode)) {
2767                 /* Do not update file c/mtime in ldiskfs. */
2768                 inode->i_flags |= S_NOCMTIME;
2769
2770                 /* For new created object, it must be consistent,
2771                  * and it is unnecessary to scrub against it. */
2772                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2773
2774                 obj->oo_inode = inode;
2775                 result = 0;
2776         } else {
2777                 if (obj->oo_hl_head != NULL) {
2778                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2779                         obj->oo_hl_head = NULL;
2780                 }
2781                 result = PTR_ERR(inode);
2782         }
2783         LINVRNT(osd_invariant(obj));
2784         return result;
2785 }
2786
2787 enum {
2788         OSD_NAME_LEN = 255
2789 };
2790
2791 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2792                      struct lu_attr *attr,
2793                      struct dt_allocation_hint *hint,
2794                      struct dt_object_format *dof,
2795                      struct thandle *th)
2796 {
2797         int result;
2798         struct osd_thandle *oth;
2799         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2800
2801         LASSERT(S_ISDIR(attr->la_mode));
2802
2803         oth = container_of(th, struct osd_thandle, ot_super);
2804         LASSERT(oth->ot_handle->h_transaction != NULL);
2805         result = osd_mkfile(info, obj, mode, hint, th);
2806
2807         return result;
2808 }
2809
2810 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2811                         struct lu_attr *attr,
2812                         struct dt_allocation_hint *hint,
2813                         struct dt_object_format *dof,
2814                         struct thandle *th)
2815 {
2816         int result;
2817         struct osd_thandle *oth;
2818         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2819
2820         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2821
2822         LASSERT(S_ISREG(attr->la_mode));
2823
2824         oth = container_of(th, struct osd_thandle, ot_super);
2825         LASSERT(oth->ot_handle->h_transaction != NULL);
2826
2827         result = osd_mkfile(info, obj, mode, hint, th);
2828         if (result == 0) {
2829                 LASSERT(obj->oo_inode != NULL);
2830                 if (feat->dif_flags & DT_IND_VARKEY)
2831                         result = iam_lvar_create(obj->oo_inode,
2832                                                  feat->dif_keysize_max,
2833                                                  feat->dif_ptrsize,
2834                                                  feat->dif_recsize_max,
2835                                                  oth->ot_handle);
2836                 else
2837                         result = iam_lfix_create(obj->oo_inode,
2838                                                  feat->dif_keysize_max,
2839                                                  feat->dif_ptrsize,
2840                                                  feat->dif_recsize_max,
2841                                                  oth->ot_handle);
2842
2843         }
2844         return result;
2845 }
2846
2847 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2848                      struct lu_attr *attr,
2849                      struct dt_allocation_hint *hint,
2850                      struct dt_object_format *dof,
2851                      struct thandle *th)
2852 {
2853         LASSERT(S_ISREG(attr->la_mode));
2854         return osd_mkfile(info, obj, (attr->la_mode &
2855                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2856 }
2857
2858 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2859                      struct lu_attr *attr,
2860                      struct dt_allocation_hint *hint,
2861                      struct dt_object_format *dof,
2862                      struct thandle *th)
2863 {
2864         LASSERT(S_ISLNK(attr->la_mode));
2865         return osd_mkfile(info, obj, (attr->la_mode &
2866                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2867 }
2868
2869 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2870                      struct lu_attr *attr,
2871                      struct dt_allocation_hint *hint,
2872                      struct dt_object_format *dof,
2873                      struct thandle *th)
2874 {
2875         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2876         int result;
2877
2878         LINVRNT(osd_invariant(obj));
2879         LASSERT(obj->oo_inode == NULL);
2880         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2881                 S_ISFIFO(mode) || S_ISSOCK(mode));
2882
2883         result = osd_mkfile(info, obj, mode, hint, th);
2884         if (result == 0) {
2885                 LASSERT(obj->oo_inode != NULL);
2886                 /*
2887                  * This inode should be marked dirty for i_rdev.  Currently
2888                  * that is done in the osd_attr_init().
2889                  */
2890                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2891                                    attr->la_rdev);
2892         }
2893         LINVRNT(osd_invariant(obj));
2894         return result;
2895 }
2896
2897 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2898                               struct lu_attr *,
2899                               struct dt_allocation_hint *hint,
2900                               struct dt_object_format *dof,
2901                               struct thandle *);
2902
2903 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2904 {
2905         osd_obj_type_f result;
2906
2907         switch (type) {
2908         case DFT_DIR:
2909                 result = osd_mkdir;
2910                 break;
2911         case DFT_REGULAR:
2912                 result = osd_mkreg;
2913                 break;
2914         case DFT_SYM:
2915                 result = osd_mksym;
2916                 break;
2917         case DFT_NODE:
2918                 result = osd_mknod;
2919                 break;
2920         case DFT_INDEX:
2921                 result = osd_mk_index;
2922                 break;
2923
2924         default:
2925                 LBUG();
2926                 break;
2927         }
2928         return result;
2929 }
2930
2931
2932 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2933                         struct dt_object *parent, struct dt_object *child,
2934                         umode_t child_mode)
2935 {
2936         LASSERT(ah);
2937
2938         ah->dah_parent = parent;
2939         ah->dah_mode = child_mode;
2940
2941         if (parent != NULL && !dt_object_remote(parent)) {
2942                 /* will help to find FID->ino at dt_insert("..") */
2943                 struct osd_object *pobj = osd_dt_obj(parent);
2944                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2945         }
2946 }
2947
2948 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2949                           struct lu_attr *attr, struct dt_object_format *dof)
2950 {
2951         struct inode   *inode = obj->oo_inode;
2952         __u64           valid = attr->la_valid;
2953         int             result;
2954
2955         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2956
2957         if (dof->dof_type != DFT_NODE)
2958                 attr->la_valid &= ~LA_RDEV;
2959         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2960                 attr->la_valid &= ~LA_ATIME;
2961         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2962                 attr->la_valid &= ~LA_CTIME;
2963         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2964                 attr->la_valid &= ~LA_MTIME;
2965
2966         result = osd_quota_transfer(inode, attr);
2967         if (result)
2968                 return;
2969
2970         if (attr->la_valid != 0) {
2971                 result = osd_inode_setattr(info->oti_env, inode, attr);
2972                 /*
2973                  * The osd_inode_setattr() should always succeed here.  The
2974                  * only error that could be returned is EDQUOT when we are
2975                  * trying to change the UID or GID of the inode. However, this
2976                  * should not happen since quota enforcement is no longer
2977                  * enabled on ldiskfs (lquota takes care of it).
2978                  */
2979                 LASSERTF(result == 0, "%d\n", result);
2980                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2981         }
2982
2983         attr->la_valid = valid;
2984 }
2985
2986 /**
2987  * Helper function for osd_object_create()
2988  *
2989  * \retval 0, on success
2990  */
2991 static int __osd_object_create(struct osd_thread_info *info,
2992                                struct osd_object *obj, struct lu_attr *attr,
2993                                struct dt_allocation_hint *hint,
2994                                struct dt_object_format *dof,
2995                                struct thandle *th)
2996 {
2997         int     result;
2998         __u32   umask;
2999
3000         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3001
3002         /* we drop umask so that permissions we pass are not affected */
3003         umask = current->fs->umask;
3004         current->fs->umask = 0;
3005
3006         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3007                                                   th);
3008         if (likely(obj->oo_inode != NULL)) {
3009                 LASSERT(obj->oo_inode->i_state & I_NEW);
3010
3011                 /* Unlock the inode before attr initialization to avoid
3012                  * unnecessary dqget operations. LU-6378 */
3013                 unlock_new_inode(obj->oo_inode);
3014         }
3015
3016         if (likely(result == 0)) {
3017                 osd_attr_init(info, obj, attr, dof);
3018                 osd_object_init0(obj);
3019         }
3020
3021         /* restore previous umask value */
3022         current->fs->umask = umask;
3023
3024         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3025
3026         return result;
3027 }
3028
3029 /**
3030  * Helper function for osd_object_create()
3031  *
3032  * \retval 0, on success
3033  */
3034 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3035                            const struct lu_fid *fid, struct thandle *th)
3036 {
3037         struct osd_thread_info *info = osd_oti_get(env);
3038         struct osd_inode_id    *id   = &info->oti_id;
3039         struct osd_device      *osd  = osd_obj2dev(obj);
3040         struct osd_thandle     *oh;
3041         int                     rc;
3042
3043         LASSERT(obj->oo_inode != NULL);
3044
3045         oh = container_of0(th, struct osd_thandle, ot_super);
3046         LASSERT(oh->ot_handle);
3047         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3048
3049         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3050         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3051                            OI_CHECK_FLD, NULL);
3052         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3053
3054         return rc;
3055 }
3056
3057 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3058                    u64 seq, struct lu_seq_range *range)
3059 {
3060         struct seq_server_site  *ss = osd_seq_site(osd);
3061
3062         if (fid_seq_is_idif(seq)) {
3063                 fld_range_set_ost(range);
3064                 range->lsr_index = idif_ost_idx(seq);
3065                 return 0;
3066         }
3067
3068         if (!fid_seq_in_fldb(seq)) {
3069                 fld_range_set_mdt(range);
3070                 if (ss != NULL)
3071                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3072                          * at all */
3073                         range->lsr_index = ss->ss_node_id;
3074                 return 0;
3075         }
3076
3077         LASSERT(ss != NULL);
3078         fld_range_set_any(range);
3079         /* OSD will only do local fld lookup */
3080         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3081 }
3082
3083 /*
3084  * Concurrency: no external locking is necessary.
3085  */
3086 static int osd_declare_object_create(const struct lu_env *env,
3087                                      struct dt_object *dt,
3088                                      struct lu_attr *attr,
3089                                      struct dt_allocation_hint *hint,
3090                                      struct dt_object_format *dof,
3091                                      struct thandle *handle)
3092 {
3093         struct osd_thandle      *oh;
3094         int                      rc;
3095         ENTRY;
3096
3097         LASSERT(handle != NULL);
3098
3099         oh = container_of0(handle, struct osd_thandle, ot_super);
3100         LASSERT(oh->ot_handle == NULL);
3101
3102         /* EA object consumes more credits than regular object: osd_mk_index
3103          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3104          * leaf_node, could involves the block, block bitmap, groups, GDT
3105          * change for each block, so add 4 * 2 credits in that case. */
3106         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3107                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3108                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3109         /* Reuse idle OI block may cause additional one OI block
3110          * to be changed. */
3111         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3112                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3113
3114         if (!attr)
3115                 RETURN(0);
3116
3117         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3118                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3119                                    false, NULL, false);
3120         if (rc != 0)
3121                 RETURN(rc);
3122
3123         /* will help to find FID->ino mapping at dt_insert() */
3124         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3125                                    osd_dt_obj(dt));
3126
3127         RETURN(rc);
3128 }
3129
3130 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3131                              struct lu_attr *attr,
3132                              struct dt_allocation_hint *hint,
3133                              struct dt_object_format *dof, struct thandle *th)
3134 {
3135         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3136         struct osd_object       *obj    = osd_dt_obj(dt);
3137         struct osd_thread_info  *info   = osd_oti_get(env);
3138         int result;
3139         ENTRY;
3140
3141         if (dt_object_exists(dt))
3142                 return -EEXIST;
3143
3144         LINVRNT(osd_invariant(obj));
3145         LASSERT(!dt_object_remote(dt));
3146         LASSERT(osd_write_locked(env, obj));
3147         LASSERT(th != NULL);
3148
3149         if (unlikely(fid_is_acct(fid)))
3150                 /* Quota files can't be created from the kernel any more,
3151                  * 'tune2fs -O quota' will take care of creating them */
3152                 RETURN(-EPERM);
3153
3154         result = __osd_object_create(info, obj, attr, hint, dof, th);
3155         if (result == 0) {
3156                 result = __osd_oi_insert(env, obj, fid, th);
3157                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3158                         obj->oo_dt.do_body_ops = &osd_body_ops;
3159         }
3160         LASSERT(ergo(result == 0,
3161                 dt_object_exists(dt) && !dt_object_remote(dt)));
3162
3163         LASSERT(osd_invariant(obj));
3164         RETURN(result);
3165 }
3166
3167 /**
3168  * Called to destroy on-disk representation of the object
3169  *
3170  * Concurrency: must be locked
3171  */
3172 static int osd_declare_object_destroy(const struct lu_env *env,
3173                                       struct dt_object *dt,
3174                                       struct thandle *th)
3175 {
3176         struct osd_object  *obj = osd_dt_obj(dt);
3177         struct inode       *inode = obj->oo_inode;
3178         struct osd_thandle *oh;
3179         int                 rc;
3180         ENTRY;
3181
3182         if (inode == NULL)
3183                 RETURN(-ENOENT);
3184
3185         oh = container_of0(th, struct osd_thandle, ot_super);
3186         LASSERT(oh->ot_handle == NULL);
3187
3188         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3189                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3190         /* Recycle idle OI leaf may cause additional three OI blocks
3191          * to be changed. */
3192         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3193                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3194                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3195         /* one less inode */
3196         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3197                                    i_projid_read(inode), -1, oh, obj, false,
3198                                    NULL, false);
3199         if (rc)
3200                 RETURN(rc);
3201         /* data to be truncated */
3202         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3203                                    i_projid_read(inode), 0, oh, obj, true,
3204                                    NULL, false);
3205         if (rc)
3206                 RETURN(rc);
3207
3208         /* will help to find FID->ino when this object is being
3209          * added to PENDING/ */
3210         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3211
3212         RETURN(rc);
3213 }
3214
3215 static int osd_object_destroy(const struct lu_env *env,
3216                               struct dt_object *dt,
3217                               struct thandle *th)
3218 {
3219         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
3220         struct osd_object      *obj = osd_dt_obj(dt);
3221         struct inode           *inode = obj->oo_inode;
3222         struct osd_device      *osd = osd_obj2dev(obj);
3223         struct osd_thandle     *oh;
3224         int                     result;
3225         ENTRY;
3226
3227         oh = container_of0(th, struct osd_thandle, ot_super);
3228         LASSERT(oh->ot_handle);
3229         LASSERT(inode);
3230         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3231
3232         if (unlikely(fid_is_acct(fid)))
3233                 RETURN(-EPERM);
3234
3235         if (S_ISDIR(inode->i_mode)) {
3236                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3237                         inode->i_nlink == 2);
3238                 /* it will check/delete the inode from remote parent,
3239                  * how to optimize it? unlink performance impaction XXX */
3240                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3241                 if (result != 0)
3242                         CERROR("%s: delete inode "DFID": rc = %d\n",
3243                                osd_name(osd), PFID(fid), result);
3244
3245                 spin_lock(&obj->oo_guard);
3246                 clear_nlink(inode);
3247                 spin_unlock(&obj->oo_guard);
3248                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3249         }
3250
3251         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3252
3253         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3254
3255         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3256                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3257                                        oh->ot_handle, OI_CHECK_FLD);
3258
3259         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3260         /* XXX: add to ext3 orphan list */
3261         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3262
3263         /* not needed in the cache anymore */
3264         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3265         obj->oo_destroyed = 1;
3266
3267         RETURN(0);
3268 }
3269
3270 /**
3271  * Put the fid into lustre_mdt_attrs, and then place the structure
3272  * inode's ea. This fid should not be altered during the life time
3273  * of the inode.
3274  *
3275  * \retval +ve, on success
3276  * \retval -ve, on error
3277  *
3278  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3279  */
3280 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3281                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3282 {
3283         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3284         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3285         int rc;
3286         ENTRY;
3287
3288         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3289                 RETURN(0);
3290
3291         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3292                 rc = -ENOMEM;
3293
3294         lustre_loa_init(loa, fid, compat, incompat);
3295         lustre_loa_swab(loa, false);
3296
3297         /* For the OST device with 256 bytes inode size by default,
3298          * the PFID EA will be stored together with LMA EA to avoid
3299          * performance trouble. Otherwise the PFID EA can be stored
3300          * independently. LU-8998 */
3301         if ((compat & LMAC_FID_ON_OST) &&
3302             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3303                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3304                                      sizeof(*loa), XATTR_CREATE);
3305         else
3306                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3307                                      sizeof(*lma), XATTR_CREATE);
3308         /* LMA may already exist, but we need to check that all the
3309          * desired compat/incompat flags have been added. */
3310         if (unlikely(rc == -EEXIST)) {
3311                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3312                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3313                 if (rc < 0)
3314                         RETURN(rc);
3315
3316                 if (rc < sizeof(*lma))
3317                         RETURN(-EINVAL);
3318
3319                 lustre_loa_swab(loa, true);
3320                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3321                     ((compat == 0 && incompat == 0) ||
3322                      (!(~lma->lma_compat & compat) &&
3323                       !(~lma->lma_incompat & incompat))))
3324                         RETURN(0);
3325
3326                 lma->lma_self_fid = *fid;
3327                 lma->lma_compat |= compat;
3328                 lma->lma_incompat |= incompat;
3329                 if (rc == sizeof(*lma)) {
3330                         lustre_lma_swab(lma);
3331                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3332                                              sizeof(*lma), XATTR_REPLACE);
3333                 } else {
3334                         lustre_loa_swab(loa, false);
3335                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3336                                              sizeof(*loa), XATTR_REPLACE);
3337                 }
3338         }
3339
3340         RETURN(rc);
3341 }
3342
3343 /**
3344  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3345  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3346  * To have compatilibility with 1.8 ldiskfs driver we need to have
3347  * magic number at start of fid data.
3348  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3349  * its inmemory API.
3350  */
3351 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3352                                          const struct lu_fid *fid)
3353 {
3354         if (!fid_is_namespace_visible(fid) ||
3355             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3356                 param->edp_magic = 0;
3357                 return;
3358         }
3359
3360         param->edp_magic = LDISKFS_LUFID_MAGIC;
3361         param->edp_len =  sizeof(struct lu_fid) + 1;
3362         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3363 }
3364
3365 /**
3366  * Try to read the fid from inode ea into dt_rec.
3367  *
3368  * \param fid object fid.
3369  *
3370  * \retval 0 on success
3371  */
3372 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3373                           __u32 ino, struct lu_fid *fid,
3374                           struct osd_inode_id *id)
3375 {
3376         struct osd_thread_info *info  = osd_oti_get(env);
3377         struct inode           *inode;
3378         ENTRY;
3379
3380         osd_id_gen(id, ino, OSD_OII_NOGEN);
3381         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3382         if (IS_ERR(inode))
3383                 RETURN(PTR_ERR(inode));
3384
3385         iput(inode);
3386         RETURN(0);
3387 }
3388
3389 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3390                                         struct inode *dir,
3391                                         struct inode *parent_dir,
3392                                         const struct lu_fid *dot_fid,
3393                                         const struct lu_fid *dot_dot_fid,
3394                                         struct osd_thandle *oth)
3395 {
3396         struct ldiskfs_dentry_param *dot_ldp;
3397         struct ldiskfs_dentry_param *dot_dot_ldp;
3398         __u32 saved_nlink = dir->i_nlink;
3399         int rc;
3400
3401         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3402         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3403
3404         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3405         dot_ldp->edp_magic = 0;
3406
3407         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3408                                     dir, dot_ldp, dot_dot_ldp);
3409         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3410          * the subseqent ref_add() will increase the dir->i_nlink
3411          * as 3. That is incorrect for new created directory.
3412          *
3413          * It looks like hack, because we want to make the OSD API
3414          * to be order-independent for new created directory object
3415          * between dt_insert(..) and ref_add() operations.
3416          *
3417          * Here, we only restore the in-RAM dir-inode's nlink attr,
3418          * becuase if the nlink attr is not 2, then there will be
3419          * ref_add() called following the dt_insert(..), such call
3420          * will make both the in-RAM and on-disk dir-inode's nlink
3421          * attr to be set as 2. LU-7447 */
3422         set_nlink(dir, saved_nlink);
3423         return rc;
3424 }
3425
3426 /**
3427  * Create an local agent inode for remote entry
3428  */
3429 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3430                                                   struct osd_device *osd,
3431                                                   struct osd_object *pobj,
3432                                                   const struct lu_fid *fid,
3433                                                   __u32 type,
3434                                                   struct thandle *th)
3435 {
3436         struct osd_thread_info  *info = osd_oti_get(env);
3437         struct inode            *local;
3438         struct osd_thandle      *oh;
3439         int                     rc;
3440         ENTRY;
3441
3442         LASSERT(th);
3443         oh = container_of(th, struct osd_thandle, ot_super);
3444         LASSERT(oh->ot_handle->h_transaction != NULL);
3445
3446         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3447         if (IS_ERR(local)) {
3448                 CERROR("%s: create local error %d\n", osd_name(osd),
3449                        (int)PTR_ERR(local));
3450                 RETURN(local);
3451         }
3452
3453         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3454          * correct gid on remote file, not agent here */
3455         local->i_gid = current_fsgid();
3456         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3457         unlock_new_inode(local);
3458
3459         /* Set special LMA flag for local agent inode */
3460         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3461         if (rc != 0) {
3462                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3463                        osd_name(osd), PFID(fid), rc);
3464                 RETURN(ERR_PTR(rc));
3465         }
3466
3467         if (!S_ISDIR(type))
3468                 RETURN(local);
3469
3470         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3471                                          lu_object_fid(&pobj->oo_dt.do_lu),
3472                                          fid, oh);
3473         if (rc != 0) {
3474                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3475                         osd_name(osd), PFID(fid), rc);
3476                 RETURN(ERR_PTR(rc));
3477         }
3478
3479         RETURN(local);
3480 }
3481
3482 /**
3483  * when direntry is deleted, we have to take care of possible agent inode
3484  * referenced by that. unfortunately we can't do this at that point:
3485  * iget() within a running transaction leads to deadlock and we better do
3486  * not call that every delete declaration to save performance. so we put
3487  * a potention agent inode on a list and process that once the transaction
3488  * is over. Notice it's not any worse in terms of real orphans as regular
3489  * object destroy doesn't put inodes on the on-disk orphan list. this should
3490  * be addressed separately
3491  */
3492 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3493                                             struct osd_thandle *oh,
3494                                             __u32 ino)
3495 {
3496         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3497         struct osd_obj_orphan *oor;
3498
3499         OBD_ALLOC_PTR(oor);
3500         if (oor == NULL)
3501                 return -ENOMEM;
3502
3503         oor->oor_ino = ino;
3504         oor->oor_env = (struct lu_env *)env;
3505         spin_lock(&osd->od_osfs_lock);
3506         list_add(&oor->oor_list, &osd->od_orphan_list);
3507         spin_unlock(&osd->od_osfs_lock);
3508
3509         oh->ot_remove_agents = 1;
3510
3511         return 0;
3512
3513 }
3514
3515 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3516                                                 struct osd_device *osd)
3517 {
3518         struct osd_thread_info *info = osd_oti_get(env);
3519         struct osd_obj_orphan *oor, *tmp;
3520         struct osd_inode_id id;
3521         struct list_head list;
3522         struct inode *inode;
3523         struct lu_fid fid;
3524         handle_t *jh;
3525         __u32 ino;
3526
3527         INIT_LIST_HEAD(&list);
3528
3529         spin_lock(&osd->od_osfs_lock);
3530         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3531                 if (oor->oor_env == env) {
3532                         list_del(&oor->oor_list);
3533                         list_add(&oor->oor_list, &list);
3534                 }
3535         }
3536         spin_unlock(&osd->od_osfs_lock);
3537
3538         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3539
3540                 ino = oor->oor_ino;
3541
3542                 list_del(&oor->oor_list);
3543                 OBD_FREE_PTR(oor);
3544
3545                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3546                 inode = osd_iget_fid(info, osd, &id, &fid);
3547                 if (IS_ERR(inode))
3548                         continue;
3549
3550                 if (!osd_remote_fid(env, osd, &fid)) {
3551                         iput(inode);
3552                         continue;
3553                 }
3554
3555                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3556                 clear_nlink(inode);
3557                 mark_inode_dirty(inode);
3558                 ldiskfs_journal_stop(jh);
3559                 iput(inode);
3560         }
3561
3562         return 0;
3563 }
3564
3565 /**
3566  * OSD layer object create function for interoperability mode (b11826).
3567  * This is mostly similar to osd_object_create(). Only difference being, fid is
3568  * inserted into inode ea here.
3569  *
3570  * \retval   0, on success
3571  * \retval -ve, on error
3572  */
3573 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3574                                 struct lu_attr *attr,
3575                                 struct dt_allocation_hint *hint,
3576                                 struct dt_object_format *dof,
3577                                 struct thandle *th)
3578 {
3579         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3580         struct osd_object       *obj    = osd_dt_obj(dt);
3581         struct osd_thread_info  *info   = osd_oti_get(env);
3582         int                      result, on_ost = 0;
3583
3584         ENTRY;
3585
3586         if (dt_object_exists(dt))
3587                 RETURN(-EEXIST);
3588
3589         LASSERT(osd_invariant(obj));
3590         LASSERT(!dt_object_remote(dt));
3591         LASSERT(osd_write_locked(env, obj));
3592         LASSERT(th != NULL);
3593
3594         if (unlikely(fid_is_acct(fid)))
3595                 /* Quota files can't be created from the kernel any more,
3596                  * 'tune2fs -O quota' will take care of creating them */
3597                 RETURN(-EPERM);
3598
3599         result = __osd_object_create(info, obj, attr, hint, dof, th);
3600         if (result == 0) {
3601                 if (fid_is_idif(fid) &&
3602                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3603                         struct lu_fid *tfid = &info->oti_fid;
3604                         struct ost_id *oi   = &info->oti_ostid;
3605
3606                         fid_to_ostid(fid, oi);
3607                         ostid_to_fid(tfid, oi, 0);
3608                         on_ost = 1;
3609                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3610                                                 LMAC_FID_ON_OST, 0);
3611                 } else {
3612                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3613                                                fid, OI_CHECK_FLD);
3614                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3615                                                 on_ost ? LMAC_FID_ON_OST : 0,
3616                                                 0);
3617                 }
3618                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3619                         obj->oo_dt.do_body_ops = &osd_body_ops;
3620         }
3621
3622         if (result == 0)
3623                 result = __osd_oi_insert(env, obj, fid, th);
3624
3625         /* a small optimization - dt_insert() isn't usually applied
3626          * to OST objects, so we don't need to cache OI mapping for
3627          * OST objects */
3628         if (result == 0 && on_ost == 0) {
3629                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3630                 result = osd_idc_find_and_init(env, osd, obj);
3631                 LASSERT(result == 0);
3632         }
3633
3634         LASSERT(ergo(result == 0,
3635                      dt_object_exists(dt) && !dt_object_remote(dt)));
3636         LINVRNT(osd_invariant(obj));
3637         RETURN(result);
3638 }
3639
3640 static int osd_declare_object_ref_add(const struct lu_env *env,
3641                                       struct dt_object *dt,
3642                                       struct thandle *handle)
3643 {
3644         struct osd_thandle       *oh;
3645
3646         /* it's possible that object doesn't exist yet */
3647         LASSERT(handle != NULL);
3648
3649         oh = container_of0(handle, struct osd_thandle, ot_super);
3650         LASSERT(oh->ot_handle == NULL);
3651
3652         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3653                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3654
3655         return 0;
3656 }
3657
3658 /*
3659  * Concurrency: @dt is write locked.
3660  */
3661 static int osd_object_ref_add(const struct lu_env *env,
3662                               struct dt_object *dt, struct thandle *th)
3663 {
3664         struct osd_object  *obj = osd_dt_obj(dt);
3665         struct inode       *inode = obj->oo_inode;
3666         struct osd_thandle *oh;
3667         int                 rc = 0;
3668
3669         if (!dt_object_exists(dt) || obj->oo_destroyed)
3670                 return -ENOENT;
3671
3672         LINVRNT(osd_invariant(obj));
3673         LASSERT(!dt_object_remote(dt));
3674         LASSERT(osd_write_locked(env, obj));
3675         LASSERT(th != NULL);
3676
3677         oh = container_of0(th, struct osd_thandle, ot_super);
3678         LASSERT(oh->ot_handle != NULL);
3679
3680         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3681
3682         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3683                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3684         /*
3685          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3686          * (65000) subdirectories by storing "1" in i_nlink if the link count
3687          * would otherwise overflow. Directory tranversal tools understand
3688          * that (st_nlink == 1) indicates that the filesystem dose not track
3689          * hard links count on the directory, and will not abort subdirectory
3690          * scanning early once (st_nlink - 2) subdirs have been found.
3691          *
3692          * This also has to properly handle the case of inodes with nlink == 0
3693          * in case they are being linked into the PENDING directory
3694          */
3695         spin_lock(&obj->oo_guard);
3696         if (unlikely(inode->i_nlink == 0))
3697                 /* inc_nlink from 0 may cause WARN_ON */
3698                 set_nlink(inode, 1);
3699         else {
3700                 ldiskfs_inc_count(oh->ot_handle, inode);
3701                 if (!S_ISDIR(inode->i_mode))
3702                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3703         }
3704         spin_unlock(&obj->oo_guard);
3705
3706         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3707         LINVRNT(osd_invariant(obj));
3708
3709         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3710
3711         return rc;
3712 }
3713
3714 static int osd_declare_object_ref_del(const struct lu_env *env,
3715                                       struct dt_object *dt,
3716                                       struct thandle *handle)
3717 {
3718         struct osd_thandle *oh;
3719
3720         if (!dt_object_exists(dt))
3721                 return -ENOENT;
3722
3723         LASSERT(!dt_object_remote(dt));
3724         LASSERT(handle != NULL);
3725
3726         oh = container_of0(handle, struct osd_thandle, ot_super);
3727         LASSERT(oh->ot_handle == NULL);
3728
3729         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3730                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3731
3732         return 0;
3733 }
3734
3735 /*
3736  * Concurrency: @dt is write locked.
3737  */
3738 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3739                               struct thandle *th)
3740 {
3741         struct osd_object       *obj = osd_dt_obj(dt);
3742         struct inode            *inode = obj->oo_inode;
3743         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3744         struct osd_thandle      *oh;
3745
3746         if (!dt_object_exists(dt))
3747                 return -ENOENT;
3748
3749         LINVRNT(osd_invariant(obj));
3750         LASSERT(!dt_object_remote(dt));
3751         LASSERT(osd_write_locked(env, obj));
3752         LASSERT(th != NULL);
3753
3754         oh = container_of0(th, struct osd_thandle, ot_super);
3755         LASSERT(oh->ot_handle != NULL);
3756
3757         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3758
3759         spin_lock(&obj->oo_guard);
3760         /* That can be result of upgrade from old Lustre version and
3761          * applied only to local files.  Just skip this ref_del call.
3762          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3763         if (inode->i_nlink == 0) {
3764                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3765                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3766                              ", maybe an upgraded file? (LU-3915)\n",
3767                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3768                 spin_unlock(&obj->oo_guard);
3769                 return 0;
3770         }
3771
3772         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3773                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3774
3775         ldiskfs_dec_count(oh->ot_handle, inode);
3776         spin_unlock(&obj->oo_guard);
3777
3778         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3779         LINVRNT(osd_invariant(obj));
3780
3781         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3782
3783         return 0;
3784 }
3785
3786 /*
3787  * Concurrency: @dt is read locked.
3788  */
3789 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3790                          struct lu_buf *buf, const char *name)
3791 {
3792         struct osd_object      *obj    = osd_dt_obj(dt);
3793         struct inode           *inode  = obj->oo_inode;
3794         struct osd_thread_info *info   = osd_oti_get(env);
3795         struct dentry          *dentry = &info->oti_obj_dentry;
3796         bool                    cache_xattr = false;
3797         int                     rc;
3798
3799         LASSERT(buf);
3800
3801         /* version get is not real XATTR but uses xattr API */
3802         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3803                 dt_obj_version_t *ver = buf->lb_buf;
3804
3805                 /* for version we are just using xattr API but change inode
3806                  * field instead */
3807                 if (buf->lb_len == 0)
3808                         return sizeof(dt_obj_version_t);
3809
3810                 if (buf->lb_len < sizeof(dt_obj_version_t))
3811                         return -ERANGE;
3812
3813                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3814                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3815
3816                 *ver = LDISKFS_I(inode)->i_fs_version;
3817
3818                 return sizeof(dt_obj_version_t);
3819         }
3820
3821         if (!dt_object_exists(dt))
3822                 return -ENOENT;
3823
3824         LASSERT(!dt_object_remote(dt));
3825         LASSERT(inode->i_op != NULL);
3826         LASSERT(inode->i_op->getxattr != NULL);
3827
3828         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3829             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3830                 cache_xattr = true;
3831
3832         if (cache_xattr) {
3833                 rc = osd_oxc_get(obj, name, buf);
3834                 if (rc != -ENOENT)
3835                         return rc;
3836         }
3837
3838         rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3839         if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3840                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3841                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3842                 struct filter_fid *ff;
3843                 struct ost_layout *ol;
3844
3845                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3846
3847                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3848                 if (rc)
3849                         return rc;
3850
3851                 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3852                         rc = -ENODATA;
3853                         goto cache;
3854                 }
3855
3856                 rc = sizeof(*ff);
3857                 if (buf->lb_len == 0 || !buf->lb_buf)
3858                         return rc;
3859
3860                 if (buf->lb_len < rc)
3861                         return -ERANGE;
3862
3863                 ff = buf->lb_buf;
3864                 ol = &ff->ff_layout;
3865                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3866                                                   PFID_STRIPE_IDX_BITS);
3867                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3868                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3869                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3870                 if (lma->lma_compat & LMAC_COMP_INFO) {
3871                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3872                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3873                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3874                 } else {
3875                         ol->ol_comp_start = 0;
3876                         ol->ol_comp_end = 0;
3877                         ol->ol_comp_id = 0;
3878                 }
3879         }
3880
3881 cache:
3882         if (cache_xattr) {
3883                 if (rc == -ENOENT || rc == -ENODATA)
3884                         osd_oxc_add(obj, name, NULL, 0);
3885                 else if (rc > 0 && buf->lb_buf != NULL)
3886                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3887         }
3888
3889         return rc;
3890 }
3891
3892 static int osd_declare_xattr_set(const struct lu_env *env,
3893                                  struct dt_object *dt,
3894                                  const struct lu_buf *buf, const char *name,
3895                                  int fl, struct thandle *handle)
3896 {
3897         struct osd_thandle *oh;
3898         int credits = 0;
3899         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3900
3901         LASSERT(handle != NULL);
3902
3903         oh = container_of0(handle, struct osd_thandle, ot_super);
3904         LASSERT(oh->ot_handle == NULL);
3905
3906         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3907                 /* For non-upgrading case, the LMA is set first and
3908                  * usually fit inode. But for upgrade case, the LMA
3909                  * may be in another separated EA block. */
3910                 if (dt_object_exists(dt)) {
3911                         if (fl == LU_XATTR_REPLACE)
3912                                 credits = 1;
3913                         else
3914                                 goto upgrade;
3915                 }
3916         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3917                 credits = 1;
3918         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3919                 /* We may need to delete the old PFID EA. */
3920                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3921                 if (fl == LU_XATTR_REPLACE)
3922                         credits += 1;
3923                 else
3924                         goto upgrade;
3925         } else {
3926
3927 upgrade:
3928                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3929
3930                 if (buf != NULL) {
3931                         ssize_t buflen;
3932
3933                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3934                                 /* learn xattr size from osd_xattr_get if
3935                                    attribute has not been read yet */
3936                                 buflen = __osd_xattr_get(
3937                                     osd_dt_obj(dt)->oo_inode,
3938                                     &osd_oti_get(env)->oti_obj_dentry,
3939                                     name, NULL, 0);
3940                                 if (buflen < 0)
3941                                         buflen = 0;
3942                         } else {
3943                                 buflen = buf->lb_len;
3944                         }
3945
3946                         if (buflen > sb->s_blocksize) {
3947                                 credits += osd_calc_bkmap_credits(
3948                                     sb, NULL, 0, -1,
3949                                     (buflen + sb->s_blocksize - 1) >>
3950                                     sb->s_blocksize_bits);
3951                         }
3952                 }
3953                 /*
3954                  * xattr set may involve inode quota change, reserve credits for
3955                  * dquot_initialize()
3956                  */
3957                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3958         }
3959
3960         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3961
3962         return 0;
3963 }
3964
3965 /*
3966  * Concurrency: @dt is write locked.
3967  */
3968 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3969                          const struct lu_buf *buf, const char *name, int fl,
3970                          struct thandle *handle)
3971 {
3972         struct osd_object *obj = osd_dt_obj(dt);
3973         struct inode *inode = obj->oo_inode;
3974         struct osd_thread_info *info = osd_oti_get(env);
3975         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3976         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3977         int fs_flags = 0;
3978         int len;
3979         int rc;
3980         ENTRY;
3981
3982         LASSERT(handle);
3983         LASSERT(buf);
3984
3985         /* version set is not real XATTR */
3986         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3987                 dt_obj_version_t *version = buf->lb_buf;
3988
3989                 /* for version we are just using xattr API but change inode
3990                  * field instead */
3991                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3992
3993                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3994                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3995
3996                 LDISKFS_I(inode)->i_fs_version = *version;
3997                 /* Version is set after all inode operations are finished,
3998                  * so we should mark it dirty here */
3999                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4000
4001                 RETURN(0);
4002         }
4003
4004         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4005                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4006
4007         len = buf->lb_len;
4008         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4009         if (fl & LU_XATTR_REPLACE)
4010                 fs_flags |= XATTR_REPLACE;
4011
4012         if (fl & LU_XATTR_CREATE)
4013                 fs_flags |= XATTR_CREATE;
4014
4015         /* For the OST device with 256 bytes inode size by default,
4016          * the PFID EA will be stored together with LMA EA to avoid
4017          * performance trouble. Otherwise the PFID EA can be stored
4018          * independently. LU-8998 */
4019         if (strcmp(name, XATTR_NAME_FID) == 0 &&
4020             LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
4021                 struct dentry *dentry = &info->oti_obj_dentry;
4022                 struct filter_fid *ff;
4023                 struct ost_layout *ol;
4024                 int fl;
4025
4026                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4027
4028                 ff = buf->lb_buf;
4029                 ol = &ff->ff_layout;
4030                 /* Old client does not send stripe information, store
4031                  * the PFID EA on disk directly. */
4032                 if (buf->lb_len == sizeof(struct lu_fid) ||
4033                     ol->ol_stripe_size == 0) {
4034                         len = sizeof(struct lu_fid);
4035                         goto set;
4036                 }
4037
4038                 if (buf->lb_len != sizeof(*ff))
4039                         RETURN(-EINVAL);
4040
4041                 rc = osd_get_lma(info, inode, dentry, loa);
4042                 if (unlikely(rc == -ENODATA)) {
4043                         /* Usually for upgarding from old device */
4044                         lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4045                                         LMAC_FID_ON_OST, 0);
4046                         fl = XATTR_CREATE;
4047                 } else if (rc) {
4048                         RETURN(rc);
4049                 } else {
4050                         fl = XATTR_REPLACE;
4051                 }
4052
4053                 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4054                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4055                                              PFID_STRIPE_IDX_BITS;
4056                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4057                 lma->lma_compat |= LMAC_STRIPE_INFO;
4058                 if (ol->ol_comp_id != 0) {
4059                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4060                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4061                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4062                         lma->lma_compat |= LMAC_COMP_INFO;
4063                 }
4064
4065                 lustre_loa_swab(loa, false);
4066
4067                 /* Remove old PFID EA entry firstly. */
4068                 ll_vfs_dq_init(inode);
4069                 rc = inode->i_op->removexattr(dentry, name);
4070                 if (rc && rc != -ENODATA)
4071                         RETURN(rc);
4072
4073                 /* Store the PFID EA inside the LMA EA. */
4074                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4075                                      sizeof(*loa), fl);
4076
4077                 RETURN(rc);
4078         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4079                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4080                 if (rc)
4081                         RETURN(rc);
4082
4083                 lma->lma_incompat |= LMAI_STRIPED;
4084                 lustre_lma_swab(lma);
4085                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4086                                      sizeof(*lma), XATTR_REPLACE);
4087                 if (rc != 0)
4088                         RETURN(rc);
4089         }
4090
4091 set:
4092         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4093         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4094
4095         if (rc == 0 &&
4096             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4097              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4098                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4099
4100         return rc;
4101 }
4102
4103 /*
4104  * Concurrency: @dt is read locked.
4105  */
4106 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4107                           const struct lu_buf *buf)
4108 {
4109         struct osd_object      *obj    = osd_dt_obj(dt);
4110         struct inode           *inode  = obj->oo_inode;
4111         struct osd_thread_info *info   = osd_oti_get(env);
4112         struct dentry          *dentry = &info->oti_obj_dentry;
4113
4114         if (!dt_object_exists(dt))
4115                 return -ENOENT;
4116
4117         LASSERT(!dt_object_remote(dt));
4118         LASSERT(inode->i_op != NULL);
4119         LASSERT(inode->i_op->listxattr != NULL);
4120
4121         dentry->d_inode = inode;
4122         dentry->d_sb = inode->i_sb;
4123         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4124 }
4125
4126 static int osd_declare_xattr_del(const struct lu_env *env,
4127                                  struct dt_object *dt, const char *name,
4128                                  struct thandle *handle)
4129 {
4130         struct osd_thandle *oh;
4131         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4132
4133         LASSERT(!dt_object_remote(dt));
4134         LASSERT(handle != NULL);
4135
4136         oh = container_of0(handle, struct osd_thandle, ot_super);
4137         LASSERT(oh->ot_handle == NULL);
4138
4139         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4140                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4141         /*
4142          * xattr del may involve inode quota change, reserve credits for
4143          * dquot_initialize()
4144          */
4145         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4146
4147         return 0;
4148 }
4149
4150 /*
4151  * Concurrency: @dt is write locked.
4152  */
4153 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4154                          const char *name, struct thandle *handle)
4155 {
4156         struct osd_object      *obj    = osd_dt_obj(dt);
4157         struct inode           *inode  = obj->oo_inode;
4158         struct osd_thread_info *info   = osd_oti_get(env);
4159         struct dentry          *dentry = &info->oti_obj_dentry;
4160         int                     rc;
4161
4162         if (!dt_object_exists(dt))
4163                 return -ENOENT;
4164
4165         LASSERT(!dt_object_remote(dt));
4166         LASSERT(inode->i_op != NULL);
4167         LASSERT(inode->i_op->removexattr != NULL);
4168         LASSERT(handle != NULL);
4169
4170         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4171
4172         ll_vfs_dq_init(inode);
4173         dentry->d_inode = inode;
4174         dentry->d_sb = inode->i_sb;
4175         rc = inode->i_op->removexattr(dentry, name);
4176         if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4177                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4178
4179                 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4180
4181                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4182                                  &info->oti_ost_attrs);
4183                 if (!rc) {
4184                         if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4185                                 rc = -ENODATA;
4186                                 goto out;
4187                         }
4188
4189                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4190                         lustre_lma_swab(lma);
4191                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4192                                              sizeof(*lma), XATTR_REPLACE);
4193                 }
4194         }
4195
4196 out:
4197         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4198
4199         if (rc == 0 &&
4200             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4201              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4202                 osd_oxc_del(obj, name);
4203
4204         return rc;
4205 }
4206
4207 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4208                            __u64 start, __u64 end)
4209 {
4210         struct osd_object       *obj    = osd_dt_obj(dt);
4211         struct inode            *inode  = obj->oo_inode;
4212         struct osd_thread_info  *info   = osd_oti_get(env);
4213         struct dentry           *dentry = &info->oti_obj_dentry;
4214         struct file             *file   = &info->oti_file;
4215         int                     rc;
4216
4217         ENTRY;
4218
4219         dentry->d_inode = inode;
4220         dentry->d_sb = inode->i_sb;
4221         file->f_path.dentry = dentry;
4222         file->f_mapping = inode->i_mapping;
4223         file->f_op = inode->i_fop;
4224         set_file_inode(file, inode);
4225
4226         rc = ll_vfs_fsync_range(file, start, end, 0);
4227
4228         RETURN(rc);
4229 }
4230
4231 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4232 {
4233         return 0;
4234 }
4235
4236 /*
4237  * Index operations.
4238  */
4239
4240 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4241                            const struct dt_index_features *feat)
4242 {
4243         struct iam_descr *descr;
4244
4245         if (osd_object_is_root(o))
4246                 return feat == &dt_directory_features;
4247
4248         LASSERT(o->oo_dir != NULL);
4249
4250         descr = o->oo_dir->od_container.ic_descr;
4251         if (feat == &dt_directory_features) {
4252                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4253                         return 1;
4254                 else
4255                         return 0;
4256         } else {
4257                 return
4258                         feat->dif_keysize_min <= descr->id_key_size &&
4259                         descr->id_key_size <= feat->dif_keysize_max &&
4260                         feat->dif_recsize_min <= descr->id_rec_size &&
4261                         descr->id_rec_size <= feat->dif_recsize_max &&
4262                         !(feat->dif_flags & (DT_IND_VARKEY |
4263                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4264                         ergo(feat->dif_flags & DT_IND_UPDATE,
4265                              1 /* XXX check that object (and file system) is
4266                                 * writable */);
4267         }
4268 }
4269
4270 static int osd_iam_container_init(const struct lu_env *env,
4271                                   struct osd_object *obj,
4272                                   struct osd_directory *dir)
4273 {
4274         struct iam_container *bag = &dir->od_container;
4275         int result;
4276
4277         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4278         if (result != 0)
4279                 return result;
4280
4281         result = iam_container_setup(bag);
4282         if (result == 0)
4283                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4284         else
4285                 iam_container_fini(bag);
4286
4287         return result;
4288 }
4289
4290
4291 /*
4292  * Concurrency: no external locking is necessary.
4293  */
4294 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4295                          const struct dt_index_features *feat)
4296 {
4297         int                      result;
4298         int                      skip_iam = 0;
4299         struct osd_object       *obj = osd_dt_obj(dt);
4300
4301         LINVRNT(osd_invariant(obj));
4302
4303         if (osd_object_is_root(obj)) {
4304                 dt->do_index_ops = &osd_index_ea_ops;
4305                 result = 0;
4306         } else if (feat == &dt_directory_features) {
4307                 dt->do_index_ops = &osd_index_ea_ops;
4308                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4309                         result = 0;
4310                 else
4311                         result = -ENOTDIR;
4312                 skip_iam = 1;
4313         } else if (unlikely(feat == &dt_otable_features)) {
4314                 dt->do_index_ops = &osd_otable_ops;
4315                 return 0;
4316         } else if (unlikely(feat == &dt_acct_features)) {
4317                 dt->do_index_ops = &osd_acct_index_ops;
4318                 result = 0;
4319                 skip_iam = 1;
4320         } else if (!osd_has_index(obj)) {
4321                 struct osd_directory *dir;
4322
4323                 OBD_ALLOC_PTR(dir);
4324                 if (dir != NULL) {
4325
4326                         spin_lock(&obj->oo_guard);
4327                         if (obj->oo_dir == NULL)
4328                                 obj->oo_dir = dir;
4329                         else
4330                                 /*
4331                                  * Concurrent thread allocated container data.
4332                                  */
4333                                 OBD_FREE_PTR(dir);
4334                         spin_unlock(&obj->oo_guard);
4335                         /*
4336                          * Now, that we have container data, serialize its
4337                          * initialization.
4338                          */
4339                         down_write(&obj->oo_ext_idx_sem);
4340                         /*
4341                          * recheck under lock.
4342                          */
4343                         if (!osd_has_index(obj))
4344                                 result = osd_iam_container_init(env, obj,
4345                                                                 obj->oo_dir);
4346                         else
4347                                 result = 0;
4348                         up_write(&obj->oo_ext_idx_sem);
4349                 } else {
4350                         result = -ENOMEM;
4351                 }
4352         } else {
4353                 result = 0;
4354         }
4355
4356         if (result == 0 && skip_iam == 0) {
4357                 if (!osd_iam_index_probe(env, obj, feat))
4358                         result = -ENOTDIR;
4359         }
4360         LINVRNT(osd_invariant(obj));
4361
4362         return result;
4363 }
4364
4365 static int osd_otable_it_attr_get(const struct lu_env *env,
4366                                  struct dt_object *dt,
4367                                  struct lu_attr *attr)
4368 {
4369         attr->la_valid = 0;
4370         return 0;
4371 }
4372
4373 static const struct dt_object_operations osd_obj_ops = {
4374         .do_read_lock         = osd_object_read_lock,
4375         .do_write_lock        = osd_object_write_lock,
4376         .do_read_unlock       = osd_object_read_unlock,
4377         .do_write_unlock      = osd_object_write_unlock,
4378         .do_write_locked      = osd_object_write_locked,
4379         .do_attr_get          = osd_attr_get,
4380         .do_declare_attr_set  = osd_declare_attr_set,
4381         .do_attr_set          = osd_attr_set,
4382         .do_ah_init           = osd_ah_init,
4383         .do_declare_create    = osd_declare_object_create,
4384         .do_create            = osd_object_create,
4385         .do_declare_destroy   = osd_declare_object_destroy,
4386         .do_destroy           = osd_object_destroy,
4387         .do_index_try         = osd_index_try,
4388         .do_declare_ref_add   = osd_declare_object_ref_add,
4389         .do_ref_add           = osd_object_ref_add,
4390         .do_declare_ref_del   = osd_declare_object_ref_del,
4391         .do_ref_del           = osd_object_ref_del,
4392         .do_xattr_get         = osd_xattr_get,
4393         .do_declare_xattr_set = osd_declare_xattr_set,
4394         .do_xattr_set         = osd_xattr_set,
4395         .do_declare_xattr_del = osd_declare_xattr_del,
4396         .do_xattr_del         = osd_xattr_del,
4397         .do_xattr_list        = osd_xattr_list,
4398         .do_object_sync       = osd_object_sync,
4399         .do_invalidate        = osd_invalidate,
4400 };
4401
4402 /**
4403  * dt_object_operations for interoperability mode
4404  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4405  */
4406 static const struct dt_object_operations osd_obj_ea_ops = {
4407         .do_read_lock         = osd_object_read_lock,
4408         .do_write_lock        = osd_object_write_lock,
4409         .do_read_unlock       = osd_object_read_unlock,
4410         .do_write_unlock      = osd_object_write_unlock,
4411         .do_write_locked      = osd_object_write_locked,
4412         .do_attr_get          = osd_attr_get,
4413         .do_declare_attr_set  = osd_declare_attr_set,
4414         .do_attr_set          = osd_attr_set,
4415         .do_ah_init           = osd_ah_init,
4416         .do_declare_create    = osd_declare_object_create,
4417         .do_create            = osd_object_ea_create,
4418         .do_declare_destroy   = osd_declare_object_destroy,
4419         .do_destroy           = osd_object_destroy,
4420         .do_index_try         = osd_index_try,
4421         .do_declare_ref_add   = osd_declare_object_ref_add,
4422         .do_ref_add           = osd_object_ref_add,
4423         .do_declare_ref_del   = osd_declare_object_ref_del,
4424         .do_ref_del           = osd_object_ref_del,
4425         .do_xattr_get         = osd_xattr_get,
4426         .do_declare_xattr_set = osd_declare_xattr_set,
4427         .do_xattr_set         = osd_xattr_set,
4428         .do_declare_xattr_del = osd_declare_xattr_del,
4429         .do_xattr_del         = osd_xattr_del,
4430         .do_xattr_list        = osd_xattr_list,
4431         .do_object_sync       = osd_object_sync,
4432         .do_invalidate        = osd_invalidate,
4433 };
4434
4435 static const struct dt_object_operations osd_obj_otable_it_ops = {
4436         .do_attr_get    = osd_otable_it_attr_get,
4437         .do_index_try   = osd_index_try,
4438 };
4439
4440 static int osd_index_declare_iam_delete(const struct lu_env *env,
4441                                         struct dt_object *dt,
4442                                         const struct dt_key *key,
4443                                         struct thandle *handle)
4444 {
4445         struct osd_thandle    *oh;
4446
4447         oh = container_of0(handle, struct osd_thandle, ot_super);
4448         LASSERT(oh->ot_handle == NULL);
4449
4450         /* Recycle  may cause additional three blocks to be changed. */
4451         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4452                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4453
4454         return 0;
4455 }
4456
4457 /**
4458  *      delete a (key, value) pair from index \a dt specified by \a key
4459  *
4460  *      \param  dt      osd index object
4461  *      \param  key     key for index
4462  *      \param  rec     record reference
4463  *      \param  handle  transaction handler
4464  *
4465  *      \retval  0  success
4466  *      \retval -ve   failure
4467  */
4468 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4469                                 const struct dt_key *key,
4470                                 struct thandle *handle)
4471 {
4472         struct osd_thread_info *oti = osd_oti_get(env);
4473         struct osd_object      *obj = osd_dt_obj(dt);
4474         struct osd_thandle     *oh;
4475         struct iam_path_descr  *ipd;
4476         struct iam_container   *bag = &obj->oo_dir->od_container;
4477         int                     rc;
4478         ENTRY;
4479
4480         if (!dt_object_exists(dt))
4481                 RETURN(-ENOENT);
4482
4483         LINVRNT(osd_invariant(obj));
4484         LASSERT(!dt_object_remote(dt));
4485         LASSERT(bag->ic_object == obj->oo_inode);
4486         LASSERT(handle != NULL);
4487
4488         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4489
4490         ipd = osd_idx_ipd_get(env, bag);
4491         if (unlikely(ipd == NULL))
4492                 RETURN(-ENOMEM);
4493
4494         oh = container_of0(handle, struct osd_thandle, ot_super);
4495         LASSERT(oh->ot_handle != NULL);
4496         LASSERT(oh->ot_handle->h_transaction != NULL);
4497
4498         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4499                 /* swab quota uid/gid provided by caller */
4500                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4501                 key = (const struct dt_key *)&oti->oti_quota_id;
4502         }
4503
4504         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4505         osd_ipd_put(env, bag, ipd);
4506         LINVRNT(osd_invariant(obj));
4507         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4508         RETURN(rc);
4509 }
4510
4511 static int osd_index_declare_ea_delete(const struct lu_env *env,
4512                                        struct dt_object *dt,
4513                                        const struct dt_key *key,
4514                                        struct thandle *handle)
4515 {
4516         struct osd_thandle *oh;
4517         struct inode       *inode;
4518         int                 rc, credits;
4519         ENTRY;
4520
4521         LASSERT(!dt_object_remote(dt));
4522         LASSERT(handle != NULL);
4523
4524         oh = container_of0(handle, struct osd_thandle, ot_super);
4525         LASSERT(oh->ot_handle == NULL);
4526
4527         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4528         if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4529                 /* '..' to a remote object has a local representative */
4530                 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4531                 /* to reset LMAI_REMOTE_PARENT */
4532                 credits += 1;
4533         }
4534         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4535
4536         inode = osd_dt_obj(dt)->oo_inode;
4537         if (inode == NULL)
4538                 RETURN(-ENOENT);
4539
4540         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4541                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4542                                    true, NULL, false);
4543         RETURN(rc);
4544 }
4545
4546 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4547                                           struct dt_rec *fid)
4548 {
4549         struct osd_fid_pack *rec;
4550         int                  rc = -ENODATA;
4551
4552         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4553                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4554                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4555                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4556                         rc = -EINVAL;
4557         }
4558         return rc;
4559 }
4560
4561 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4562                           const struct lu_fid *fid)
4563 {
4564         struct seq_server_site  *ss = osd_seq_site(osd);
4565         ENTRY;
4566
4567         /* FID seqs not in FLDB, must be local seq */
4568         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4569                 RETURN(0);
4570
4571         /* If FLD is not being initialized yet, it only happens during the
4572          * initialization, likely during mgs initialization, and we assume
4573          * this is local FID. */
4574         if (ss == NULL || ss->ss_server_fld == NULL)
4575                 RETURN(0);
4576
4577         /* Only check the local FLDB here */
4578         if (osd_seq_exists(env, osd, fid_seq(fid)))
4579                 RETURN(0);
4580
4581         RETURN(1);
4582 }
4583
4584 /**
4585  * Index delete function for interoperability mode (b11826).
4586  * It will remove the directory entry added by osd_index_ea_insert().
4587  * This entry is needed to maintain name->fid mapping.
4588  *
4589  * \param key,  key i.e. file entry to be deleted
4590  *
4591  * \retval   0, on success
4592  * \retval -ve, on error
4593  */
4594 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4595                                const struct dt_key *key, struct thandle *handle)
4596 {
4597         struct osd_object          *obj = osd_dt_obj(dt);
4598         struct inode               *dir = obj->oo_inode;
4599         struct dentry              *dentry;
4600         struct osd_thandle         *oh;
4601         struct ldiskfs_dir_entry_2 *de = NULL;
4602         struct buffer_head         *bh;
4603         struct htree_lock          *hlock = NULL;
4604         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
4605         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4606         int                        rc;
4607         ENTRY;
4608
4609         if (!dt_object_exists(dt))
4610                 RETURN(-ENOENT);
4611
4612         LINVRNT(osd_invariant(obj));
4613         LASSERT(!dt_object_remote(dt));
4614         LASSERT(handle != NULL);
4615
4616         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4617
4618         oh = container_of(handle, struct osd_thandle, ot_super);
4619         LASSERT(oh->ot_handle != NULL);
4620         LASSERT(oh->ot_handle->h_transaction != NULL);
4621
4622         ll_vfs_dq_init(dir);
4623         dentry = osd_child_dentry_get(env, obj,
4624                                       (char *)key, strlen((char *)key));
4625
4626         if (obj->oo_hl_head != NULL) {
4627                 hlock = osd_oti_get(env)->oti_hlock;
4628                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4629                                    dir, LDISKFS_HLOCK_DEL);
4630         } else {
4631                 down_write(&obj->oo_ext_idx_sem);
4632         }
4633
4634         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4635         if (!IS_ERR(bh)) {
4636                 /* If this is not the ".." entry, it might be a remote DNE
4637                  * entry and  we need to check if the FID is for a remote
4638                  * MDT.  If the FID is  not in the directory entry (e.g.
4639                  * upgraded 1.8 filesystem without dirdata enabled) then
4640                  * we need to get the FID from the LMA. For a remote directory
4641                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4642                  * case.
4643                  *
4644                  * Delete the entry before the agent inode in order to
4645                  * simplify error handling.  At worst an error after deleting
4646                  * the entry first might leak the agent inode afterward. The
4647                  * reverse would need filesystem abort in case of error deleting
4648                  * the entry after the agent had been removed, or leave a
4649                  * dangling entry pointing at a random inode. */
4650                 if (strcmp((char *)key, dotdot) != 0) {
4651                         LASSERT(de != NULL);
4652                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4653                         if (rc == -ENODATA) {
4654                                 /* can't get FID, postpone to the end of the
4655                                  * transaction when iget() is safe */
4656                                 osd_schedule_agent_inode_removal(env, oh,
4657                                                 le32_to_cpu(de->inode));
4658                         } else if (rc == 0 &&
4659                                    unlikely(osd_remote_fid(env, osd, fid))) {
4660                                 osd_schedule_agent_inode_removal(env, oh,
4661                                                 le32_to_cpu(de->inode));
4662                         }
4663                 }
4664                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4665                 brelse(bh);
4666         } else {
4667                 rc = PTR_ERR(bh);
4668         }
4669         if (hlock != NULL)
4670                 ldiskfs_htree_unlock(hlock);
4671         else
4672                 up_write(&obj->oo_ext_idx_sem);
4673
4674         if (rc != 0)
4675                 GOTO(out, rc);
4676
4677         /* For inode on the remote MDT, .. will point to
4678          * /Agent directory, Check whether it needs to delete
4679          * from agent directory */
4680         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4681                 int ret;
4682
4683                 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4684                                                     obj, oh);
4685                 if (ret != 0)
4686                         /* Sigh, the entry has been deleted, and
4687                          * it is not easy to revert it back, so
4688                          * let's keep this error private, and let
4689                          * LFSCK fix it. XXX */
4690                         CERROR("%s: delete remote parent "DFID": rc = %d\n",
4691                                osd_name(osd), PFID(fid), ret);
4692         }
4693 out:
4694         LASSERT(osd_invariant(obj));
4695         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4696         RETURN(rc);
4697 }
4698
4699 /**
4700  *      Lookup index for \a key and copy record to \a rec.
4701  *
4702  *      \param  dt      osd index object
4703  *      \param  key     key for index
4704  *      \param  rec     record reference
4705  *
4706  *      \retval  +ve  success : exact mach
4707  *      \retval  0    return record with key not greater than \a key
4708  *      \retval -ve   failure
4709  */
4710 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4711                                 struct dt_rec *rec, const struct dt_key *key)
4712 {
4713         struct osd_object      *obj = osd_dt_obj(dt);
4714         struct iam_path_descr  *ipd;
4715         struct iam_container   *bag = &obj->oo_dir->od_container;
4716         struct osd_thread_info *oti = osd_oti_get(env);
4717         struct iam_iterator    *it = &oti->oti_idx_it;
4718         struct iam_rec         *iam_rec;
4719         int                     rc;
4720         ENTRY;
4721
4722         if (!dt_object_exists(dt))
4723                 RETURN(-ENOENT);
4724
4725         LASSERT(osd_invariant(obj));
4726         LASSERT(!dt_object_remote(dt));
4727         LASSERT(bag->ic_object == obj->oo_inode);
4728
4729         ipd = osd_idx_ipd_get(env, bag);
4730         if (IS_ERR(ipd))
4731                 RETURN(-ENOMEM);
4732
4733         /* got ipd now we can start iterator. */
4734         iam_it_init(it, bag, 0, ipd);
4735
4736         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4737                 /* swab quota uid/gid provided by caller */
4738                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4739                 key = (const struct dt_key *)&oti->oti_quota_id;
4740         }
4741
4742         rc = iam_it_get(it, (struct iam_key *)key);
4743         if (rc >= 0) {
4744                 if (S_ISDIR(obj->oo_inode->i_mode))
4745                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4746                 else
4747                         iam_rec = (struct iam_rec *) rec;
4748
4749                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4750
4751                 if (S_ISDIR(obj->oo_inode->i_mode))
4752                         osd_fid_unpack((struct lu_fid *) rec,
4753                                        (struct osd_fid_pack *)iam_rec);
4754                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4755                         osd_quota_unpack(obj, rec);
4756         }
4757
4758         iam_it_put(it);
4759         iam_it_fini(it);
4760         osd_ipd_put(env, bag, ipd);
4761
4762         LINVRNT(osd_invariant(obj));
4763
4764         RETURN(rc);
4765 }
4766
4767 static int osd_index_declare_iam_insert(const struct lu_env *env,
4768                                         struct dt_object *dt,
4769                                         const struct dt_rec *rec,
4770                                         const struct dt_key *key,
4771                                         struct thandle *handle)
4772 {
4773         struct osd_thandle *oh;
4774
4775         LASSERT(handle != NULL);
4776
4777         oh = container_of0(handle, struct osd_thandle, ot_super);
4778         LASSERT(oh->ot_handle == NULL);
4779
4780         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4781                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4782
4783         return 0;
4784 }
4785
4786 /**
4787  *      Inserts (key, value) pair in \a dt index object.
4788  *
4789  *      \param  dt      osd index object
4790  *      \param  key     key for index
4791  *      \param  rec     record reference
4792  *      \param  th      transaction handler
4793  *
4794  *      \retval  0  success
4795  *      \retval -ve failure
4796  */
4797 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4798                                 const struct dt_rec *rec,
4799                                 const struct dt_key *key, struct thandle *th,
4800                                 int ignore_quota)
4801 {
4802         struct osd_object     *obj = osd_dt_obj(dt);
4803         struct iam_path_descr *ipd;
4804         struct osd_thandle    *oh;
4805         struct iam_container  *bag;
4806         struct osd_thread_info *oti = osd_oti_get(env);
4807         struct iam_rec         *iam_rec;
4808         int                     rc;
4809         ENTRY;
4810
4811         if (!dt_object_exists(dt))
4812                 RETURN(-ENOENT);
4813
4814         LINVRNT(osd_invariant(obj));
4815         LASSERT(!dt_object_remote(dt));
4816
4817         bag = &obj->oo_dir->od_container;
4818         LASSERT(bag->ic_object == obj->oo_inode);
4819         LASSERT(th != NULL);
4820
4821         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4822
4823         ipd = osd_idx_ipd_get(env, bag);
4824         if (unlikely(ipd == NULL))
4825                 RETURN(-ENOMEM);
4826
4827         oh = container_of0(th, struct osd_thandle, ot_super);
4828         LASSERT(oh->ot_handle != NULL);
4829         LASSERT(oh->ot_handle->h_transaction != NULL);
4830         if (S_ISDIR(obj->oo_inode->i_mode)) {
4831                 iam_rec = (struct iam_rec *)oti->oti_ldp;
4832                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4833         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4834                 /* pack quota uid/gid */
4835                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4836                 key = (const struct dt_key *)&oti->oti_quota_id;
4837                 /* pack quota record */
4838                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4839                 iam_rec = (struct iam_rec *)rec;
4840         } else {
4841                 iam_rec = (struct iam_rec *)rec;
4842         }
4843
4844         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4845                         iam_rec, ipd);
4846         osd_ipd_put(env, bag, ipd);
4847         LINVRNT(osd_invariant(obj));
4848         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4849         RETURN(rc);
4850 }
4851
4852 /**
4853  * Calls ldiskfs_add_entry() to add directory entry
4854  * into the directory. This is required for
4855  * interoperability mode (b11826)
4856  *
4857  * \retval   0, on success
4858  * \retval -ve, on error
4859  */
4860 static int __osd_ea_add_rec(struct osd_thread_info *info,
4861                             struct osd_object *pobj, struct inode  *cinode,
4862                             const char *name, const struct lu_fid *fid,
4863                             struct htree_lock *hlock, struct thandle *th)
4864 {
4865         struct ldiskfs_dentry_param *ldp;
4866         struct dentry               *child;
4867         struct osd_thandle          *oth;
4868         int                          rc;
4869
4870         oth = container_of(th, struct osd_thandle, ot_super);
4871         LASSERT(oth->ot_handle != NULL);
4872         LASSERT(oth->ot_handle->h_transaction != NULL);
4873         LASSERT(pobj->oo_inode);
4874
4875         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4876         if (unlikely(pobj->oo_inode ==
4877                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4878                 ldp->edp_magic = 0;
4879         else
4880                 osd_get_ldiskfs_dirent_param(ldp, fid);
4881         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4882         child->d_fsdata = (void *)ldp;
4883         ll_vfs_dq_init(pobj->oo_inode);
4884         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4885                                    child, cinode, hlock);
4886         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4887                 struct ldiskfs_dir_entry_2      *de;
4888                 struct buffer_head              *bh;
4889                 int                              rc1;
4890
4891                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4892                                             NULL, hlock);
4893                 if (!IS_ERR(bh)) {
4894                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4895                                                                bh);
4896                         if (rc1 == 0) {
4897                                 if (S_ISDIR(cinode->i_mode))
4898                                         de->file_type = LDISKFS_DIRENT_LUFID |
4899                                                         LDISKFS_FT_REG_FILE;
4900                                 else
4901                                         de->file_type = LDISKFS_DIRENT_LUFID |
4902                                                         LDISKFS_FT_DIR;
4903                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4904                                                               NULL, bh);
4905                         }
4906                         brelse(bh);
4907                 }
4908         }
4909
4910         RETURN(rc);
4911 }
4912
4913 /**
4914  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4915  * into the directory.Also sets flags into osd object to
4916  * indicate dot and dotdot are created. This is required for
4917  * interoperability mode (b11826)
4918  *
4919  * \param dir   directory for dot and dotdot fixup.
4920  * \param obj   child object for linking
4921  *
4922  * \retval   0, on success
4923  * \retval -ve, on error
4924  */
4925 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4926                               struct osd_object *dir,
4927                               struct inode *parent_dir, const char *name,
4928                               const struct lu_fid *dot_fid,
4929                               const struct lu_fid *dot_dot_fid,
4930                               struct thandle *th)
4931 {
4932         struct inode                *inode = dir->oo_inode;
4933         struct osd_thandle          *oth;
4934         int result = 0;
4935
4936         oth = container_of(th, struct osd_thandle, ot_super);
4937         LASSERT(oth->ot_handle->h_transaction != NULL);
4938         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4939
4940         if (strcmp(name, dot) == 0) {
4941                 if (dir->oo_compat_dot_created) {
4942                         result = -EEXIST;
4943                 } else {
4944                         LASSERT(inode->i_ino == parent_dir->i_ino);
4945                         dir->oo_compat_dot_created = 1;
4946                         result = 0;
4947                 }
4948         } else if (strcmp(name, dotdot) == 0) {
4949                 if (!dir->oo_compat_dot_created)
4950                         return -EINVAL;
4951                 /* in case of rename, dotdot is already created */
4952                 if (dir->oo_compat_dotdot_created) {
4953                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4954                                                 dot_dot_fid, NULL, th);
4955                 }
4956
4957                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4958                         struct lu_fid tfid = *dot_dot_fid;
4959
4960                         tfid.f_oid--;
4961                         result = osd_add_dot_dotdot_internal(info,
4962                                         dir->oo_inode, parent_dir, dot_fid,
4963                                         &tfid, oth);
4964                 } else {
4965                         result = osd_add_dot_dotdot_internal(info,
4966                                         dir->oo_inode, parent_dir, dot_fid,
4967                                         dot_dot_fid, oth);
4968                 }
4969
4970                 if (result == 0)
4971                         dir->oo_compat_dotdot_created = 1;
4972         }
4973
4974         return result;
4975 }
4976
4977
4978 /**
4979  * It will call the appropriate osd_add* function and return the
4980  * value, return by respective functions.
4981  */
4982 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4983                           struct inode *cinode, const char *name,
4984                           const struct lu_fid *fid, struct thandle *th)
4985 {
4986         struct osd_thread_info *info   = osd_oti_get(env);
4987         struct htree_lock      *hlock;
4988         int                     rc;
4989
4990         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4991
4992         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4993                                                    name[2] =='\0'))) {
4994                 if (hlock != NULL) {
4995                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4996                                            pobj->oo_inode, 0);
4997                 } else {
4998                         down_write(&pobj->oo_ext_idx_sem);
4999                 }
5000
5001                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5002                                         lu_object_fid(&pobj->oo_dt.do_lu),
5003                                         fid, th);
5004         } else {
5005                 if (hlock != NULL) {
5006                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5007                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5008                 } else {
5009                         down_write(&pobj->oo_ext_idx_sem);
5010                 }
5011
5012                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5013                         struct lu_fid *tfid = &info->oti_fid;
5014
5015                         *tfid = *fid;
5016                         tfid->f_ver = ~0;
5017                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5018                                               tfid, hlock, th);
5019                 } else {
5020                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5021                                               hlock, th);
5022                 }
5023         }
5024         if (hlock != NULL)
5025                 ldiskfs_htree_unlock(hlock);
5026         else
5027                 up_write(&pobj->oo_ext_idx_sem);
5028
5029         return rc;
5030 }
5031
5032 static int
5033 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5034                       struct osd_idmap_cache *oic)
5035 {
5036         struct osd_scrub *scrub = &dev->od_scrub;
5037         struct lu_fid *fid = &oic->oic_fid;
5038         struct osd_inode_id *id = &oic->oic_lid;
5039         struct inode *inode = NULL;
5040         int once  = 0;
5041         bool insert;
5042         int rc;
5043         ENTRY;
5044
5045         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5046                 RETURN(0);
5047
5048         if (scrub->os_pos_current > id->oii_ino)
5049                 RETURN(0);
5050
5051 again:
5052         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5053         if (rc == -ENOENT) {
5054                 __u32 gen = id->oii_gen;
5055
5056                 insert = true;
5057                 if (inode != NULL)
5058                         goto trigger;
5059
5060                 inode = osd_iget(oti, dev, id);
5061                 /* The inode has been removed (by race maybe). */
5062                 if (IS_ERR(inode)) {
5063                         rc = PTR_ERR(inode);
5064
5065                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5066                 }
5067
5068                 /* The OI mapping is lost. */
5069                 if (gen != OSD_OII_NOGEN)
5070                         goto trigger;
5071
5072                 iput(inode);
5073                 /* The inode may has been reused by others, we do not know,
5074                  * leave it to be handled by subsequent osd_fid_lookup(). */
5075                 RETURN(0);
5076         } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
5077                 RETURN(rc);
5078         } else {
5079                 insert = false;
5080         }
5081
5082 trigger:
5083         if (thread_is_running(&scrub->os_thread)) {
5084                 if (inode == NULL) {
5085                         inode = osd_iget(oti, dev, id);
5086                         /* The inode has been removed (by race maybe). */
5087                         if (IS_ERR(inode)) {
5088                                 rc = PTR_ERR(inode);
5089
5090                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5091                         }
5092                 }
5093
5094                 rc = osd_oii_insert(dev, oic, insert);
5095                 /* There is race condition between osd_oi_lookup and OI scrub.
5096                  * The OI scrub finished just after osd_oi_lookup() failure.
5097                  * Under such case, it is unnecessary to trigger OI scrub again,
5098                  * but try to call osd_oi_lookup() again. */
5099                 if (unlikely(rc == -EAGAIN))
5100                         goto again;
5101
5102                 if (!S_ISDIR(inode->i_mode))
5103                         rc = 0;
5104                 else
5105                         rc = osd_check_lmv(oti, dev, inode, oic);
5106
5107                 iput(inode);
5108                 RETURN(rc);
5109         }
5110
5111         if (!dev->od_noscrub && ++once == 1) {
5112                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5113                                      SS_CLEAR_FAILOUT);
5114                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5115                        "%.16s: trigger partial OI scrub for RPC inconsistency "
5116                        "checking FID "DFID": rc = %d\n",
5117                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
5118                        PFID(fid), rc);
5119                 if (rc == 0 || rc == -EALREADY)
5120                         goto again;
5121         }
5122
5123         if (inode != NULL)
5124                 iput(inode);
5125
5126         RETURN(rc);
5127 }
5128
5129 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5130                                struct osd_device *dev,
5131                                struct osd_idmap_cache *oic,
5132                                struct lu_fid *fid, __u32 ino)
5133 {
5134         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5135         struct inode *inode;
5136         int rc;
5137
5138         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5139         inode = osd_iget(oti, dev, &oic->oic_lid);
5140         if (IS_ERR(inode)) {
5141                 fid_zero(&oic->oic_fid);
5142                 return PTR_ERR(inode);
5143         }
5144
5145         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5146         iput(inode);
5147         if (rc != 0)
5148                 fid_zero(&oic->oic_fid);
5149         else
5150                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5151         return rc;
5152 }
5153
5154 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5155                       struct osd_inode_id *id, const struct lu_fid *fid)
5156 {
5157         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5158                id->oii_ino, id->oii_gen, info);
5159         info->oti_cache.oic_lid = *id;
5160         info->oti_cache.oic_fid = *fid;
5161         info->oti_cache.oic_dev = osd;
5162 }
5163
5164 /**
5165  * Get parent FID from the linkEA.
5166  *
5167  * For a directory which parent resides on remote MDT, to satisfy the
5168  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5169  * the other hand, to make the lookup(..) on the directory can return
5170  * the real parent FID, we append the real parent FID after its ".."
5171  * name entry in the /REMOTE_PARENT_DIR.
5172  *
5173  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5174  * backup. So after the restore, we cannot get the right parent FID from
5175  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5176  * we have stored the real parent FID in the directory object's linkEA,
5177  * we can parse the linkEA for the real parent FID.
5178  *
5179  * \param[in] env       pointer to the thread context
5180  * \param[in] obj       pointer to the object to be handled
5181  * \param[out]fid       pointer to the buffer to hold the parent FID
5182  *
5183  * \retval              0 for getting the real parent FID successfully
5184  * \retval              negative error number on failure
5185  */
5186 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5187                                     struct osd_object *obj,
5188                                     struct lu_fid *fid)
5189 {
5190         struct osd_thread_info  *oti    = osd_oti_get(env);
5191         struct lu_buf           *buf    = &oti->oti_big_buf;
5192         struct dentry           *dentry = &oti->oti_obj_dentry;
5193         struct inode            *inode  = obj->oo_inode;
5194         struct linkea_data       ldata  = { NULL };
5195         int                      rc;
5196         ENTRY;
5197
5198         fid_zero(fid);
5199         if (!S_ISDIR(inode->i_mode))
5200                 RETURN(-EIO);
5201
5202 again:
5203         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5204                              buf->lb_buf, buf->lb_len);
5205         if (rc == -ERANGE) {
5206                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5207                                      NULL, 0);
5208                 if (rc > 0) {
5209                         lu_buf_realloc(buf, rc);
5210                         if (buf->lb_buf == NULL)
5211                                 RETURN(-ENOMEM);
5212
5213                         goto again;
5214                 }
5215         }
5216
5217         if (unlikely(rc == 0))
5218                 RETURN(-ENODATA);
5219
5220         if (rc < 0)
5221                 RETURN(rc);
5222
5223         if (unlikely(buf->lb_buf == NULL)) {
5224                 lu_buf_realloc(buf, rc);
5225                 if (buf->lb_buf == NULL)
5226                         RETURN(-ENOMEM);
5227
5228                 goto again;
5229         }
5230
5231         ldata.ld_buf = buf;
5232         rc = linkea_init_with_rec(&ldata);
5233         if (!rc) {
5234                 linkea_first_entry(&ldata);
5235                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5236         }
5237
5238         RETURN(rc);
5239 }
5240
5241 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5242                                     struct inode *inode,
5243                                     const struct lu_fid *pfid,
5244                                     const char *name, const int namelen)
5245 {
5246         struct osd_thread_info *oti = osd_oti_get(env);
5247         struct lu_buf *buf = &oti->oti_big_buf;
5248         struct dentry *dentry = &oti->oti_obj_dentry;
5249         struct linkea_data ldata = { NULL };
5250         struct lu_name cname = { .ln_name = name,
5251                                  .ln_namelen = namelen };
5252         int rc;
5253         ENTRY;
5254
5255 again:
5256         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5257                              buf->lb_buf, buf->lb_len);
5258         if (rc == -ERANGE)
5259                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5260
5261         if (rc < 0)
5262                 RETURN(rc);
5263
5264         if (unlikely(rc == 0))
5265                 RETURN(-ENODATA);
5266
5267         if (buf->lb_len < rc) {
5268                 lu_buf_realloc(buf, rc);
5269                 if (buf->lb_buf == NULL)
5270                         RETURN(-ENOMEM);
5271
5272                 goto again;
5273         }
5274
5275         ldata.ld_buf = buf;
5276         rc = linkea_init_with_rec(&ldata);
5277         if (!rc)
5278                 rc = linkea_links_find(&ldata, &cname, pfid);
5279
5280         RETURN(rc);
5281 }
5282
5283 /**
5284  * Calls ->lookup() to find dentry. From dentry get inode and
5285  * read inode's ea to get fid. This is required for  interoperability
5286  * mode (b11826)
5287  *
5288  * \retval   0, on success
5289  * \retval -ve, on error
5290  */
5291 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5292                              struct dt_rec *rec, const struct dt_key *key)
5293 {
5294         struct inode                    *dir    = obj->oo_inode;
5295         struct dentry                   *dentry;
5296         struct ldiskfs_dir_entry_2      *de;
5297         struct buffer_head              *bh;
5298         struct lu_fid                   *fid = (struct lu_fid *) rec;
5299         struct htree_lock               *hlock = NULL;
5300         int                             ino;
5301         int                             rc;
5302         ENTRY;
5303
5304         LASSERT(dir->i_op != NULL);
5305         LASSERT(dir->i_op->lookup != NULL);
5306
5307         dentry = osd_child_dentry_get(env, obj,
5308                                       (char *)key, strlen((char *)key));
5309
5310         if (obj->oo_hl_head != NULL) {
5311                 hlock = osd_oti_get(env)->oti_hlock;
5312                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5313                                    dir, LDISKFS_HLOCK_LOOKUP);
5314         } else {
5315                 down_read(&obj->oo_ext_idx_sem);
5316         }
5317
5318         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5319         if (!IS_ERR(bh)) {
5320                 struct osd_thread_info *oti = osd_oti_get(env);
5321                 struct osd_inode_id *id = &oti->oti_id;
5322                 struct osd_idmap_cache *oic = &oti->oti_cache;
5323                 struct osd_device *dev = osd_obj2dev(obj);
5324
5325                 ino = le32_to_cpu(de->inode);
5326                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5327                         brelse(bh);
5328                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5329                         GOTO(out, rc);
5330                 }
5331
5332                 rc = osd_get_fid_from_dentry(de, rec);
5333
5334                 /* done with de, release bh */
5335                 brelse(bh);
5336                 if (rc != 0) {
5337                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
5338                                 const char *name = (const char *)key;
5339
5340                                 /* If the parent is on remote MDT, and there
5341                                  * is no FID-in-dirent, then we have to get
5342                                  * the parent FID from the linkEA.  */
5343                                 if (likely(strlen(name) == 2 &&
5344                                            name[0] == '.' && name[1] == '.'))
5345                                         rc = osd_get_pfid_from_linkea(env, obj,
5346                                                                       fid);
5347                         } else {
5348                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5349                         }
5350                 } else {
5351                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5352                 }
5353
5354                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5355                         fid_zero(&oic->oic_fid);
5356
5357                         GOTO(out, rc);
5358                 }
5359
5360                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5361                 rc = osd_consistency_check(oti, dev, oic);
5362                 if (rc != 0)
5363                         fid_zero(&oic->oic_fid);
5364         } else {
5365                 rc = PTR_ERR(bh);
5366         }
5367
5368         GOTO(out, rc);
5369
5370 out:
5371         if (hlock != NULL)
5372                 ldiskfs_htree_unlock(hlock);
5373         else
5374                 up_read(&obj->oo_ext_idx_sem);
5375         return rc;
5376 }
5377
5378 static int osd_index_declare_ea_insert(const struct lu_env *env,
5379                                        struct dt_object *dt,
5380                                        const struct dt_rec *rec,
5381                                        const struct dt_key *key,
5382                                        struct thandle *handle)
5383 {
5384         struct osd_thandle      *oh;
5385         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5386         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5387         const struct lu_fid     *fid    = rec1->rec_fid;
5388         int                      credits, rc = 0;
5389         struct osd_idmap_cache  *idc;
5390         ENTRY;
5391
5392         LASSERT(!dt_object_remote(dt));
5393         LASSERT(handle != NULL);
5394         LASSERT(fid != NULL);
5395         LASSERT(rec1->rec_type != 0);
5396
5397         oh = container_of0(handle, struct osd_thandle, ot_super);
5398         LASSERT(oh->ot_handle == NULL);
5399
5400         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5401
5402         /* we can't call iget() while a transactions is running
5403          * (this can lead to a deadlock), but we need to know
5404          * inum and object type. so we find this information at
5405          * declaration and cache in per-thread info */
5406         idc = osd_idc_find_or_init(env, osd, fid);
5407         if (IS_ERR(idc))
5408                 RETURN(PTR_ERR(idc));
5409         if (idc->oic_remote) {
5410                 /* a reference to remote inode is represented by an
5411                  * agent inode which we have to create */
5412                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5413                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5414         }
5415
5416         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5417
5418         if (osd_dt_obj(dt)->oo_inode != NULL) {
5419                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5420
5421                 /* We ignore block quota on meta pool (MDTs), so needn't
5422                  * calculate how many blocks will be consumed by this index
5423                  * insert */
5424                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5425                                            i_gid_read(inode),
5426                                            i_projid_read(inode), 0,
5427                                            oh, osd_dt_obj(dt), true,
5428                                            NULL, false);
5429         }
5430
5431         RETURN(rc);
5432 }
5433
5434 /**
5435  * Index add function for interoperability mode (b11826).
5436  * It will add the directory entry.This entry is needed to
5437  * maintain name->fid mapping.
5438  *
5439  * \param key it is key i.e. file entry to be inserted
5440  * \param rec it is value of given key i.e. fid
5441  *
5442  * \retval   0, on success
5443  * \retval -ve, on error
5444  */
5445 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5446                                const struct dt_rec *rec,
5447                                const struct dt_key *key, struct thandle *th,
5448                                int ignore_quota)
5449 {
5450         struct osd_object       *obj = osd_dt_obj(dt);
5451         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5452         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5453         const struct lu_fid     *fid    = rec1->rec_fid;
5454         const char              *name = (const char *)key;
5455         struct osd_thread_info  *oti   = osd_oti_get(env);
5456         struct inode            *child_inode = NULL;
5457         struct osd_idmap_cache  *idc;
5458         int                     rc;
5459         ENTRY;
5460
5461         if (!dt_object_exists(dt))
5462                 RETURN(-ENOENT);
5463
5464         LASSERT(osd_invariant(obj));
5465         LASSERT(!dt_object_remote(dt));
5466         LASSERT(th != NULL);
5467
5468         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5469
5470         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5471
5472         idc = osd_idc_find(env, osd, fid);
5473         if (unlikely(idc == NULL)) {
5474                 /* this dt_insert() wasn't declared properly, so
5475                  * FID is missing in OI cache. we better do not
5476                  * lookup FID in FLDB/OI and don't risk to deadlock,
5477                  * but in some special cases (lfsck testing, etc)
5478                  * it's much simpler than fixing a caller */
5479                 CERROR("%s: "DFID" wasn't declared for insert\n",
5480                        osd_name(osd), PFID(fid));
5481                 dump_stack();
5482                 idc = osd_idc_find_or_init(env, osd, fid);
5483                 if (IS_ERR(idc))
5484                         RETURN(PTR_ERR(idc));
5485         }
5486
5487         if (idc->oic_remote) {
5488                 /* Insert remote entry */
5489                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5490                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
5491                         struct osd_thandle      *oh;
5492
5493                         /* If parent on remote MDT, we need put this object
5494                          * under AGENT */
5495                         oh = container_of(th, typeof(*oh), ot_super);
5496                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
5497                         if (rc != 0) {
5498                                 CERROR("%s: add "DFID" error: rc = %d\n",
5499                                        osd_name(osd),
5500                                        PFID(lu_object_fid(&dt->do_lu)), rc);
5501                                 RETURN(rc);
5502                         }
5503
5504                         child_inode = igrab(omm->omm_remote_parent->d_inode);
5505                 } else {
5506                         child_inode = osd_create_local_agent_inode(env, osd,
5507                                         obj, fid, rec1->rec_type & S_IFMT, th);
5508                         if (IS_ERR(child_inode))
5509                                 RETURN(PTR_ERR(child_inode));
5510                 }
5511         } else {
5512                 /* Insert local entry */
5513                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5514                         /* for a reason OI cache wasn't filled properly */
5515                         CERROR("%s: OIC for "DFID" isn't filled\n",
5516                                osd_name(osd), PFID(fid));
5517                         RETURN(-EINVAL);
5518                 }
5519                 child_inode = oti->oti_inode;
5520                 if (unlikely(child_inode == NULL)) {
5521                         struct ldiskfs_inode_info *lii;
5522                         OBD_ALLOC_PTR(lii);
5523                         if (lii == NULL)
5524                                 RETURN(-ENOMEM);
5525                         child_inode = oti->oti_inode = &lii->vfs_inode;
5526                 }
5527                 child_inode->i_sb = osd_sb(osd);
5528                 child_inode->i_ino = idc->oic_lid.oii_ino;
5529                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5530         }
5531
5532         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5533
5534         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5535                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5536
5537         if (child_inode && child_inode != oti->oti_inode)
5538                 iput(child_inode);
5539         LASSERT(osd_invariant(obj));
5540         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5541         RETURN(rc);
5542 }
5543
5544 /**
5545  *  Initialize osd Iterator for given osd index object.
5546  *
5547  *  \param  dt      osd index object
5548  */
5549
5550 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5551                                      struct dt_object *dt,
5552                                      __u32 unused)
5553 {
5554         struct osd_it_iam      *it;
5555         struct osd_object      *obj = osd_dt_obj(dt);
5556         struct lu_object       *lo  = &dt->do_lu;
5557         struct iam_path_descr  *ipd;
5558         struct iam_container   *bag = &obj->oo_dir->od_container;
5559
5560         if (!dt_object_exists(dt))
5561                 return ERR_PTR(-ENOENT);
5562
5563         OBD_ALLOC_PTR(it);
5564         if (it == NULL)
5565                 return ERR_PTR(-ENOMEM);
5566
5567         ipd = osd_it_ipd_get(env, bag);
5568         if (likely(ipd != NULL)) {
5569                 it->oi_obj = obj;
5570                 it->oi_ipd = ipd;
5571                 lu_object_get(lo);
5572                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5573                 return (struct dt_it *)it;
5574         } else {
5575                 OBD_FREE_PTR(it);
5576                 return ERR_PTR(-ENOMEM);
5577         }
5578 }
5579
5580 /**
5581  * free given Iterator.
5582  */
5583 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5584 {
5585         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5586         struct osd_object       *obj = it->oi_obj;
5587
5588         iam_it_fini(&it->oi_it);
5589         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5590         osd_object_put(env, obj);
5591         OBD_FREE_PTR(it);
5592 }
5593
5594 /**
5595  *  Move Iterator to record specified by \a key
5596  *
5597  *  \param  di      osd iterator
5598  *  \param  key     key for index
5599  *
5600  *  \retval +ve  di points to record with least key not larger than key
5601  *  \retval  0   di points to exact matched key
5602  *  \retval -ve  failure
5603  */
5604
5605 static int osd_it_iam_get(const struct lu_env *env,
5606                           struct dt_it *di, const struct dt_key *key)
5607 {
5608         struct osd_thread_info  *oti = osd_oti_get(env);
5609         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5610
5611         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5612                 /* swab quota uid/gid */
5613                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5614                 key = (struct dt_key *)&oti->oti_quota_id;
5615         }
5616
5617         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5618 }
5619
5620 /**
5621  *  Release Iterator
5622  *
5623  *  \param  di      osd iterator
5624  */
5625 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5626 {
5627         struct osd_it_iam *it = (struct osd_it_iam *)di;
5628
5629         iam_it_put(&it->oi_it);
5630 }
5631
5632 /**
5633  *  Move iterator by one record
5634  *
5635  *  \param  di      osd iterator
5636  *
5637  *  \retval +1   end of container reached
5638  *  \retval  0   success
5639  *  \retval -ve  failure
5640  */
5641
5642 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5643 {
5644         struct osd_it_iam *it = (struct osd_it_iam *)di;
5645
5646         return iam_it_next(&it->oi_it);
5647 }
5648
5649 /**
5650  * Return pointer to the key under iterator.
5651  */
5652
5653 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5654                                  const struct dt_it *di)
5655 {
5656         struct osd_thread_info *oti = osd_oti_get(env);
5657         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5658         struct osd_object      *obj = it->oi_obj;
5659         struct dt_key          *key;
5660
5661         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5662
5663         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5664                 /* swab quota uid/gid */
5665                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5666                 key = (struct dt_key *)&oti->oti_quota_id;
5667         }
5668
5669         return key;
5670 }
5671
5672 /**
5673  * Return size of key under iterator (in bytes)
5674  */
5675
5676 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5677 {
5678         struct osd_it_iam *it = (struct osd_it_iam *)di;
5679
5680         return iam_it_key_size(&it->oi_it);
5681 }
5682
5683 static inline void
5684 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5685 {
5686         /* check if file type is required */
5687         if (ent->lde_attrs & LUDA_TYPE) {
5688                 struct luda_type *lt;
5689                 int align = sizeof(*lt) - 1;
5690
5691                 len = (len + align) & ~align;
5692                 lt = (struct luda_type *)(ent->lde_name + len);
5693                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5694         }
5695
5696         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5697 }
5698
5699 /**
5700  * build lu direct from backend fs dirent.
5701  */
5702
5703 static inline void
5704 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5705                    char *name, __u16 namelen, __u16 type, __u32 attr)
5706 {
5707         ent->lde_attrs = attr | LUDA_FID;
5708         fid_cpu_to_le(&ent->lde_fid, fid);
5709
5710         ent->lde_hash = cpu_to_le64(offset);
5711         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5712
5713         strncpy(ent->lde_name, name, namelen);
5714         ent->lde_name[namelen] = '\0';
5715         ent->lde_namelen = cpu_to_le16(namelen);
5716
5717         /* append lustre attributes */
5718         osd_it_append_attrs(ent, namelen, type);
5719 }
5720
5721 /**
5722  * Return pointer to the record under iterator.
5723  */
5724 static int osd_it_iam_rec(const struct lu_env *env,
5725                           const struct dt_it *di,
5726                           struct dt_rec *dtrec, __u32 attr)
5727 {
5728         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5729         struct osd_thread_info *info = osd_oti_get(env);
5730         ENTRY;
5731
5732         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5733                 const struct osd_fid_pack *rec;
5734                 struct lu_fid             *fid = &info->oti_fid;
5735                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5736                 char                      *name;
5737                 int                        namelen;
5738                 __u64                      hash;
5739                 int                        rc;
5740
5741                 name = (char *)iam_it_key_get(&it->oi_it);
5742                 if (IS_ERR(name))
5743                         RETURN(PTR_ERR(name));
5744
5745                 namelen = iam_it_key_size(&it->oi_it);
5746
5747                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5748                 if (IS_ERR(rec))
5749                         RETURN(PTR_ERR(rec));
5750
5751                 rc = osd_fid_unpack(fid, rec);
5752                 if (rc)
5753                         RETURN(rc);
5754
5755                 hash = iam_it_store(&it->oi_it);
5756
5757                 /* IAM does not store object type in IAM index (dir) */
5758                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5759                                    0, LUDA_FID);
5760         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5761                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5762                            (struct iam_rec *)dtrec);
5763                 osd_quota_unpack(it->oi_obj, dtrec);
5764         } else {
5765                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5766                            (struct iam_rec *)dtrec);
5767         }
5768
5769         RETURN(0);
5770 }
5771
5772 /**
5773  * Returns cookie for current Iterator position.
5774  */
5775 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5776 {
5777         struct osd_it_iam *it = (struct osd_it_iam *)di;
5778
5779         return iam_it_store(&it->oi_it);
5780 }
5781
5782 /**
5783  * Restore iterator from cookie.
5784  *
5785  * \param  di      osd iterator
5786  * \param  hash    Iterator location cookie
5787  *
5788  * \retval +ve  di points to record with least key not larger than key.
5789  * \retval  0   di points to exact matched key
5790  * \retval -ve  failure
5791  */
5792
5793 static int osd_it_iam_load(const struct lu_env *env,
5794                            const struct dt_it *di, __u64 hash)
5795 {
5796         struct osd_it_iam *it = (struct osd_it_iam *)di;
5797
5798         return iam_it_load(&it->oi_it, hash);
5799 }
5800
5801 static const struct dt_index_operations osd_index_iam_ops = {
5802         .dio_lookup         = osd_index_iam_lookup,
5803         .dio_declare_insert = osd_index_declare_iam_insert,
5804         .dio_insert         = osd_index_iam_insert,
5805         .dio_declare_delete = osd_index_declare_iam_delete,
5806         .dio_delete         = osd_index_iam_delete,
5807         .dio_it     = {
5808                 .init     = osd_it_iam_init,
5809                 .fini     = osd_it_iam_fini,
5810                 .get      = osd_it_iam_get,
5811                 .put      = osd_it_iam_put,
5812                 .next     = osd_it_iam_next,
5813                 .key      = osd_it_iam_key,
5814                 .key_size = osd_it_iam_key_size,
5815                 .rec      = osd_it_iam_rec,
5816                 .store    = osd_it_iam_store,
5817                 .load     = osd_it_iam_load
5818         }
5819 };
5820
5821
5822 /**
5823  * Creates or initializes iterator context.
5824  *
5825  * \retval struct osd_it_ea, iterator structure on success
5826  *
5827  */
5828 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5829                                     struct dt_object *dt,
5830                                     __u32 attr)
5831 {
5832         struct osd_object       *obj  = osd_dt_obj(dt);
5833         struct osd_thread_info  *info = osd_oti_get(env);
5834         struct osd_it_ea        *oie;
5835         struct file             *file;
5836         struct lu_object        *lo   = &dt->do_lu;
5837         struct dentry           *obj_dentry;
5838         ENTRY;
5839
5840         if (!dt_object_exists(dt) || obj->oo_destroyed)
5841                 RETURN(ERR_PTR(-ENOENT));
5842
5843         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5844         if (oie == NULL)
5845                 RETURN(ERR_PTR(-ENOMEM));
5846         obj_dentry = &oie->oie_dentry;
5847
5848         obj_dentry->d_inode = obj->oo_inode;
5849         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5850         obj_dentry->d_name.hash = 0;
5851
5852         oie->oie_rd_dirent       = 0;
5853         oie->oie_it_dirent       = 0;
5854         oie->oie_dirent          = NULL;
5855         if (unlikely(!info->oti_it_ea_buf_used)) {
5856                 oie->oie_buf = info->oti_it_ea_buf;
5857                 info->oti_it_ea_buf_used = 1;
5858         } else {
5859                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5860                 if (oie->oie_buf == NULL)
5861                         RETURN(ERR_PTR(-ENOMEM));
5862         }
5863         oie->oie_obj             = obj;
5864
5865         file = &oie->oie_file;
5866
5867         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5868         if (attr & LUDA_64BITHASH)
5869                 file->f_mode    = FMODE_64BITHASH;
5870         else
5871                 file->f_mode    = FMODE_32BITHASH;
5872         file->f_path.dentry     = obj_dentry;
5873         file->f_mapping         = obj->oo_inode->i_mapping;
5874         file->f_op              = obj->oo_inode->i_fop;
5875         set_file_inode(file, obj->oo_inode);
5876
5877         lu_object_get(lo);
5878         RETURN((struct dt_it *) oie);
5879 }
5880
5881 /**
5882  * Destroy or finishes iterator context.
5883  *
5884  * \param di iterator structure to be destroyed
5885  */
5886 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5887 {
5888         struct osd_thread_info  *info = osd_oti_get(env);
5889         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
5890         struct osd_object       *obj    = oie->oie_obj;
5891         struct inode            *inode  = obj->oo_inode;
5892
5893         ENTRY;
5894         oie->oie_file.f_op->release(inode, &oie->oie_file);
5895         osd_object_put(env, obj);
5896         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5897                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5898         else
5899                 info->oti_it_ea_buf_used = 0;
5900         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5901         EXIT;
5902 }
5903
5904 /**
5905  * It position the iterator at given key, so that next lookup continues from
5906  * that key Or it is similar to dio_it->load() but based on a key,
5907  * rather than file position.
5908  *
5909  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5910  * to the beginning.
5911  *
5912  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5913  */
5914 static int osd_it_ea_get(const struct lu_env *env,
5915                          struct dt_it *di, const struct dt_key *key)
5916 {
5917         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
5918
5919         ENTRY;
5920         LASSERT(((const char *)key)[0] == '\0');
5921         it->oie_file.f_pos      = 0;
5922         it->oie_rd_dirent       = 0;
5923         it->oie_it_dirent       = 0;
5924         it->oie_dirent          = NULL;
5925
5926         RETURN(+1);
5927 }
5928
5929 /**
5930  * Does nothing
5931  */
5932 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5933 {
5934 }
5935
5936 struct osd_filldir_cbs {
5937 #ifdef HAVE_DIR_CONTEXT
5938         struct dir_context ctx;
5939 #endif
5940         struct osd_it_ea  *it;
5941 };
5942 /**
5943  * It is called internally by ->readdir(). It fills the
5944  * iterator's in-memory data structure with required
5945  * information i.e. name, namelen, rec_size etc.
5946  *
5947  * \param buf in which information to be filled in.
5948  * \param name name of the file in given dir
5949  *
5950  * \retval 0 on success
5951  * \retval 1 on buffer full
5952  */
5953 #ifdef HAVE_FILLDIR_USE_CTX
5954 static int osd_ldiskfs_filldir(struct dir_context *buf,
5955 #else
5956 static int osd_ldiskfs_filldir(void *buf,
5957 #endif
5958                                const char *name, int namelen,
5959                                loff_t offset, __u64 ino, unsigned d_type)
5960 {
5961         struct osd_it_ea        *it   =
5962                 ((struct osd_filldir_cbs *)buf)->it;
5963         struct osd_object       *obj  = it->oie_obj;
5964         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5965         struct lu_fid           *fid  = &ent->oied_fid;
5966         struct osd_fid_pack     *rec;
5967         ENTRY;
5968
5969         /* this should never happen */
5970         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5971                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5972                 RETURN(-EIO);
5973         }
5974
5975         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5976             OSD_IT_EA_BUFSIZE)
5977                 RETURN(1);
5978
5979         /* "." is just the object itself. */
5980         if (namelen == 1 && name[0] == '.') {
5981                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5982         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5983                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5984                 if (osd_fid_unpack(fid, rec) != 0)
5985                         fid_zero(fid);
5986         } else {
5987                 fid_zero(fid);
5988         }
5989         d_type &= ~LDISKFS_DIRENT_LUFID;
5990
5991         /* NOT export local root. */
5992         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5993                 ino = obj->oo_inode->i_ino;
5994                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5995         }
5996
5997         ent->oied_ino     = ino;
5998         ent->oied_off     = offset;
5999         ent->oied_namelen = namelen;
6000         ent->oied_type    = d_type;
6001
6002         memcpy(ent->oied_name, name, namelen);
6003
6004         it->oie_rd_dirent++;
6005         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6006         RETURN(0);
6007 }
6008
6009 /**
6010  * Calls ->readdir() to load a directory entry at a time
6011  * and stored it in iterator's in-memory data structure.
6012  *
6013  * \param di iterator's in memory structure
6014  *
6015  * \retval   0 on success
6016  * \retval -ve on error
6017  * \retval +1 reach the end of entry
6018  */
6019 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6020                                const struct dt_it *di)
6021 {
6022         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
6023         struct osd_object  *obj   = it->oie_obj;
6024         struct inode       *inode = obj->oo_inode;
6025         struct htree_lock  *hlock = NULL;
6026         struct file        *filp  = &it->oie_file;
6027         int                 rc = 0;
6028         struct osd_filldir_cbs buf = {
6029 #ifdef HAVE_DIR_CONTEXT
6030                 .ctx.actor = osd_ldiskfs_filldir,
6031 #endif
6032                 .it = it
6033         };
6034
6035         ENTRY;
6036         it->oie_dirent = it->oie_buf;
6037         it->oie_rd_dirent = 0;
6038
6039         if (obj->oo_hl_head != NULL) {
6040                 hlock = osd_oti_get(env)->oti_hlock;
6041                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6042                                    inode, LDISKFS_HLOCK_READDIR);
6043         } else {
6044                 down_read(&obj->oo_ext_idx_sem);
6045         }
6046
6047 #ifdef HAVE_DIR_CONTEXT
6048         buf.ctx.pos = filp->f_pos;
6049         rc = inode->i_fop->iterate(filp, &buf.ctx);
6050         filp->f_pos = buf.ctx.pos;
6051 #else
6052         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6053 #endif
6054
6055         if (hlock != NULL)
6056                 ldiskfs_htree_unlock(hlock);
6057         else
6058                 up_read(&obj->oo_ext_idx_sem);
6059
6060         if (it->oie_rd_dirent == 0) {
6061                 /*If it does not get any dirent, it means it has been reached
6062                  *to the end of the dir */
6063                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6064                 if (rc == 0)
6065                         rc = 1;
6066         } else {
6067                 it->oie_dirent = it->oie_buf;
6068                 it->oie_it_dirent = 1;
6069         }
6070
6071         RETURN(rc);
6072 }
6073
6074 /**
6075  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6076  * to load a directory entry at a time and stored it in
6077  * iterator's in-memory data structure.
6078  *
6079  * \param di iterator's in memory structure
6080  *
6081  * \retval +ve iterator reached to end
6082  * \retval   0 iterator not reached to end
6083  * \retval -ve on error
6084  */
6085 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6086 {
6087         struct osd_it_ea *it = (struct osd_it_ea *)di;
6088         int rc;
6089
6090         ENTRY;
6091
6092         if (it->oie_it_dirent < it->oie_rd_dirent) {
6093                 it->oie_dirent =
6094                         (void *) it->oie_dirent +
6095                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6096                                        it->oie_dirent->oied_namelen);
6097                 it->oie_it_dirent++;
6098                 RETURN(0);
6099         } else {
6100                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6101                         rc = +1;
6102                 else
6103                         rc = osd_ldiskfs_it_fill(env, di);
6104         }
6105
6106         RETURN(rc);
6107 }
6108
6109 /**
6110  * Returns the key at current position from iterator's in memory structure.
6111  *
6112  * \param di iterator's in memory structure
6113  *
6114  * \retval key i.e. struct dt_key on success
6115  */
6116 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6117                                     const struct dt_it *di)
6118 {
6119         struct osd_it_ea *it = (struct osd_it_ea *)di;
6120
6121         return (struct dt_key *)it->oie_dirent->oied_name;
6122 }
6123
6124 /**
6125  * Returns the key's size at current position from iterator's in memory structure.
6126  *
6127  * \param di iterator's in memory structure
6128  *
6129  * \retval key_size i.e. struct dt_key on success
6130  */
6131 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6132 {
6133         struct osd_it_ea *it = (struct osd_it_ea *)di;
6134
6135         return it->oie_dirent->oied_namelen;
6136 }
6137
6138 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6139 {
6140         if (LDISKFS_DIR_REC_LEN(de) >=
6141             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6142                 return true;
6143
6144         return false;
6145 }
6146
6147 static inline bool
6148 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6149                      unsigned blocksize, bool dotdot)
6150 {
6151         if (dotdot)
6152                 return osd_dotdot_has_space(de);
6153
6154         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6155             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6156                 return true;
6157
6158         return false;
6159 }
6160
6161 static int
6162 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6163                     handle_t *jh, struct dentry *dentry,
6164                     const struct lu_fid *fid, struct buffer_head *bh,
6165                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6166                     bool dotdot)
6167 {
6168         struct inode                *dir        = dentry->d_parent->d_inode;
6169         struct inode                *inode      = dentry->d_inode;
6170         struct osd_fid_pack         *rec;
6171         struct ldiskfs_dentry_param *ldp;
6172         int                          namelen    = dentry->d_name.len;
6173         int                          rc;
6174         struct osd_thread_info     *info        = osd_oti_get(env);
6175         ENTRY;
6176
6177         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6178                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6179                 RETURN(0);
6180
6181         /* There is enough space to hold the FID-in-dirent. */
6182         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6183                 rc = ldiskfs_journal_get_write_access(jh, bh);
6184                 if (rc != 0)
6185                         RETURN(rc);
6186
6187                 de->name[namelen] = 0;
6188                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6189                 rec->fp_len = sizeof(struct lu_fid) + 1;
6190                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6191                 de->file_type |= LDISKFS_DIRENT_LUFID;
6192                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6193
6194                 RETURN(rc);
6195         }
6196
6197         LASSERT(!dotdot);
6198
6199         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6200         if (rc != 0)
6201                 RETURN(rc);
6202
6203         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6204         osd_get_ldiskfs_dirent_param(ldp, fid);
6205         dentry->d_fsdata = (void *)ldp;
6206         ll_vfs_dq_init(dir);
6207         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6208         /* It is too bad, we cannot reinsert the name entry back.
6209          * That means we lose it! */
6210         if (rc != 0)
6211                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6212                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6213                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6214                        dir->i_ino, dir->i_generation, namelen,
6215                        dentry->d_name.name, PFID(fid), rc);
6216
6217         RETURN(rc);
6218 }
6219
6220 static int
6221 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6222                         struct osd_it_ea *it, struct lu_fid *fid,
6223                         struct osd_inode_id *id, __u32 *attr)
6224 {
6225         struct osd_thread_info     *info        = osd_oti_get(env);
6226         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6227         struct osd_device          *dev         = osd_obj2dev(obj);
6228         struct super_block         *sb          = osd_sb(dev);
6229         const char                 *devname     = osd_name(dev);
6230         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6231         struct inode               *dir         = obj->oo_inode;
6232         struct htree_lock          *hlock       = NULL;
6233         struct buffer_head         *bh          = NULL;
6234         handle_t                   *jh          = NULL;
6235         struct ldiskfs_dir_entry_2 *de;
6236         struct dentry              *dentry;
6237         struct inode               *inode;
6238         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6239         int                         credits;
6240         int                         rc;
6241         bool                        dotdot      = false;
6242         bool                        dirty       = false;
6243         ENTRY;
6244
6245         if (ent->oied_name[0] == '.') {
6246                 if (ent->oied_namelen == 1)
6247                         RETURN(0);
6248
6249                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6250                         dotdot = true;
6251         }
6252
6253         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6254         inode = osd_iget(info, dev, id);
6255         if (IS_ERR(inode)) {
6256                 rc = PTR_ERR(inode);
6257                 if (rc == -ENOENT || rc == -ESTALE) {
6258                         /* Maybe dangling name entry, or
6259                          * corrupted directory entry. */
6260                         *attr |= LUDA_UNKNOWN;
6261                         rc = 0;
6262                 } else {
6263                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6264                                "check_repair, dir = %lu/%u, name = %.*s, "
6265                                "ino = %llu, rc = %d\n",
6266                                devname, dir->i_ino, dir->i_generation,
6267                                ent->oied_namelen, ent->oied_name,
6268                                ent->oied_ino, rc);
6269                 }
6270
6271                 RETURN(rc);
6272         }
6273
6274         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6275                                            ent->oied_namelen);
6276         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6277         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6278                 lma = NULL;
6279         else if (rc != 0)
6280                 GOTO(out, rc);
6281
6282         /* We need to ensure that the name entry is still valid.
6283          * Because it may be removed or renamed by other already.
6284          *
6285          * The unlink or rename operation will start journal before PDO lock,
6286          * so to avoid deadlock, here we need to start journal handle before
6287          * related PDO lock also. But because we do not know whether there
6288          * will be something to be repaired before PDO lock, we just start
6289          * journal without conditions.
6290          *
6291          * We may need to remove the name entry firstly, then insert back.
6292          * One credit is for user quota file update.
6293          * One credit is for group quota file update.
6294          * Two credits are for dirty inode. */
6295         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6296                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6297
6298         if (dev->od_dirent_journal != 0) {
6299
6300 again:
6301                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6302                 if (IS_ERR(jh)) {
6303                         rc = PTR_ERR(jh);
6304                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6305                                "check_repair, dir = %lu/%u, credits = %d, "
6306                                "name = %.*s, ino = %llu: rc = %d\n",
6307                                devname, dir->i_ino, dir->i_generation, credits,
6308                                ent->oied_namelen, ent->oied_name,
6309                                ent->oied_ino, rc);
6310
6311                         GOTO(out_inode, rc);
6312                 }
6313
6314                 if (obj->oo_hl_head != NULL) {
6315                         hlock = osd_oti_get(env)->oti_hlock;
6316                         /* "0" means exclusive lock for the whole directory.
6317                          * We need to prevent others access such name entry
6318                          * during the delete + insert. Neither HLOCK_ADD nor
6319                          * HLOCK_DEL cannot guarantee the atomicity. */
6320                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6321                 } else {
6322                         down_write(&obj->oo_ext_idx_sem);
6323                 }
6324         } else {
6325                 if (obj->oo_hl_head != NULL) {
6326                         hlock = osd_oti_get(env)->oti_hlock;
6327                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6328                                            LDISKFS_HLOCK_LOOKUP);
6329                 } else {
6330                         down_read(&obj->oo_ext_idx_sem);
6331                 }
6332         }
6333
6334         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6335         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6336                 *attr |= LUDA_IGNORE;
6337
6338                 GOTO(out, rc = 0);
6339         }
6340
6341         /* For dotdot entry, if there is not enough space to hold the
6342          * FID-in-dirent, just keep them there. It only happens when the
6343          * device upgraded from 1.8 or restored from MDT file-level backup.
6344          * For the whole directory, only dotdot entry have no FID-in-dirent
6345          * and needs to get FID from LMA when readdir, it will not affect the
6346          * performance much. */
6347         if (dotdot && !osd_dotdot_has_space(de)) {
6348                 *attr |= LUDA_UNKNOWN;
6349
6350                 GOTO(out, rc = 0);
6351         }
6352
6353         if (lma != NULL) {
6354                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6355                         GOTO(out, rc = 0);
6356
6357                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6358                         struct lu_fid *tfid = &lma->lma_self_fid;
6359
6360                         if (likely(dotdot &&
6361                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6362                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6363                                 /* It must be REMOTE_PARENT_DIR and as the
6364                                  * 'dotdot' entry of remote directory */
6365                                 *attr |= LUDA_IGNORE;
6366                         } else {
6367                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6368                                        "parent directory, but got %.*s under "
6369                                        "dir = %lu/%u with the FID "DFID"\n",
6370                                        devname, ent->oied_namelen,
6371                                        ent->oied_name, dir->i_ino,
6372                                        dir->i_generation, PFID(tfid));
6373
6374                                 *attr |= LUDA_UNKNOWN;
6375                         }
6376
6377                         GOTO(out, rc = 0);
6378                 }
6379         }
6380
6381         if (!fid_is_zero(fid)) {
6382                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6383                                               ent->oied_namelen);
6384                 if (rc == -ENOENT ||
6385                     (rc == -ENODATA &&
6386                      !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6387                         /* linkEA does not recognize the dirent entry,
6388                          * it may because the dirent entry corruption
6389                          * and points to other's inode. */
6390                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6391                                "recognize the dirent, dir = %lu/%u, "
6392                                " name = %.*s, ino = %llu, "
6393                                DFID": rc = %d\n", devname, dir->i_ino,
6394                                dir->i_generation, ent->oied_namelen,
6395                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6396                         *attr |= LUDA_UNKNOWN;
6397
6398                         GOTO(out, rc = 0);
6399                 }
6400
6401                 if (rc && rc != -ENODATA) {
6402                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6403                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6404                                DFID": rc = %d\n", devname, dir->i_ino,
6405                                dir->i_generation, ent->oied_namelen,
6406                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6407                         *attr |= LUDA_UNKNOWN;
6408
6409                         GOTO(out, rc = 0);
6410                 }
6411         }
6412
6413         if (lma != NULL) {
6414                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6415                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6416                 if (*attr & LUDA_VERIFY_DRYRUN) {
6417                         *fid = lma->lma_self_fid;
6418                         *attr |= LUDA_REPAIR;
6419
6420                         GOTO(out, rc = 0);
6421                 }
6422
6423                 if (jh == NULL) {
6424                         brelse(bh);
6425                         dev->od_dirent_journal = 1;
6426                         if (hlock != NULL) {
6427                                 ldiskfs_htree_unlock(hlock);
6428                                 hlock = NULL;
6429                         } else {
6430                                 up_read(&obj->oo_ext_idx_sem);
6431                         }
6432
6433                         goto again;
6434                 }
6435
6436                 *fid = lma->lma_self_fid;
6437                 dirty = true;
6438                 /* Update or append the FID-in-dirent. */
6439                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6440                                          bh, de, hlock, dotdot);
6441                 if (rc == 0)
6442                         *attr |= LUDA_REPAIR;
6443                 else
6444                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6445                                "the dirent, dir = %lu/%u, name = %.*s, "
6446                                "ino = %llu, "DFID": rc = %d\n",
6447                                devname, dir->i_ino, dir->i_generation,
6448                                ent->oied_namelen, ent->oied_name,
6449                                ent->oied_ino, PFID(fid), rc);
6450         } else {
6451                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6452                 if (*attr & LUDA_VERIFY_DRYRUN) {
6453                         if (fid_is_sane(fid)) {
6454                                 *attr |= LUDA_REPAIR;
6455                         } else if (dev->od_index == 0) {
6456                                 lu_igif_build(fid, inode->i_ino,
6457                                               inode->i_generation);
6458                                 *attr |= LUDA_UPGRADE;
6459                         }
6460
6461                         GOTO(out, rc = 0);
6462                 }
6463
6464                 if (jh == NULL) {
6465                         brelse(bh);
6466                         dev->od_dirent_journal = 1;
6467                         if (hlock != NULL) {
6468                                 ldiskfs_htree_unlock(hlock);
6469                                 hlock = NULL;
6470                         } else {
6471                                 up_read(&obj->oo_ext_idx_sem);
6472                         }
6473
6474                         goto again;
6475                 }
6476
6477                 dirty = true;
6478                 if (unlikely(fid_is_sane(fid))) {
6479                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6480                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6481                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6482                         if (rc == 0)
6483                                 *attr |= LUDA_REPAIR;
6484                         else
6485                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6486                                        "update dirent, dir = %lu/%u, "
6487                                        "name = %.*s, ino = %llu, "
6488                                        DFID": rc = %d\n",
6489                                        devname, dir->i_ino, dir->i_generation,
6490                                        ent->oied_namelen, ent->oied_name,
6491                                        ent->oied_ino, PFID(fid), rc);
6492                 } else if (dev->od_index == 0) {
6493                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6494                         /* It is probably IGIF object. Only aappend the
6495                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6496                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6497                                                  bh, de, hlock, dotdot);
6498                         if (rc == 0)
6499                                 *attr |= LUDA_UPGRADE;
6500                         else
6501                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6502                                        "after the dirent, dir = %lu/%u, "
6503                                        "name = %.*s, ino = %llu, "
6504                                        DFID": rc = %d\n",
6505                                        devname, dir->i_ino, dir->i_generation,
6506                                        ent->oied_namelen, ent->oied_name,
6507                                        ent->oied_ino, PFID(fid), rc);
6508                 }
6509         }
6510
6511         GOTO(out, rc);
6512
6513 out:
6514         if (!IS_ERR(bh))
6515                 brelse(bh);
6516         if (hlock != NULL) {
6517                 ldiskfs_htree_unlock(hlock);
6518         } else {
6519                 if (dev->od_dirent_journal != 0)
6520                         up_write(&obj->oo_ext_idx_sem);
6521                 else
6522                         up_read(&obj->oo_ext_idx_sem);
6523         }
6524
6525         if (jh != NULL)
6526                 ldiskfs_journal_stop(jh);
6527
6528 out_inode:
6529         iput(inode);
6530         if (rc >= 0 && !dirty)
6531                 dev->od_dirent_journal = 0;
6532
6533         return rc;
6534 }
6535
6536 /**
6537  * Returns the value at current position from iterator's in memory structure.
6538  *
6539  * \param di struct osd_it_ea, iterator's in memory structure
6540  * \param attr attr requested for dirent.
6541  * \param lde lustre dirent
6542  *
6543  * \retval   0 no error and \param lde has correct lustre dirent.
6544  * \retval -ve on error
6545  */
6546 static inline int osd_it_ea_rec(const struct lu_env *env,
6547                                 const struct dt_it *di,
6548                                 struct dt_rec *dtrec, __u32 attr)
6549 {
6550         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6551         struct osd_object      *obj   = it->oie_obj;
6552         struct osd_device      *dev   = osd_obj2dev(obj);
6553         struct osd_thread_info *oti   = osd_oti_get(env);
6554         struct osd_inode_id    *id    = &oti->oti_id;
6555         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6556         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6557         __u32                   ino   = it->oie_dirent->oied_ino;
6558         int                     rc    = 0;
6559         ENTRY;
6560
6561         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6562
6563         if (attr & LUDA_VERIFY) {
6564                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6565                         attr |= LUDA_IGNORE;
6566                         /* If the parent is on remote MDT, and there
6567                          * is no FID-in-dirent, then we have to get
6568                          * the parent FID from the linkEA.  */
6569                         if (!fid_is_sane(fid) &&
6570                             it->oie_dirent->oied_namelen == 2 &&
6571                             it->oie_dirent->oied_name[0] == '.' &&
6572                             it->oie_dirent->oied_name[1] == '.')
6573                                 osd_get_pfid_from_linkea(env, obj, fid);
6574                 } else {
6575                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6576                                                      &attr);
6577                 }
6578
6579                 if (!fid_is_sane(fid))
6580                         attr |= LUDA_UNKNOWN;
6581         } else {
6582                 attr &= ~LU_DIRENT_ATTRS_MASK;
6583                 if (!fid_is_sane(fid)) {
6584                         bool is_dotdot = false;
6585                         if (it->oie_dirent->oied_namelen == 2 &&
6586                             it->oie_dirent->oied_name[0] == '.' &&
6587                             it->oie_dirent->oied_name[1] == '.')
6588                                 is_dotdot = true;
6589                         /* If the parent is on remote MDT, and there
6590                          * is no FID-in-dirent, then we have to get
6591                          * the parent FID from the linkEA.  */
6592                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6593                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6594                         } else {
6595                                 if (is_dotdot == false &&
6596                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6597                                         RETURN(-ENOENT);
6598
6599                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6600                         }
6601                 } else {
6602                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6603                 }
6604         }
6605
6606         /* Pack the entry anyway, at least the offset is right. */
6607         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6608                            it->oie_dirent->oied_name,
6609                            it->oie_dirent->oied_namelen,
6610                            it->oie_dirent->oied_type, attr);
6611
6612         if (rc < 0)
6613                 RETURN(rc);
6614
6615         if (osd_remote_fid(env, dev, fid))
6616                 RETURN(0);
6617
6618         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6619                 osd_add_oi_cache(oti, dev, id, fid);
6620
6621         RETURN(rc > 0 ? 0 : rc);
6622 }
6623
6624 /**
6625  * Returns the record size size at current position.
6626  *
6627  * This function will return record(lu_dirent) size in bytes.
6628  *
6629  * \param[in] env       execution environment
6630  * \param[in] di        iterator's in memory structure
6631  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6632  *                      calculate the lu_dirent size.
6633  *
6634  * \retval      record size(in bytes & in memory) of the current lu_dirent
6635  *              entry.
6636  */
6637 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6638                               __u32 attr)
6639 {
6640         struct osd_it_ea *it = (struct osd_it_ea *)di;
6641
6642         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6643 }
6644
6645 /**
6646  * Returns a cookie for current position of the iterator head, so that
6647  * user can use this cookie to load/start the iterator next time.
6648  *
6649  * \param di iterator's in memory structure
6650  *
6651  * \retval cookie for current position, on success
6652  */
6653 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6654 {
6655         struct osd_it_ea *it = (struct osd_it_ea *)di;
6656
6657         return it->oie_dirent->oied_off;
6658 }
6659
6660 /**
6661  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6662  * to load a directory entry at a time and stored it i inn,
6663  * in iterator's in-memory data structure.
6664  *
6665  * \param di struct osd_it_ea, iterator's in memory structure
6666  *
6667  * \retval +ve on success
6668  * \retval -ve on error
6669  */
6670 static int osd_it_ea_load(const struct lu_env *env,
6671                           const struct dt_it *di, __u64 hash)
6672 {
6673         struct osd_it_ea *it = (struct osd_it_ea *)di;
6674         int rc;
6675
6676         ENTRY;
6677         it->oie_file.f_pos = hash;
6678
6679         rc =  osd_ldiskfs_it_fill(env, di);
6680         if (rc > 0)
6681                 rc = -ENODATA;
6682
6683         if (rc == 0)
6684                 rc = +1;
6685
6686         RETURN(rc);
6687 }
6688
6689 /**
6690  * Index lookup function for interoperability mode (b11826).
6691  *
6692  * \param key,  key i.e. file name to be searched
6693  *
6694  * \retval +ve, on success
6695  * \retval -ve, on error
6696  */
6697 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6698                                struct dt_rec *rec, const struct dt_key *key)
6699 {
6700         struct osd_object *obj = osd_dt_obj(dt);
6701         int rc = 0;
6702
6703         ENTRY;
6704
6705         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6706         LINVRNT(osd_invariant(obj));
6707
6708         rc = osd_ea_lookup_rec(env, obj, rec, key);
6709         if (rc == 0)
6710                 rc = +1;
6711         RETURN(rc);
6712 }
6713
6714 /**
6715  * Index and Iterator operations for interoperability
6716  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6717  */
6718 static const struct dt_index_operations osd_index_ea_ops = {
6719         .dio_lookup         = osd_index_ea_lookup,
6720         .dio_declare_insert = osd_index_declare_ea_insert,
6721         .dio_insert         = osd_index_ea_insert,
6722         .dio_declare_delete = osd_index_declare_ea_delete,
6723         .dio_delete         = osd_index_ea_delete,
6724         .dio_it     = {
6725                 .init     = osd_it_ea_init,
6726                 .fini     = osd_it_ea_fini,
6727                 .get      = osd_it_ea_get,
6728                 .put      = osd_it_ea_put,
6729                 .next     = osd_it_ea_next,
6730                 .key      = osd_it_ea_key,
6731                 .key_size = osd_it_ea_key_size,
6732                 .rec      = osd_it_ea_rec,
6733                 .rec_size = osd_it_ea_rec_size,
6734                 .store    = osd_it_ea_store,
6735                 .load     = osd_it_ea_load
6736         }
6737 };
6738
6739 static void *osd_key_init(const struct lu_context *ctx,
6740                           struct lu_context_key *key)
6741 {
6742         struct osd_thread_info *info;
6743
6744         OBD_ALLOC_PTR(info);
6745         if (info == NULL)
6746                 return ERR_PTR(-ENOMEM);
6747
6748         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6749         if (info->oti_it_ea_buf == NULL)
6750                 goto out_free_info;
6751
6752         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6753
6754         info->oti_hlock = ldiskfs_htree_lock_alloc();
6755         if (info->oti_hlock == NULL)
6756                 goto out_free_ea;
6757
6758         return info;
6759
6760  out_free_ea:
6761         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6762  out_free_info:
6763         OBD_FREE_PTR(info);
6764         return ERR_PTR(-ENOMEM);
6765 }
6766
6767 static void osd_key_fini(const struct lu_context *ctx,
6768                          struct lu_context_key *key, void* data)
6769 {
6770         struct osd_thread_info *info = data;
6771         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6772         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6773
6774         if (info->oti_inode != NULL)
6775                 OBD_FREE_PTR(lli);
6776         if (info->oti_hlock != NULL)
6777                 ldiskfs_htree_lock_free(info->oti_hlock);
6778         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6779         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6780         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6781         lu_buf_free(&info->oti_big_buf);
6782         if (idc != NULL) {
6783                 LASSERT(info->oti_ins_cache_size > 0);
6784                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6785                 info->oti_ins_cache = NULL;
6786                 info->oti_ins_cache_size = 0;
6787         }
6788         OBD_FREE_PTR(info);
6789 }
6790
6791 static void osd_key_exit(const struct lu_context *ctx,
6792                          struct lu_context_key *key, void *data)
6793 {
6794         struct osd_thread_info *info = data;
6795
6796         LASSERT(info->oti_r_locks == 0);
6797         LASSERT(info->oti_w_locks == 0);
6798         LASSERT(info->oti_txns    == 0);
6799 }
6800
6801 /* type constructor/destructor: osd_type_init, osd_type_fini */
6802 LU_TYPE_INIT_FINI(osd, &osd_key);
6803
6804 struct lu_context_key osd_key = {
6805         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6806         .lct_init = osd_key_init,
6807         .lct_fini = osd_key_fini,
6808         .lct_exit = osd_key_exit
6809 };
6810
6811
6812 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6813                            const char *name, struct lu_device *next)
6814 {
6815         struct osd_device *osd = osd_dev(d);
6816
6817         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6818             >= sizeof(osd->od_svname))
6819                 return -E2BIG;
6820         return osd_procfs_init(osd, name);
6821 }
6822
6823 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6824 {
6825         struct seq_server_site  *ss = osd_seq_site(osd);
6826         int                     rc;
6827         ENTRY;
6828
6829         if (osd->od_is_ost || osd->od_cl_seq != NULL)
6830                 RETURN(0);
6831
6832         if (unlikely(ss == NULL))
6833                 RETURN(-ENODEV);
6834
6835         OBD_ALLOC_PTR(osd->od_cl_seq);
6836         if (osd->od_cl_seq == NULL)
6837                 RETURN(-ENOMEM);
6838
6839         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6840                              osd->od_svname, ss->ss_server_seq);
6841         if (rc != 0) {
6842                 OBD_FREE_PTR(osd->od_cl_seq);
6843                 osd->od_cl_seq = NULL;
6844                 RETURN(rc);
6845         }
6846
6847         if (ss->ss_node_id == 0) {
6848                 /* If the OSD on the sequence controller(MDT0), then allocate
6849                  * sequence here, otherwise allocate sequence after connected
6850                  * to MDT0 (see mdt_register_lwp_callback()). */
6851                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6852                                    &osd->od_cl_seq->lcs_space, env);
6853         }
6854
6855         RETURN(rc);
6856 }
6857
6858 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6859 {
6860         if (osd->od_cl_seq == NULL)
6861                 return;
6862
6863         seq_client_fini(osd->od_cl_seq);
6864         OBD_FREE_PTR(osd->od_cl_seq);
6865         osd->od_cl_seq = NULL;
6866 }
6867
6868 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6869 {
6870         ENTRY;
6871
6872         /* shutdown quota slave instance associated with the device */
6873         if (o->od_quota_slave != NULL) {
6874                 qsd_fini(env, o->od_quota_slave);
6875                 o->od_quota_slave = NULL;
6876         }
6877
6878         osd_fid_fini(env, o);
6879
6880         RETURN(0);
6881 }
6882
6883 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6884 {
6885         ENTRY;
6886
6887         if (o->od_mnt != NULL) {
6888                 shrink_dcache_sb(osd_sb(o));
6889                 osd_sync(env, &o->od_dt_dev);
6890
6891                 mntput(o->od_mnt);
6892                 o->od_mnt = NULL;
6893         }
6894
6895         EXIT;
6896 }
6897
6898 static int osd_mount(const struct lu_env *env,
6899                      struct osd_device *o, struct lustre_cfg *cfg)
6900 {
6901         const char              *name  = lustre_cfg_string(cfg, 0);
6902         const char              *dev  = lustre_cfg_string(cfg, 1);
6903         const char              *opts;
6904         unsigned long            page, s_flags, lmd_flags = 0;
6905         struct page             *__page;
6906         struct file_system_type *type;
6907         char                    *options = NULL;
6908         char                    *str;
6909         struct osd_thread_info  *info = osd_oti_get(env);
6910         struct lu_fid           *fid = &info->oti_fid;
6911         struct inode            *inode;
6912         int                      rc = 0, force_over_512tb = 0;
6913         ENTRY;
6914
6915         if (o->od_mnt != NULL)
6916                 RETURN(0);
6917
6918         if (strlen(dev) >= sizeof(o->od_mntdev))
6919                 RETURN(-E2BIG);
6920         strcpy(o->od_mntdev, dev);
6921
6922         str = lustre_cfg_string(cfg, 2);
6923         s_flags = simple_strtoul(str, NULL, 0);
6924         str = strstr(str, ":");
6925         if (str)
6926                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6927         opts = lustre_cfg_string(cfg, 3);
6928 #ifdef __BIG_ENDIAN
6929         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6930                 CERROR("%s: device %s extents feature is not guaranteed to "
6931                        "work on big-endian systems. Use \"bigendian_extents\" "
6932                        "mount option to override.\n", name, dev);
6933                 RETURN(-EINVAL);
6934         }
6935 #endif
6936 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6937         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6938                 CWARN("force_over_128tb option is deprecated. "
6939                       "Filesystems less than 512TB can be created without any "
6940                       "force options. Use force_over_512tb option for "
6941                       "filesystems greater than 512TB.\n");
6942         }
6943 #endif
6944 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6945         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6946                 CWARN("force_over_256tb option is deprecated. "
6947                       "Filesystems less than 512TB can be created without any "
6948                       "force options. Use force_over_512tb option for "
6949                       "filesystems greater than 512TB.\n");
6950         }
6951 #endif
6952
6953         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6954                 force_over_512tb = 1;
6955
6956         __page = alloc_page(GFP_KERNEL);
6957         if (__page == NULL)
6958                 GOTO(out, rc = -ENOMEM);
6959         page = (unsigned long)page_address(__page);
6960         options = (char *)page;
6961         *options = '\0';
6962         if (opts != NULL) {
6963                 /* strip out the options for back compatiblity */
6964                 static char *sout[] = {
6965                         "mballoc",
6966                         "iopen",
6967                         "noiopen",
6968                         "iopen_nopriv",
6969                         "extents",
6970                         "noextents",
6971                         /* strip out option we processed in osd */
6972                         "bigendian_extents",
6973                         "force_over_128tb",
6974                         "force_over_256tb",
6975                         "force_over_512tb",
6976                         NULL
6977                 };
6978                 strcat(options, opts);
6979                 for (rc = 0, str = options; sout[rc]; ) {
6980                         char *op = strstr(str, sout[rc]);
6981                         if (op == NULL) {
6982                                 rc++;
6983                                 str = options;
6984                                 continue;
6985                         }
6986                         if (op == options || *(op - 1) == ',') {
6987                                 str = op + strlen(sout[rc]);
6988                                 if (*str == ',' || *str == '\0') {
6989                                         *str == ',' ? str++ : str;
6990                                         memmove(op, str, strlen(str) + 1);
6991                                 }
6992                         }
6993                         for (str = op; *str != ',' && *str != '\0'; str++)
6994                                 ;
6995                 }
6996         } else {
6997                 strncat(options, "user_xattr,acl", 14);
6998         }
6999
7000         /* Glom up mount options */
7001         if (*options != '\0')
7002                 strcat(options, ",");
7003         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7004
7005         type = get_fs_type("ldiskfs");
7006         if (!type) {
7007                 CERROR("%s: cannot find ldiskfs module\n", name);
7008                 GOTO(out, rc = -ENODEV);
7009         }
7010
7011         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7012         module_put(type->owner);
7013
7014         if (IS_ERR(o->od_mnt)) {
7015                 rc = PTR_ERR(o->od_mnt);
7016                 o->od_mnt = NULL;
7017                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7018                 GOTO(out, rc);
7019         }
7020
7021         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7022                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7023                                  force_over_512tb == 0) {
7024                 CERROR("%s: device %s LDISKFS does not support filesystems "
7025                        "greater than 512TB and can cause data corruption. "
7026                        "Use \"force_over_512tb\" mount option to override.\n",
7027                        name, dev);
7028                 GOTO(out_mnt, rc = -EINVAL);
7029         }
7030
7031         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7032 #ifdef HAVE_DEV_SET_RDONLY
7033                 dev_set_rdonly(osd_sb(o)->s_bdev);
7034                 o->od_dt_dev.dd_rdonly = 1;
7035                 LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name);
7036 #else
7037                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7038                               name);
7039
7040                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7041 #endif
7042         } else {
7043 #ifdef HAVE_DEV_SET_RDONLY
7044                 if (dev_check_rdonly(osd_sb(o)->s_bdev)) {
7045                         CERROR("%s: underlying device %s is marked as "
7046                                "read-only. Setup failed\n", name, dev);
7047
7048                         GOTO(out_mnt, rc = -EROFS);
7049                 }
7050 #endif
7051         }
7052
7053         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7054                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7055                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7056                 GOTO(out_mnt, rc = -EINVAL);
7057         }
7058
7059 #ifdef LDISKFS_MOUNT_DIRDATA
7060         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7061                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7062                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7063         else if (!o->od_is_ost)
7064                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7065                       "enabling the dirdata feature. If you do not want to "
7066                       "downgrade to Lustre-1.x again, you can enable it via "
7067                       "'tune2fs -O dirdata device'\n", name, dev);
7068 #endif
7069         inode = osd_sb(o)->s_root->d_inode;
7070         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7071         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7072         if (rc != 0) {
7073                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7074                 GOTO(out_mnt, rc);
7075         }
7076
7077         if (lmd_flags & LMD_FLG_NOSCRUB)
7078                 o->od_noscrub = 1;
7079
7080         GOTO(out, rc = 0);
7081
7082 out_mnt:
7083         mntput(o->od_mnt);
7084         o->od_mnt = NULL;
7085
7086 out:
7087         if (__page)
7088                 __free_page(__page);
7089
7090         return rc;
7091 }
7092
7093 static struct lu_device *osd_device_fini(const struct lu_env *env,
7094                                          struct lu_device *d)
7095 {
7096         struct osd_device *o = osd_dev(d);
7097         ENTRY;
7098
7099         osd_shutdown(env, o);
7100         osd_procfs_fini(o);
7101         osd_scrub_cleanup(env, o);
7102         osd_obj_map_fini(o);
7103         osd_umount(env, o);
7104
7105         RETURN(NULL);
7106 }
7107
7108 static int osd_device_init0(const struct lu_env *env,
7109                             struct osd_device *o,
7110                             struct lustre_cfg *cfg)
7111 {
7112         struct lu_device        *l = osd2lu_dev(o);
7113         struct osd_thread_info *info;
7114         int                     rc;
7115         int                     cplen = 0;
7116
7117         /* if the module was re-loaded, env can loose its keys */
7118         rc = lu_env_refill((struct lu_env *) env);
7119         if (rc)
7120                 GOTO(out, rc);
7121         info = osd_oti_get(env);
7122         LASSERT(info);
7123
7124         l->ld_ops = &osd_lu_ops;
7125         o->od_dt_dev.dd_ops = &osd_dt_ops;
7126
7127         spin_lock_init(&o->od_osfs_lock);
7128         mutex_init(&o->od_otable_mutex);
7129         INIT_LIST_HEAD(&o->od_orphan_list);
7130
7131         o->od_read_cache = 1;
7132         o->od_writethrough_cache = 1;
7133         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7134
7135         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7136                         sizeof(o->od_svname));
7137         if (cplen >= sizeof(o->od_svname)) {
7138                 rc = -E2BIG;
7139                 GOTO(out, rc);
7140         }
7141
7142         o->od_index = -1; /* -1 means index is invalid */
7143         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7144         if (rc == LDD_F_SV_TYPE_OST)
7145                 o->od_is_ost = 1;
7146
7147         o->od_full_scrub_ratio = OFSR_DEFAULT;
7148         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7149         rc = osd_mount(env, o, cfg);
7150         if (rc != 0)
7151                 GOTO(out, rc);
7152
7153         rc = osd_obj_map_init(env, o);
7154         if (rc != 0)
7155                 GOTO(out_mnt, rc);
7156
7157         rc = lu_site_init(&o->od_site, l);
7158         if (rc != 0)
7159                 GOTO(out_compat, rc);
7160         o->od_site.ls_bottom_dev = l;
7161
7162         rc = lu_site_init_finish(&o->od_site);
7163         if (rc != 0)
7164                 GOTO(out_site, rc);
7165
7166         INIT_LIST_HEAD(&o->od_ios_list);
7167         /* setup scrub, including OI files initialization */
7168         rc = osd_scrub_setup(env, o);
7169         if (rc < 0)
7170                 GOTO(out_site, rc);
7171
7172         rc = osd_procfs_init(o, o->od_svname);
7173         if (rc != 0) {
7174                 CERROR("%s: can't initialize procfs: rc = %d\n",
7175                        o->od_svname, rc);
7176                 GOTO(out_scrub, rc);
7177         }
7178
7179         LASSERT(l->ld_site->ls_linkage.next != NULL);
7180         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7181
7182         /* initialize quota slave instance */
7183         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7184                                      o->od_proc_entry);
7185         if (IS_ERR(o->od_quota_slave)) {
7186                 rc = PTR_ERR(o->od_quota_slave);
7187                 o->od_quota_slave = NULL;
7188                 GOTO(out_procfs, rc);
7189         }
7190
7191         RETURN(0);
7192
7193 out_procfs:
7194         osd_procfs_fini(o);
7195 out_scrub:
7196         osd_scrub_cleanup(env, o);
7197 out_site:
7198         lu_site_fini(&o->od_site);
7199 out_compat:
7200         osd_obj_map_fini(o);
7201 out_mnt:
7202         osd_umount(env, o);
7203 out:
7204         return rc;
7205 }
7206
7207 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7208                                           struct lu_device_type *t,
7209                                           struct lustre_cfg *cfg)
7210 {
7211         struct osd_device *o;
7212         int                rc;
7213
7214         OBD_ALLOC_PTR(o);
7215         if (o == NULL)
7216                 return ERR_PTR(-ENOMEM);
7217
7218         rc = dt_device_init(&o->od_dt_dev, t);
7219         if (rc == 0) {
7220                 /* Because the ctx might be revived in dt_device_init,
7221                  * refill the env here */
7222                 lu_env_refill((struct lu_env *)env);
7223                 rc = osd_device_init0(env, o, cfg);
7224                 if (rc)
7225                         dt_device_fini(&o->od_dt_dev);
7226         }
7227
7228         if (unlikely(rc != 0))
7229                 OBD_FREE_PTR(o);
7230
7231         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7232 }
7233
7234 static struct lu_device *osd_device_free(const struct lu_env *env,
7235                                          struct lu_device *d)
7236 {
7237         struct osd_device *o = osd_dev(d);
7238         ENTRY;
7239
7240         /* XXX: make osd top device in order to release reference */
7241         d->ld_site->ls_top_dev = d;
7242         lu_site_purge(env, d->ld_site, -1);
7243         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7244                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7245                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7246         }
7247         lu_site_fini(&o->od_site);
7248         dt_device_fini(&o->od_dt_dev);
7249         OBD_FREE_PTR(o);
7250         RETURN(NULL);
7251 }
7252
7253 static int osd_process_config(const struct lu_env *env,
7254                               struct lu_device *d, struct lustre_cfg *cfg)
7255 {
7256         struct osd_device               *o = osd_dev(d);
7257         int                             rc;
7258         ENTRY;
7259
7260         switch (cfg->lcfg_command) {
7261         case LCFG_SETUP:
7262                 rc = osd_mount(env, o, cfg);
7263                 break;
7264         case LCFG_CLEANUP:
7265                 lu_dev_del_linkage(d->ld_site, d);
7266                 rc = osd_shutdown(env, o);
7267                 break;
7268         case LCFG_PARAM:
7269                 LASSERT(&o->od_dt_dev);
7270                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7271                                               cfg, &o->od_dt_dev);
7272                 if (rc > 0 || rc == -ENOSYS)
7273                         rc = class_process_proc_param(PARAM_OST,
7274                                                       lprocfs_osd_obd_vars,
7275                                                       cfg, &o->od_dt_dev);
7276                 break;
7277         default:
7278                 rc = -ENOSYS;
7279         }
7280
7281         RETURN(rc);
7282 }
7283
7284 static int osd_recovery_complete(const struct lu_env *env,
7285                                  struct lu_device *d)
7286 {
7287         struct osd_device       *osd = osd_dev(d);
7288         int                      rc = 0;
7289         ENTRY;
7290
7291         if (osd->od_quota_slave == NULL)
7292                 RETURN(0);
7293
7294         /* start qsd instance on recovery completion, this notifies the quota
7295          * slave code that we are about to process new requests now */
7296         rc = qsd_start(env, osd->od_quota_slave);
7297         RETURN(rc);
7298 }
7299
7300 /*
7301  * we use exports to track all osd users
7302  */
7303 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7304                            struct obd_device *obd, struct obd_uuid *cluuid,
7305                            struct obd_connect_data *data, void *localdata)
7306 {
7307         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7308         struct lustre_handle  conn;
7309         int                   rc;
7310         ENTRY;
7311
7312         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7313
7314         rc = class_connect(&conn, obd, cluuid);
7315         if (rc)
7316                 RETURN(rc);
7317
7318         *exp = class_conn2export(&conn);
7319
7320         spin_lock(&osd->od_osfs_lock);
7321         osd->od_connects++;
7322         spin_unlock(&osd->od_osfs_lock);
7323
7324         RETURN(0);
7325 }
7326
7327 /*
7328  * once last export (we don't count self-export) disappeared
7329  * osd can be released
7330  */
7331 static int osd_obd_disconnect(struct obd_export *exp)
7332 {
7333         struct obd_device *obd = exp->exp_obd;
7334         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7335         int                rc, release = 0;
7336         ENTRY;
7337
7338         /* Only disconnect the underlying layers on the final disconnect. */
7339         spin_lock(&osd->od_osfs_lock);
7340         osd->od_connects--;
7341         if (osd->od_connects == 0)
7342                 release = 1;
7343         spin_unlock(&osd->od_osfs_lock);
7344
7345         rc = class_disconnect(exp); /* bz 9811 */
7346
7347         if (rc == 0 && release)
7348                 class_manual_cleanup(obd);
7349         RETURN(rc);
7350 }
7351
7352 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7353                        struct lu_device *dev)
7354 {
7355         struct osd_device       *osd    = osd_dev(dev);
7356         struct lr_server_data   *lsd    =
7357                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7358         int                      result = 0;
7359         ENTRY;
7360
7361         if (osd->od_quota_slave != NULL) {
7362                 /* set up quota slave objects */
7363                 result = qsd_prepare(env, osd->od_quota_slave);
7364                 if (result != 0)
7365                         RETURN(result);
7366         }
7367
7368         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7369 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7370                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7371                         osd->od_index_in_idif = 1;
7372                 } else {
7373                         osd->od_index_in_idif = 0;
7374                         result = osd_register_proc_index_in_idif(osd);
7375                         if (result != 0)
7376                                 RETURN(result);
7377                 }
7378 #else
7379                 osd->od_index_in_idif = 1;
7380 #endif
7381         }
7382
7383         result = osd_fid_init(env, osd);
7384
7385         RETURN(result);
7386 }
7387
7388 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7389                          struct lu_fid *fid, struct md_op_data *op_data)
7390 {
7391         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7392
7393         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7394 }
7395
7396 static const struct lu_object_operations osd_lu_obj_ops = {
7397         .loo_object_init      = osd_object_init,
7398         .loo_object_delete    = osd_object_delete,
7399         .loo_object_release   = osd_object_release,
7400         .loo_object_free      = osd_object_free,
7401         .loo_object_print     = osd_object_print,
7402         .loo_object_invariant = osd_object_invariant
7403 };
7404
7405 const struct lu_device_operations osd_lu_ops = {
7406         .ldo_object_alloc      = osd_object_alloc,
7407         .ldo_process_config    = osd_process_config,
7408         .ldo_recovery_complete = osd_recovery_complete,
7409         .ldo_prepare           = osd_prepare,
7410 };
7411
7412 static const struct lu_device_type_operations osd_device_type_ops = {
7413         .ldto_init = osd_type_init,
7414         .ldto_fini = osd_type_fini,
7415
7416         .ldto_start = osd_type_start,
7417         .ldto_stop  = osd_type_stop,
7418
7419         .ldto_device_alloc = osd_device_alloc,
7420         .ldto_device_free  = osd_device_free,
7421
7422         .ldto_device_init    = osd_device_init,
7423         .ldto_device_fini    = osd_device_fini
7424 };
7425
7426 static struct lu_device_type osd_device_type = {
7427         .ldt_tags     = LU_DEVICE_DT,
7428         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7429         .ldt_ops      = &osd_device_type_ops,
7430         .ldt_ctx_tags = LCT_LOCAL,
7431 };
7432
7433 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7434 {
7435         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7436         struct super_block *sb = osd_sb(osd);
7437
7438         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7439 }
7440
7441 /*
7442  * lprocfs legacy support.
7443  */
7444 static struct obd_ops osd_obd_device_ops = {
7445         .o_owner = THIS_MODULE,
7446         .o_connect      = osd_obd_connect,
7447         .o_disconnect   = osd_obd_disconnect,
7448         .o_fid_alloc    = osd_fid_alloc,
7449         .o_health_check = osd_health_check,
7450 };
7451
7452 static int __init osd_init(void)
7453 {
7454         int rc;
7455
7456         LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7457 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7458         /* please, try to keep osd_thread_info smaller than a page */
7459         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7460 #endif
7461
7462         osd_oi_mod_init();
7463
7464         rc = lu_kmem_init(ldiskfs_caches);
7465         if (rc)
7466                 return rc;
7467
7468         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7469                                  lprocfs_osd_module_vars,
7470                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7471         if (rc)
7472                 lu_kmem_fini(ldiskfs_caches);
7473         return rc;
7474 }
7475
7476 static void __exit osd_exit(void)
7477 {
7478         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7479         lu_kmem_fini(ldiskfs_caches);
7480 }
7481
7482 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7483 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7484 MODULE_VERSION(LUSTRE_VERSION_STRING);
7485 MODULE_LICENSE("GPL");
7486
7487 module_init(osd_init);
7488 module_exit(osd_exit);