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, 2017, 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/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157         return oti->oti_w_locks > 0 && o->oo_owner == env;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't access mutable data
 162  */
 163 static int osd_root_get(const struct lu_env *env,
 164                         struct dt_device *dev, struct lu_fid *f)
 165 {
 166         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 167         return 0;
 168 }
 169
 170 /*
 171  * the following set of functions are used to maintain per-thread
 172  * cache of FID->ino mapping. this mechanism is needed to resolve
 173  * FID to inode at dt_insert() which in turn stores ino in the
 174  * directory entries to keep ldiskfs compatible with ext[34].
 175  * due to locking-originated restrictions we can't lookup ino
 176  * using LU cache (deadlock is possible). lookup using OI is quite
 177  * expensive. so instead we maintain this cache and methods like
 178  * dt_create() fill it. so in the majority of cases dt_insert() is
 179  * able to find needed mapping in lockless manner.
 180  */
 181 static struct osd_idmap_cache *
 182 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 183              const struct lu_fid *fid)
 184 {
 185         struct osd_thread_info  *oti   = osd_oti_get(env);
 186         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
 187         int i;
 188         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 189                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 190                         continue;
 191                 if (idc[i].oic_dev != osd)
 192                         continue;
 193
 194                 return idc + i;
 195         }
 196
 197         return NULL;
 198 }
 199
 200 static struct osd_idmap_cache *
 201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 202             const struct lu_fid *fid)
 203 {
 204         struct osd_thread_info  *oti   = osd_oti_get(env);
 205         struct osd_idmap_cache  *idc;
 206         int i;
 207
 208         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 209                 i = oti->oti_ins_cache_size * 2;
 210                 if (i == 0)
 211                         i = OSD_INS_CACHE_SIZE;
 212                 OBD_ALLOC(idc, sizeof(*idc) * i);
 213                 if (idc == NULL)
 214                         return ERR_PTR(-ENOMEM);
 215                 if (oti->oti_ins_cache != NULL) {
 216                         memcpy(idc, oti->oti_ins_cache,
 217                                oti->oti_ins_cache_used * sizeof(*idc));
 218                         OBD_FREE(oti->oti_ins_cache,
 219                                  oti->oti_ins_cache_used * sizeof(*idc));
 220                 }
 221                 oti->oti_ins_cache = idc;
 222                 oti->oti_ins_cache_size = i;
 223         }
 224
 225         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 226         idc->oic_fid = *fid;
 227         idc->oic_dev = osd;
 228         idc->oic_lid.oii_ino = 0;
 229         idc->oic_lid.oii_gen = 0;
 230         idc->oic_remote = 0;
 231
 232         return idc;
 233 }
 234
 235 /*
 236  * lookup mapping for the given fid in the cache, initialize a
 237  * new one if not found. the initialization checks whether the
 238  * object is local or remote. for local objects, OI is used to
 239  * learn ino/generation. the function is used when the caller
 240  * has no information about the object, e.g. at dt_insert().
 241  */
 242 static struct osd_idmap_cache *
 243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 244                      const struct lu_fid *fid)
 245 {
 246         struct osd_idmap_cache *idc;
 247         int rc;
 248
 249         idc = osd_idc_find(env, osd, fid);
 250         LASSERT(!IS_ERR(idc));
 251         if (idc != NULL)
 252                 return idc;
 253
 254         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 255                osd->od_svname, PFID(fid));
 256
 257         /* new mapping is needed */
 258         idc = osd_idc_add(env, osd, fid);
 259         if (IS_ERR(idc)) {
 260                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 261                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 262                 return idc;
 263         }
 264
 265         /* initialize it */
 266         rc = osd_remote_fid(env, osd, fid);
 267         if (unlikely(rc < 0))
 268                 return ERR_PTR(rc);
 269
 270         if (rc == 0) {
 271                 /* the object is local, lookup in OI */
 272                 /* XXX: probably cheaper to lookup in LU first? */
 273                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 274                                    &idc->oic_lid, 0);
 275                 if (unlikely(rc < 0)) {
 276                         CERROR("can't lookup: rc = %d\n", rc);
 277                         return ERR_PTR(rc);
 278                 }
 279         } else {
 280                 /* the object is remote */
 281                 idc->oic_remote = 1;
 282         }
 283
 284         return idc;
 285 }
 286
 287 /*
 288  * lookup mapping for given FID and fill it from the given object.
 289  * the object is lolcal by definition.
 290  */
 291 static int osd_idc_find_and_init(const struct lu_env *env,
 292                                  struct osd_device *osd,
 293                                  struct osd_object *obj)
 294 {
 295         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 296         struct osd_idmap_cache  *idc;
 297
 298         idc = osd_idc_find(env, osd, fid);
 299         LASSERT(!IS_ERR(idc));
 300         if (idc != NULL) {
 301                 if (obj->oo_inode == NULL)
 302                         return 0;
 303                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 304                         LASSERT(idc->oic_lid.oii_ino == 0);
 305                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 306                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 307                 }
 308                 return 0;
 309         }
 310
 311         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 312                osd->od_svname, PFID(fid));
 313
 314         /* new mapping is needed */
 315         idc = osd_idc_add(env, osd, fid);
 316         if (IS_ERR(idc)) {
 317                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 318                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 319                 return PTR_ERR(idc);
 320         }
 321
 322         if (obj->oo_inode != NULL) {
 323                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 324                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 325         }
 326         return 0;
 327 }
 328
 329 /*
 330  * OSD object methods.
 331  */
 332
 333 /*
 334  * Concurrency: no concurrent access is possible that early in object
 335  * life-cycle.
 336  */
 337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 338                                           const struct lu_object_header *hdr,
 339                                           struct lu_device *d)
 340 {
 341         struct osd_object *mo;
 342
 343         OBD_ALLOC_PTR(mo);
 344         if (mo != NULL) {
 345                 struct lu_object *l;
 346                 struct lu_object_header *h;
 347                 struct osd_device *o = osd_dev(d);
 348
 349                 l = &mo->oo_dt.do_lu;
 350                 if (unlikely(o->od_in_init)) {
 351                         OBD_ALLOC_PTR(h);
 352                         if (!h) {
 353                                 OBD_FREE_PTR(mo);
 354                                 return NULL;
 355                         }
 356
 357                         lu_object_header_init(h);
 358                         lu_object_init(l, h, d);
 359                         lu_object_add_top(h, l);
 360                         mo->oo_header = h;
 361                 } else {
 362                         dt_object_init(&mo->oo_dt, NULL, d);
 363                         mo->oo_header = NULL;
 364                 }
 365
 366                 mo->oo_dt.do_ops = &osd_obj_ops;
 367                 l->lo_ops = &osd_lu_obj_ops;
 368                 init_rwsem(&mo->oo_sem);
 369                 init_rwsem(&mo->oo_ext_idx_sem);
 370                 spin_lock_init(&mo->oo_guard);
 371                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 372                 return l;
 373         } else {
 374                 return NULL;
 375         }
 376 }
 377
 378 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 379                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 380 {
 381         int rc;
 382
 383         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 384                              (void *)loa, sizeof(*loa));
 385         if (rc > 0) {
 386                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 387
 388                 if (rc < sizeof(*lma))
 389                         return -EINVAL;
 390
 391                 rc = 0;
 392                 lustre_loa_swab(loa, true);
 393                 /* Check LMA compatibility */
 394                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 395                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 396                               "for fid = "DFID", ino = %lu\n",
 397                               osd_ino2name(inode),
 398                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 399                               PFID(&lma->lma_self_fid), inode->i_ino);
 400                         rc = -EOPNOTSUPP;
 401                 }
 402         } else if (rc == 0) {
 403                 rc = -ENODATA;
 404         }
 405
 406         return rc;
 407 }
 408
 409 /*
 410  * retrieve object from backend ext fs.
 411  **/
 412 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 413                        struct osd_inode_id *id)
 414 {
 415         int rc;
 416         struct inode *inode = NULL;
 417
 418         /* if we look for an inode withing a running
 419          * transaction, then we risk to deadlock */
 420         /* osd_dirent_check_repair() breaks this */
 421         /*LASSERT(current->journal_info == NULL);*/
 422
 423         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 424         if (IS_ERR(inode)) {
 425                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 426                        id->oii_ino, PTR_ERR(inode));
 427         } else if (id->oii_gen != OSD_OII_NOGEN &&
 428                    inode->i_generation != id->oii_gen) {
 429                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 430                        "i_generation = %u\n",
 431                        id->oii_ino, id->oii_gen, inode->i_generation);
 432                 iput(inode);
 433                 inode = ERR_PTR(-ESTALE);
 434         } else if (inode->i_nlink == 0) {
 435                 /* due to parallel readdir and unlink,
 436                 * we can have dead inode here. */
 437                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 438                 iput(inode);
 439                 inode = ERR_PTR(-ESTALE);
 440         } else if (is_bad_inode(inode)) {
 441                 CWARN("%s: bad inode: ino = %u\n",
 442                 osd_dev2name(dev), id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ENOENT);
 445         } else if ((rc = osd_attach_jinode(inode))) {
 446                 iput(inode);
 447                 inode = ERR_PTR(rc);
 448         } else {
 449                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 450                 if (id->oii_gen == OSD_OII_NOGEN)
 451                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 452
 453                 /* Do not update file c/mtime in ldiskfs.
 454                  * NB: we don't have any lock to protect this because we don't
 455                  * have reference on osd_object now, but contention with
 456                  * another lookup + attr_set can't happen in the tiny window
 457                  * between if (...) and set S_NOCMTIME. */
 458                 if (!(inode->i_flags & S_NOCMTIME))
 459                         inode->i_flags |= S_NOCMTIME;
 460         }
 461         return inode;
 462 }
 463
 464 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 465                           handle_t *handle, struct dentry *child,
 466                           struct inode *inode, struct htree_lock *hlock)
 467 {
 468         int rc, rc2;
 469
 470         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 471         if (rc == -ENOBUFS || rc == -ENOSPC) {
 472                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 473                 struct inode *parent = child->d_parent->d_inode;
 474                 struct lu_fid *fid = NULL;
 475
 476                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 477                 if (!rc2) {
 478                         fid = &loa->loa_lma.lma_self_fid;
 479                 } else if (rc2 == -ENODATA) {
 480                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 481                                 fid = &info->oti_fid3;
 482                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 483                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 484                                 fid = &info->oti_fid3;
 485                                 lu_igif_build(fid, parent->i_ino,
 486                                               parent->i_generation);
 487                         }
 488                 }
 489
 490                 if (fid != NULL)
 491                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 492                               "maximum entry limit\n",
 493                               osd_name(osd), parent->i_ino, PFID(fid),
 494                               rc == -ENOSPC ? "has reached" : "is approaching");
 495                 else
 496                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 497                               "maximum entry limit\n",
 498                               osd_name(osd), parent->i_ino,
 499                               rc == -ENOSPC ? "has reached" : "is approaching");
 500
 501                 /* ignore such error now */
 502                 if (rc == -ENOBUFS)
 503                         rc = 0;
 504         }
 505
 506         return rc;
 507 }
 508
 509
 510 struct inode *
 511 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 512              struct osd_inode_id *id, struct lu_fid *fid)
 513 {
 514         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 515         struct inode *inode;
 516         int rc;
 517
 518         inode = osd_iget(info, dev, id);
 519         if (IS_ERR(inode))
 520                 return inode;
 521
 522         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 523         if (!rc) {
 524                 *fid = loa->loa_lma.lma_self_fid;
 525         } else if (rc == -ENODATA) {
 526                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 527                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 528                 else
 529                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 530         } else {
 531                 iput(inode);
 532                 inode = ERR_PTR(rc);
 533         }
 534         return inode;
 535 }
 536
 537 static struct inode *osd_iget_check(struct osd_thread_info *info,
 538                                     struct osd_device *dev,
 539                                     const struct lu_fid *fid,
 540                                     struct osd_inode_id *id,
 541                                     bool trusted)
 542 {
 543         struct inode *inode;
 544         int rc = 0;
 545         ENTRY;
 546
 547         /* The cached OI mapping is trustable. If we cannot locate the inode
 548          * via the cached OI mapping, then return the failure to the caller
 549          * directly without further OI checking. */
 550
 551 again:
 552         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 553         if (IS_ERR(inode)) {
 554                 rc = PTR_ERR(inode);
 555                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 556                         goto check_oi;
 557
 558                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 559                        PFID(fid), id->oii_ino, rc);
 560                 GOTO(put, rc);
 561         }
 562
 563         if (is_bad_inode(inode)) {
 564                 rc = -ENOENT;
 565                 if (!trusted)
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 569                        PFID(fid), id->oii_ino);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (id->oii_gen != OSD_OII_NOGEN &&
 574             inode->i_generation != id->oii_gen) {
 575                 rc = -ESTALE;
 576                 if (!trusted)
 577                         goto check_oi;
 578
 579                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 580                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 581                        id->oii_ino, id->oii_gen, inode->i_generation);
 582                 GOTO(put, rc);
 583         }
 584
 585         if (inode->i_nlink == 0) {
 586                 rc = -ENOENT;
 587                 if (!trusted)
 588                         goto check_oi;
 589
 590                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 591                        PFID(fid), id->oii_ino);
 592                 GOTO(put, rc);
 593         }
 594
 595         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 596
 597 check_oi:
 598         if (rc != 0) {
 599                 __u32 saved_ino = id->oii_ino;
 600                 __u32 saved_gen = id->oii_gen;
 601
 602                 LASSERT(!trusted);
 603                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 604
 605                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 606                 /* XXX: There are four possible cases:
 607                  *      1. rc = 0.
 608                  *         Backup/restore caused the OI invalid.
 609                  *      2. rc = 0.
 610                  *         Someone unlinked the object but NOT removed
 611                  *         the OI mapping, such as mount target device
 612                  *         as ldiskfs, and modify something directly.
 613                  *      3. rc = -ENOENT.
 614                  *         Someone just removed the object between the
 615                  *         former oi_lookup and the iget. It is normal.
 616                  *      4. Other failure cases.
 617                  *
 618                  *      Generally, when the device is mounted, it will
 619                  *      auto check whether the system is restored from
 620                  *      file-level backup or not. We trust such detect
 621                  *      to distinguish the 1st case from the 2nd case:
 622                  *      if the OI files are consistent but may contain
 623                  *      stale OI mappings because of case 2, if iget()
 624                  *      returns -ENOENT or -ESTALE, then it should be
 625                  *      the case 2. */
 626                 if (rc != 0)
 627                         /* If the OI mapping was in OI file before the
 628                          * osd_iget_check(), but now, it is disappear,
 629                          * then it must be removed by race. That is a
 630                          * normal race case. */
 631                         GOTO(put, rc);
 632
 633                 /* It is the OI scrub updated the OI mapping by race.
 634                  * The new OI mapping must be valid. */
 635                 if (saved_ino != id->oii_ino ||
 636                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 637                         if (!IS_ERR(inode))
 638                                 iput(inode);
 639
 640                         trusted = true;
 641                         goto again;
 642                 }
 643
 644                 if (IS_ERR(inode)) {
 645                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 646                             SF_INCONSISTENT)
 647                                 /* It still can be the case 2, but we cannot
 648                                  * distinguish it from the case 1. So return
 649                                  * -EREMCHG to block current operation until
 650                                  *  OI scrub rebuilt the OI mappings. */
 651                                 rc = -EREMCHG;
 652                         else
 653                                 rc = -ENOENT;
 654
 655                         GOTO(put, rc);
 656                 }
 657
 658                 if (inode->i_generation == id->oii_gen)
 659                         rc = -ENOENT;
 660                 else
 661                         rc = -EREMCHG;
 662         } else {
 663                 if (id->oii_gen == OSD_OII_NOGEN)
 664                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 665
 666                 /* Do not update file c/mtime in ldiskfs.
 667                  * NB: we don't have any lock to protect this because we don't
 668                  * have reference on osd_object now, but contention with
 669                  * another lookup + attr_set can't happen in the tiny window
 670                  * between if (...) and set S_NOCMTIME. */
 671                 if (!(inode->i_flags & S_NOCMTIME))
 672                         inode->i_flags |= S_NOCMTIME;
 673         }
 674
 675         GOTO(put, rc);
 676
 677 put:
 678         if (rc != 0) {
 679                 if (!IS_ERR(inode))
 680                         iput(inode);
 681
 682                 inode = ERR_PTR(rc);
 683         }
 684
 685         return inode;
 686 }
 687
 688 /**
 689  * \retval +v: new filter_fid, does not contain self-fid
 690  * \retval 0:  filter_fid_old, contains self-fid
 691  * \retval -v: other failure cases
 692  */
 693 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 694                  struct dentry *dentry, struct lu_fid *fid)
 695 {
 696         struct filter_fid_old   *ff     = &info->oti_ff;
 697         struct ost_id           *ostid  = &info->oti_ostid;
 698         int                      rc;
 699
 700         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 701         if (rc == sizeof(*ff)) {
 702                 rc = 0;
 703                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 704                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 705                 /*
 706                  * XXX: use 0 as the index for compatibility, the caller will
 707                  *      handle index related issues when necessary.
 708                  */
 709                 if (!rc)
 710                         ostid_to_fid(fid, ostid, 0);
 711         } else if (rc == sizeof(struct filter_fid)) {
 712                 rc = 1;
 713         } else if (rc >= 0) {
 714                 rc = -EINVAL;
 715         }
 716
 717         return rc;
 718 }
 719
 720 static int osd_lma_self_repair(struct osd_thread_info *info,
 721                                struct osd_device *osd, struct inode *inode,
 722                                const struct lu_fid *fid, __u32 compat)
 723 {
 724         handle_t *jh;
 725         int       rc;
 726
 727         LASSERT(current->journal_info == NULL);
 728
 729         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 730                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 731         if (IS_ERR(jh)) {
 732                 rc = PTR_ERR(jh);
 733                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 734                       osd_name(osd), rc);
 735                 return rc;
 736         }
 737
 738         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 739         if (rc != 0)
 740                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 741                       osd_name(osd), rc);
 742         ldiskfs_journal_stop(jh);
 743         return rc;
 744 }
 745
 746 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 747 {
 748         struct osd_thread_info  *info   = osd_oti_get(env);
 749         struct osd_device       *osd    = osd_obj2dev(obj);
 750         struct lustre_ost_attrs *loa    = &info->oti_ost_attrs;
 751         struct lustre_mdt_attrs *lma    = &loa->loa_lma;
 752         struct inode            *inode  = obj->oo_inode;
 753         struct dentry           *dentry = &info->oti_obj_dentry;
 754         struct lu_fid           *fid    = NULL;
 755         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 756         int                      rc;
 757         ENTRY;
 758
 759         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 760                              (void *)loa, sizeof(*loa));
 761         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 762                 fid = &lma->lma_self_fid;
 763                 rc = osd_get_idif(info, inode, dentry, fid);
 764                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 765                         /* For the given OST-object, if it has neither LMA nor
 766                          * FID in XATTR_NAME_FID, then the given FID (which is
 767                          * contained in the @obj, from client RPC for locating
 768                          * the OST-object) is trusted. We use it to generate
 769                          * the LMA. */
 770                         osd_lma_self_repair(info, osd, inode, rfid,
 771                                             LMAC_FID_ON_OST);
 772                         RETURN(0);
 773                 }
 774         }
 775
 776         if (rc < 0)
 777                 RETURN(rc);
 778
 779         if (rc > 0) {
 780                 rc = 0;
 781                 lustre_lma_swab(lma);
 782                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 783                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 784                               S_ISREG(inode->i_mode)))) {
 785                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 786                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 787                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 788                               PFID(rfid), inode->i_ino);
 789                         rc = -EOPNOTSUPP;
 790                 } else {
 791                         fid = &lma->lma_self_fid;
 792                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 793                             osd->od_is_ost)
 794                                 obj->oo_pfid_in_lma = 1;
 795                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 796                             !osd->od_is_ost)
 797                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 798                 }
 799         }
 800
 801         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 802                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 803                         struct ost_id   *oi   = &info->oti_ostid;
 804                         struct lu_fid   *fid1 = &info->oti_fid3;
 805                         __u32            idx  = fid_idif_ost_idx(rfid);
 806
 807                         /* For old IDIF, the OST index is not part of the IDIF,
 808                          * Means that different OSTs may have the same IDIFs.
 809                          * Under such case, we need to make some compatible
 810                          * check to make sure to trigger OI scrub properly. */
 811                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 812                                 /* Given @rfid is new, LMA is old. */
 813                                 fid_to_ostid(fid, oi);
 814                                 ostid_to_fid(fid1, oi, idx);
 815                                 if (lu_fid_eq(fid1, rfid)) {
 816                                         if (osd->od_index_in_idif)
 817                                                 osd_lma_self_repair(info, osd,
 818                                                         inode, rfid,
 819                                                         LMAC_FID_ON_OST);
 820                                         RETURN(0);
 821                                 }
 822                         }
 823                 }
 824
 825                 rc = -EREMCHG;
 826         }
 827
 828         RETURN(rc);
 829 }
 830
 831 struct osd_check_lmv_buf {
 832 #ifdef HAVE_DIR_CONTEXT
 833         /* please keep it as first member */
 834         struct dir_context       ctx;
 835 #endif
 836         struct osd_thread_info  *oclb_info;
 837         struct osd_device       *oclb_dev;
 838         struct osd_idmap_cache  *oclb_oic;
 839         int                      oclb_items;
 840         bool                     oclb_found;
 841 };
 842
 843 /**
 844  * It is called internally by ->readdir() to filter out the
 845  * local slave object's FID of the striped directory.
 846  *
 847  * \retval      1 found the local slave's FID
 848  * \retval      0 continue to check next item
 849  * \retval      -ve for failure
 850  */
 851 #ifdef HAVE_FILLDIR_USE_CTX
 852 static int osd_stripe_dir_filldir(struct dir_context *buf,
 853 #else
 854 static int osd_stripe_dir_filldir(void *buf,
 855 #endif
 856                                   const char *name, int namelen,
 857                                   loff_t offset, __u64 ino, unsigned d_type)
 858 {
 859         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 860         struct osd_thread_info *oti = oclb->oclb_info;
 861         struct lu_fid *fid = &oti->oti_fid3;
 862         struct osd_inode_id *id = &oti->oti_id3;
 863         struct osd_device *dev = oclb->oclb_dev;
 864         struct osd_idmap_cache *oic = oclb->oclb_oic;
 865         struct inode *inode;
 866
 867         oclb->oclb_items++;
 868
 869         if (name[0] == '.')
 870                 return 0;
 871
 872         fid_zero(fid);
 873         sscanf(name + 1, SFID, RFID(fid));
 874         if (!fid_is_sane(fid))
 875                 return 0;
 876
 877         if (osd_remote_fid(oti->oti_env, dev, fid))
 878                 return 0;
 879
 880         osd_id_gen(id, ino, OSD_OII_NOGEN);
 881         inode = osd_iget(oti, dev, id);
 882         if (IS_ERR(inode))
 883                 return PTR_ERR(inode);
 884
 885         iput(inode);
 886         osd_add_oi_cache(oti, dev, id, fid);
 887         oic->oic_fid = *fid;
 888         oic->oic_lid = *id;
 889         oic->oic_dev = dev;
 890         osd_oii_insert(dev, oic, true);
 891         oclb->oclb_found = true;
 892
 893         return 1;
 894 }
 895
 896 /* When lookup item under striped directory, we need to locate the master
 897  * MDT-object of the striped directory firstly, then the client will send
 898  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 899  * and the item's name. If the system is restored from MDT file level backup,
 900  * then before the OI scrub completely built the OI files, the OI mappings of
 901  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 902  * not a problem for the master MDT-object. Because when locate the master
 903  * MDT-object, we will do name based lookup (for the striped directory itself)
 904  * firstly, during such process we can setup the correct OI mapping for the
 905  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 906  * the client will not trigger name based lookup on the MDT to locate the slave
 907  * MDT-object before locating item under the striped directory, then when
 908  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 909  * is invalid and does not know what the right OI mapping is, then the MDT has
 910  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 911  * the OI file, related OI mapping is unknown yet, please try again later. And
 912  * then client will re-try the RPC again and again until related OI mapping has
 913  * been updated. That is quite inefficient.
 914  *
 915  * To resolve above trouble, we will handle it as the following two cases:
 916  *
 917  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 918  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 919  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 920  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 921  *    directly. Please check osd_fid_lookup().
 922  *
 923  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 924  *    Under such case, during lookup the master MDT-object, we will lookup the
 925  *    slave MDT-object via readdir against the master MDT-object, because the
 926  *    slave MDT-objects information are stored as sub-directories with the name
 927  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 928  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 929  *    the correct OI mapping for the slave MDT-object. */
 930 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 931                          struct inode *inode, struct osd_idmap_cache *oic)
 932 {
 933         struct lu_buf *buf = &oti->oti_big_buf;
 934         struct dentry *dentry = &oti->oti_obj_dentry;
 935         struct file *filp = &oti->oti_file;
 936         const struct file_operations *fops;
 937         struct lmv_mds_md_v1 *lmv1;
 938         struct osd_check_lmv_buf oclb = {
 939 #ifdef HAVE_DIR_CONTEXT
 940                 .ctx.actor = osd_stripe_dir_filldir,
 941 #endif
 942                 .oclb_info = oti,
 943                 .oclb_dev = dev,
 944                 .oclb_oic = oic,
 945                 .oclb_found = false,
 946         };
 947         int rc = 0;
 948         ENTRY;
 949
 950 again:
 951         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 952                              buf->lb_len);
 953         if (rc == -ERANGE) {
 954                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 955                 if (rc > 0) {
 956                         lu_buf_realloc(buf, rc);
 957                         if (buf->lb_buf == NULL)
 958                                 GOTO(out, rc = -ENOMEM);
 959
 960                         goto again;
 961                 }
 962         }
 963
 964         if (unlikely(rc == 0 || rc == -ENODATA))
 965                 GOTO(out, rc = 0);
 966
 967         if (rc < 0)
 968                 GOTO(out, rc);
 969
 970         if (unlikely(buf->lb_buf == NULL)) {
 971                 lu_buf_realloc(buf, rc);
 972                 if (buf->lb_buf == NULL)
 973                         GOTO(out, rc = -ENOMEM);
 974
 975                 goto again;
 976         }
 977
 978         lmv1 = buf->lb_buf;
 979         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 980                 GOTO(out, rc = 0);
 981
 982         fops = inode->i_fop;
 983         dentry->d_inode = inode;
 984         dentry->d_sb = inode->i_sb;
 985         filp->f_pos = 0;
 986         filp->f_path.dentry = dentry;
 987         filp->f_mode = FMODE_64BITHASH;
 988         filp->f_mapping = inode->i_mapping;
 989         filp->f_op = fops;
 990         filp->private_data = NULL;
 991         set_file_inode(filp, inode);
 992
 993         do {
 994                 oclb.oclb_items = 0;
 995 #ifdef HAVE_DIR_CONTEXT
 996                 oclb.ctx.pos = filp->f_pos;
 997                 rc = fops->iterate(filp, &oclb.ctx);
 998                 filp->f_pos = oclb.ctx.pos;
 999 #else
1000                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1001 #endif
1002         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1003                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1004         fops->release(inode, filp);
1005
1006 out:
1007         if (rc < 0)
1008                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1009                        DFID": rc = %d\n", osd_ino2name(inode),
1010                        inode->i_ino, inode->i_generation,
1011                        PFID(&oic->oic_fid), rc);
1012         else
1013                 rc = 0;
1014
1015         RETURN(rc);
1016 }
1017
1018 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1019                           const struct lu_fid *fid,
1020                           const struct lu_object_conf *conf)
1021 {
1022         struct osd_thread_info *info;
1023         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1024         struct osd_device *dev;
1025         struct osd_idmap_cache *oic;
1026         struct osd_inode_id *id;
1027         struct inode *inode = NULL;
1028         struct lustre_scrub *scrub;
1029         struct scrub_file *sf;
1030         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1031         __u32 saved_ino;
1032         __u32 saved_gen;
1033         int result = 0;
1034         int rc1 = 0;
1035         bool remote = false;
1036         bool trusted = true;
1037         bool updated = false;
1038         bool checked = false;
1039         ENTRY;
1040
1041         LINVRNT(osd_invariant(obj));
1042         LASSERT(obj->oo_inode == NULL);
1043         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1044
1045         dev = osd_dev(ldev);
1046         scrub = &dev->od_scrub.os_scrub;
1047         sf = &scrub->os_file;
1048         info = osd_oti_get(env);
1049         LASSERT(info);
1050         oic = &info->oti_cache;
1051
1052         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1053                 RETURN(-ENOENT);
1054
1055         /* For the object is created as locking anchor, or for the object to
1056          * be created on disk. No need to osd_oi_lookup() at here because FID
1057          * shouldn't never be re-used, if it's really a duplicate FID from
1058          * unexpected reason, we should be able to detect it later by calling
1059          * do_create->osd_oi_insert(). */
1060         if (conf && conf->loc_flags & LOC_F_NEW)
1061                 GOTO(out, result = 0);
1062
1063         /* Search order: 1. per-thread cache. */
1064         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1065                 id = &oic->oic_lid;
1066                 goto iget;
1067         }
1068
1069         id = &info->oti_id;
1070         if (!list_empty(&scrub->os_inconsistent_items)) {
1071                 /* Search order: 2. OI scrub pending list. */
1072                 result = osd_oii_lookup(dev, fid, id);
1073                 if (!result)
1074                         goto iget;
1075         }
1076
1077         /* The OI mapping in the OI file can be updated by the OI scrub
1078          * when we locate the inode via FID. So it may be not trustable. */
1079         trusted = false;
1080
1081         /* Search order: 3. OI files. */
1082         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1083         if (result == -ENOENT) {
1084                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1085                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1086                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1087                                       sf->sf_oi_bitmap))
1088                         GOTO(out, result = 0);
1089
1090                 goto trigger;
1091         }
1092
1093         if (result)
1094                 GOTO(out, result);
1095
1096 iget:
1097         obj->oo_inode = NULL;
1098         /* for later passes through checks, not true on first pass */
1099         if (!IS_ERR_OR_NULL(inode))
1100                 iput(inode);
1101
1102         inode = osd_iget_check(info, dev, fid, id, trusted);
1103         if (!IS_ERR(inode)) {
1104                 obj->oo_inode = inode;
1105                 result = 0;
1106                 if (remote)
1107                         goto trigger;
1108
1109                 goto check_lma;
1110         }
1111
1112         result = PTR_ERR(inode);
1113         if (result == -ENOENT || result == -ESTALE)
1114                 GOTO(out, result = 0);
1115
1116         if (result != -EREMCHG)
1117                 GOTO(out, result);
1118
1119 trigger:
1120         /* We still have chance to get the valid inode: for the
1121          * object which is referenced by remote name entry, the
1122          * object on the local MDT will be linked under the dir
1123          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1124          *
1125          * We do not know whether the object for the given FID
1126          * is referenced by some remote name entry or not, and
1127          * especially for DNE II, a multiple-linked object may
1128          * have many name entries reside on many MDTs.
1129          *
1130          * To simplify the operation, OSD will not distinguish
1131          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1132          * only happened for the RPC from other MDT during the
1133          * OI scrub, or for the client side RPC with FID only,
1134          * such as FID to path, or from old connected client. */
1135         if (!remote) {
1136                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1137                 if (!rc1) {
1138                         remote = true;
1139                         trusted = true;
1140                         flags |= SS_AUTO_PARTIAL;
1141                         flags &= ~SS_AUTO_FULL;
1142                         goto iget;
1143                 }
1144         }
1145
1146         if (thread_is_running(&scrub->os_thread)) {
1147                 if (scrub->os_partial_scan && !scrub->os_in_join)
1148                         goto join;
1149
1150                 if (IS_ERR_OR_NULL(inode) || result)
1151                         GOTO(out, result = -EINPROGRESS);
1152
1153                 LASSERT(remote);
1154                 LASSERT(obj->oo_inode == inode);
1155
1156                 osd_add_oi_cache(info, dev, id, fid);
1157                 osd_oii_insert(dev, oic, true);
1158                 goto found;
1159         }
1160
1161         if (dev->od_auto_scrub_interval == AS_NEVER) {
1162                 if (!remote)
1163                         GOTO(out, result = -EREMCHG);
1164
1165                 LASSERT(!result);
1166                 LASSERT(obj->oo_inode == inode);
1167
1168                 osd_add_oi_cache(info, dev, id, fid);
1169                 goto found;
1170         }
1171
1172 join:
1173         rc1 = osd_scrub_start(env, dev, flags);
1174         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1175                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1176         if (rc1 && rc1 != -EALREADY)
1177                 GOTO(out, result = -EREMCHG);
1178
1179         if (IS_ERR_OR_NULL(inode) || result)
1180                 GOTO(out, result = -EINPROGRESS);
1181
1182         LASSERT(remote);
1183         LASSERT(obj->oo_inode == inode);
1184
1185         osd_add_oi_cache(info, dev, id, fid);
1186         osd_oii_insert(dev, oic, true);
1187         goto found;
1188
1189 check_lma:
1190         checked = true;
1191         if (unlikely(obj->oo_header))
1192                 goto found;
1193
1194         result = osd_check_lma(env, obj);
1195         if (!result)
1196                 goto found;
1197
1198         LASSERTF(id->oii_ino == inode->i_ino &&
1199                  id->oii_gen == inode->i_generation,
1200                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1201                  PFID(fid), id->oii_ino, id->oii_gen,
1202                  inode->i_ino, inode->i_generation);
1203
1204         saved_ino = inode->i_ino;
1205         saved_gen = inode->i_generation;
1206
1207         if (unlikely(result == -ENODATA)) {
1208                 /* If the OI scrub updated the OI mapping by race, it
1209                  * must be valid. Trust the inode that has no LMA EA. */
1210                 if (updated)
1211                         goto found;
1212
1213                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1214                 if (!result) {
1215                         /* The OI mapping is still there, the inode is still
1216                          * valid. It is just becaues the inode has no LMA EA. */
1217                         if (saved_ino == id->oii_ino &&
1218                             saved_gen == id->oii_gen)
1219                                 goto found;
1220
1221                         /* It is the OI scrub updated the OI mapping by race.
1222                          * The new OI mapping must be valid. */
1223                         trusted = true;
1224                         updated = true;
1225                         goto iget;
1226                 }
1227
1228                 /* "result == -ENOENT" means that the OI mappinghas been
1229                  * removed by race, so the inode belongs to other object.
1230                  *
1231                  * Others error can be returned  directly. */
1232                 if (result == -ENOENT) {
1233                         LASSERT(trusted);
1234
1235                         obj->oo_inode = NULL;
1236                         result = 0;
1237                 }
1238         }
1239
1240         if (result != -EREMCHG)
1241                 GOTO(out, result);
1242
1243         LASSERT(!updated);
1244
1245         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1246         /* "result == -ENOENT" means the cached OI mapping has been removed
1247          * from the OI file by race, above inode belongs to other object. */
1248         if (result == -ENOENT) {
1249                 LASSERT(trusted);
1250
1251                 obj->oo_inode = NULL;
1252                 GOTO(out, result = 0);
1253         }
1254
1255         if (result)
1256                 GOTO(out, result);
1257
1258         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1259                 result = -EREMCHG;
1260                 goto trigger;
1261         }
1262
1263         /* It is the OI scrub updated the OI mapping by race.
1264          * The new OI mapping must be valid. */
1265         trusted = true;
1266         updated = true;
1267         goto iget;
1268
1269 found:
1270         if (!checked) {
1271                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1272                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1273
1274                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1275                 if (!result) {
1276                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1277                             dev->od_is_ost)
1278                                 obj->oo_pfid_in_lma = 1;
1279                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1280                             !dev->od_is_ost)
1281                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1282                 } else if (result != -ENODATA) {
1283                         GOTO(out, result);
1284                 }
1285         }
1286
1287         obj->oo_compat_dot_created = 1;
1288         obj->oo_compat_dotdot_created = 1;
1289
1290         if (S_ISDIR(inode->i_mode) &&
1291             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1292                 osd_check_lmv(info, dev, inode, oic);
1293
1294         result = osd_attach_jinode(inode);
1295         if (result)
1296                 GOTO(out, result);
1297
1298         if (!ldiskfs_pdo)
1299                 GOTO(out, result = 0);
1300
1301         LASSERT(!obj->oo_hl_head);
1302         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1303
1304         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1305
1306 out:
1307         if (result || !obj->oo_inode) {
1308                 if (!IS_ERR_OR_NULL(inode))
1309                         iput(inode);
1310
1311                 obj->oo_inode = NULL;
1312                 if (trusted)
1313                         fid_zero(&oic->oic_fid);
1314         }
1315
1316         LINVRNT(osd_invariant(obj));
1317         return result;
1318 }
1319
1320 /*
1321  * Concurrency: shouldn't matter.
1322  */
1323 static void osd_object_init0(struct osd_object *obj)
1324 {
1325         LASSERT(obj->oo_inode != NULL);
1326         obj->oo_dt.do_body_ops = &osd_body_ops;
1327         obj->oo_dt.do_lu.lo_header->loh_attr |=
1328                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1329 }
1330
1331 /*
1332  * Concurrency: no concurrent access is possible that early in object
1333  * life-cycle.
1334  */
1335 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1336                            const struct lu_object_conf *conf)
1337 {
1338         struct osd_object *obj = osd_obj(l);
1339         int result;
1340
1341         LINVRNT(osd_invariant(obj));
1342
1343         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1344                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1345                 l->lo_header->loh_attr |= LOHA_EXISTS;
1346                 return 0;
1347         }
1348
1349         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1350         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1351         if (result == 0 && obj->oo_inode != NULL) {
1352                 struct osd_thread_info *oti = osd_oti_get(env);
1353                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1354
1355                 osd_object_init0(obj);
1356                 if (unlikely(obj->oo_header))
1357                         return 0;
1358
1359                 result = osd_get_lma(oti, obj->oo_inode,
1360                                      &oti->oti_obj_dentry, loa);
1361                 if (!result) {
1362                         /* Convert LMAI flags to lustre LMA flags
1363                          * and cache it to oo_lma_flags */
1364                         obj->oo_lma_flags =
1365                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1366                 } else if (result == -ENODATA) {
1367                         result = 0;
1368                 }
1369         }
1370
1371         LINVRNT(osd_invariant(obj));
1372         return result;
1373 }
1374
1375 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1376  * The left part is for value, binary mode. */
1377 struct osd_xattr_entry {
1378         struct list_head        oxe_list;
1379         size_t                  oxe_len;
1380         size_t                  oxe_namelen;
1381         bool                    oxe_exist;
1382         struct rcu_head         oxe_rcu;
1383         char                    oxe_buf[0];
1384 };
1385
1386 static int osd_oxc_get(struct osd_object *obj, const char *name,
1387                        struct lu_buf *buf)
1388 {
1389         struct osd_xattr_entry *tmp;
1390         struct osd_xattr_entry *oxe = NULL;
1391         size_t namelen = strlen(name);
1392         int rc;
1393         ENTRY;
1394
1395         rcu_read_lock();
1396         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1397                 if (namelen == tmp->oxe_namelen &&
1398                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1399                         oxe = tmp;
1400                         break;
1401                 }
1402         }
1403
1404         if (oxe == NULL)
1405                 GOTO(out, rc = -ENOENT);
1406
1407         if (!oxe->oxe_exist)
1408                 GOTO(out, rc = -ENODATA);
1409
1410         /* vallen */
1411         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1412         LASSERT(rc > 0);
1413
1414         if (buf->lb_buf == NULL)
1415                 GOTO(out, rc);
1416
1417         if (buf->lb_len < rc)
1418                 GOTO(out, rc = -ERANGE);
1419
1420         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1421         EXIT;
1422 out:
1423         rcu_read_unlock();
1424
1425         return rc;
1426 }
1427
1428 static void osd_oxc_free(struct rcu_head *head)
1429 {
1430         struct osd_xattr_entry *oxe;
1431
1432         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1433         OBD_FREE(oxe, oxe->oxe_len);
1434 }
1435
1436 static void osd_oxc_add(struct osd_object *obj, const char *name,
1437                         const char *buf, int buflen)
1438 {
1439         struct osd_xattr_entry *oxe;
1440         struct osd_xattr_entry *old = NULL;
1441         struct osd_xattr_entry *tmp;
1442         size_t namelen = strlen(name);
1443         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1444
1445         OBD_ALLOC(oxe, len);
1446         if (oxe == NULL)
1447                 return;
1448
1449         INIT_LIST_HEAD(&oxe->oxe_list);
1450         oxe->oxe_len = len;
1451         oxe->oxe_namelen = namelen;
1452         memcpy(oxe->oxe_buf, name, namelen);
1453         if (buflen > 0) {
1454                 LASSERT(buf != NULL);
1455                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1456                 oxe->oxe_exist = true;
1457         } else {
1458                 oxe->oxe_exist = false;
1459         }
1460
1461         /* this should be rarely called, just remove old and add new */
1462         spin_lock(&obj->oo_guard);
1463         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1464                 if (namelen == tmp->oxe_namelen &&
1465                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1466                         old = tmp;
1467                         break;
1468                 }
1469         }
1470         if (old != NULL) {
1471                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1472                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1473         } else {
1474                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1475         }
1476         spin_unlock(&obj->oo_guard);
1477 }
1478
1479 static void osd_oxc_del(struct osd_object *obj, const char *name)
1480 {
1481         struct osd_xattr_entry *oxe;
1482         size_t namelen = strlen(name);
1483
1484         spin_lock(&obj->oo_guard);
1485         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1486                 if (namelen == oxe->oxe_namelen &&
1487                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1488                         list_del_rcu(&oxe->oxe_list);
1489                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1490                         break;
1491                 }
1492         }
1493         spin_unlock(&obj->oo_guard);
1494 }
1495
1496 static void osd_oxc_fini(struct osd_object *obj)
1497 {
1498         struct osd_xattr_entry *oxe, *next;
1499
1500         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1501                 list_del(&oxe->oxe_list);
1502                 OBD_FREE(oxe, oxe->oxe_len);
1503         }
1504 }
1505
1506 /*
1507  * Concurrency: no concurrent access is possible that late in object
1508  * life-cycle.
1509  */
1510 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1511 {
1512         struct osd_object *obj = osd_obj(l);
1513         struct lu_object_header *h = obj->oo_header;
1514
1515         LINVRNT(osd_invariant(obj));
1516
1517         osd_oxc_fini(obj);
1518         dt_object_fini(&obj->oo_dt);
1519         if (obj->oo_hl_head != NULL)
1520                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1521         OBD_FREE_PTR(obj);
1522         if (unlikely(h)) {
1523                 lu_object_header_fini(h);
1524                 OBD_FREE_PTR(h);
1525         }
1526 }
1527
1528 /*
1529  * Concurrency: no concurrent access is possible that late in object
1530  * life-cycle.
1531  */
1532 static void osd_index_fini(struct osd_object *o)
1533 {
1534         struct iam_container *bag;
1535
1536         if (o->oo_dir != NULL) {
1537                 bag = &o->oo_dir->od_container;
1538                 if (o->oo_inode != NULL) {
1539                         if (bag->ic_object == o->oo_inode)
1540                                 iam_container_fini(bag);
1541                 }
1542                 OBD_FREE_PTR(o->oo_dir);
1543                 o->oo_dir = NULL;
1544         }
1545 }
1546
1547 /*
1548  * Concurrency: no concurrent access is possible that late in object
1549  * life-cycle (for all existing callers, that is. New callers have to provide
1550  * their own locking.)
1551  */
1552 static int osd_inode_unlinked(const struct inode *inode)
1553 {
1554         return inode->i_nlink == 0;
1555 }
1556
1557 enum {
1558         OSD_TXN_OI_DELETE_CREDITS    = 20,
1559         OSD_TXN_INODE_DELETE_CREDITS = 20
1560 };
1561
1562 /*
1563  * Journal
1564  */
1565
1566 #if OSD_THANDLE_STATS
1567 /**
1568  * Set time when the handle is allocated
1569  */
1570 static void osd_th_alloced(struct osd_thandle *oth)
1571 {
1572         oth->oth_alloced = ktime_get();
1573 }
1574
1575 /**
1576  * Set time when the handle started
1577  */
1578 static void osd_th_started(struct osd_thandle *oth)
1579 {
1580         oth->oth_started = ktime_get();
1581 }
1582
1583 /**
1584  * Check whether the we deal with this handle for too long.
1585  */
1586 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1587                                 ktime_t alloced, ktime_t started,
1588                                 ktime_t closed)
1589 {
1590         ktime_t now = ktime_get();
1591
1592         LASSERT(dev != NULL);
1593
1594         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1595                             ktime_us_delta(started, alloced));
1596         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1597                             ktime_us_delta(closed, started));
1598         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1599                             ktime_us_delta(now, closed));
1600
1601         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1602                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1603                       oth, now, alloced, started, closed);
1604                 libcfs_debug_dumpstack(NULL);
1605         }
1606 }
1607
1608 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1609 {                                                                       \
1610         ktime_t __closed = ktime_get();                                 \
1611         ktime_t __alloced = oth->oth_alloced;                           \
1612         ktime_t __started = oth->oth_started;                           \
1613                                                                         \
1614         expr;                                                           \
1615         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1616 }
1617
1618 #else /* OSD_THANDLE_STATS */
1619
1620 #define osd_th_alloced(h)                  do {} while(0)
1621 #define osd_th_started(h)                  do {} while(0)
1622 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1623
1624 #endif /* OSD_THANDLE_STATS */
1625
1626 /*
1627  * Concurrency: doesn't access mutable data.
1628  */
1629 static int osd_param_is_not_sane(const struct osd_device *dev,
1630                                  const struct thandle *th)
1631 {
1632         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1633
1634         return oh->ot_credits > osd_transaction_size(dev);
1635 }
1636
1637 /*
1638  * Concurrency: shouldn't matter.
1639  */
1640 static void osd_trans_commit_cb(struct super_block *sb,
1641                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1642 {
1643         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1644         struct thandle     *th  = &oh->ot_super;
1645         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1646         struct dt_txn_commit_cb *dcb, *tmp;
1647
1648         LASSERT(oh->ot_handle == NULL);
1649
1650         if (error)
1651                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1652
1653         dt_txn_hook_commit(th);
1654
1655         /* call per-transaction callbacks if any */
1656         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1657                                  dcb_linkage) {
1658                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1659                          "commit callback entry: magic=%x name='%s'\n",
1660                          dcb->dcb_magic, dcb->dcb_name);
1661                 list_del_init(&dcb->dcb_linkage);
1662                 dcb->dcb_func(NULL, th, dcb, error);
1663         }
1664
1665         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1666         lu_device_put(lud);
1667         th->th_dev = NULL;
1668
1669         OBD_FREE_PTR(oh);
1670 }
1671
1672 #ifndef HAVE_SB_START_WRITE
1673 # define sb_start_write(sb) do {} while (0)
1674 # define sb_end_write(sb) do {} while (0)
1675 #endif
1676
1677 static struct thandle *osd_trans_create(const struct lu_env *env,
1678                                         struct dt_device *d)
1679 {
1680         struct osd_thread_info  *oti = osd_oti_get(env);
1681         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1682         struct osd_thandle      *oh;
1683         struct thandle          *th;
1684         ENTRY;
1685
1686         if (d->dd_rdonly) {
1687                 CERROR("%s: someone try to start transaction under "
1688                        "readonly mode, should be disabled.\n",
1689                        osd_name(osd_dt_dev(d)));
1690                 dump_stack();
1691                 RETURN(ERR_PTR(-EROFS));
1692         }
1693
1694         /* on pending IO in this thread should left from prev. request */
1695         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1696
1697         sb_start_write(osd_sb(osd_dt_dev(d)));
1698
1699         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1700         if (!oh) {
1701                 sb_end_write(osd_sb(osd_dt_dev(d)));
1702                 RETURN(ERR_PTR(-ENOMEM));
1703         }
1704
1705         oh->ot_quota_trans = &oti->oti_quota_trans;
1706         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1707         th = &oh->ot_super;
1708         th->th_dev = d;
1709         th->th_result = 0;
1710         oh->ot_credits = 0;
1711         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1712         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1713         osd_th_alloced(oh);
1714
1715         memset(oti->oti_declare_ops, 0,
1716                sizeof(oti->oti_declare_ops));
1717         memset(oti->oti_declare_ops_cred, 0,
1718                sizeof(oti->oti_declare_ops_cred));
1719         memset(oti->oti_declare_ops_used, 0,
1720                sizeof(oti->oti_declare_ops_used));
1721
1722         oti->oti_ins_cache_depth++;
1723
1724         RETURN(th);
1725 }
1726
1727 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1728 {
1729         struct osd_thread_info  *oti = osd_oti_get(env);
1730         struct osd_thandle      *oh;
1731
1732         oh = container_of0(th, struct osd_thandle, ot_super);
1733         LASSERT(oh != NULL);
1734
1735         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1736               oti->oti_declare_ops[OSD_OT_CREATE],
1737               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1738               oti->oti_declare_ops_used[OSD_OT_CREATE],
1739               oti->oti_declare_ops[OSD_OT_DESTROY],
1740               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1741               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1742         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1743               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1744               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1745               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1746               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1747               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1748               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1749         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1750               oti->oti_declare_ops[OSD_OT_WRITE],
1751               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1752               oti->oti_declare_ops_used[OSD_OT_WRITE],
1753               oti->oti_declare_ops[OSD_OT_PUNCH],
1754               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1755               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1756               oti->oti_declare_ops[OSD_OT_QUOTA],
1757               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1758               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1759         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1760               oti->oti_declare_ops[OSD_OT_INSERT],
1761               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1762               oti->oti_declare_ops_used[OSD_OT_INSERT],
1763               oti->oti_declare_ops[OSD_OT_DELETE],
1764               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1765               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1766         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1767               oti->oti_declare_ops[OSD_OT_REF_ADD],
1768               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1769               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1770               oti->oti_declare_ops[OSD_OT_REF_DEL],
1771               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1772               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1773 }
1774
1775 /*
1776  * Concurrency: shouldn't matter.
1777  */
1778 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1779                            struct thandle *th)
1780 {
1781         struct osd_thread_info *oti = osd_oti_get(env);
1782         struct osd_device  *dev = osd_dt_dev(d);
1783         handle_t           *jh;
1784         struct osd_thandle *oh;
1785         int rc;
1786
1787         ENTRY;
1788
1789         LASSERT(current->journal_info == NULL);
1790
1791         oh = container_of0(th, struct osd_thandle, ot_super);
1792         LASSERT(oh != NULL);
1793         LASSERT(oh->ot_handle == NULL);
1794
1795         rc = dt_txn_hook_start(env, d, th);
1796         if (rc != 0)
1797                 GOTO(out, rc);
1798
1799         if (unlikely(osd_param_is_not_sane(dev, th))) {
1800                 static unsigned long last_printed;
1801                 static int last_credits;
1802
1803                 /* don't make noise on a tiny testing systems
1804                  * actual credits misuse will be caught anyway */
1805                 if (last_credits != oh->ot_credits &&
1806                     time_after(jiffies, last_printed +
1807                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1808                     osd_transaction_size(dev) > 512) {
1809                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1810                               oh->ot_credits, osd_transaction_size(dev));
1811                         osd_trans_dump_creds(env, th);
1812                         libcfs_debug_dumpstack(NULL);
1813                         last_credits = oh->ot_credits;
1814                         last_printed = jiffies;
1815                 }
1816                 /* XXX Limit the credits to 'max_transaction_buffers', and
1817                  *     let the underlying filesystem to catch the error if
1818                  *     we really need so many credits.
1819                  *
1820                  *     This should be removed when we can calculate the
1821                  *     credits precisely. */
1822                 oh->ot_credits = osd_transaction_size(dev);
1823         } else if (ldiskfs_track_declares_assert != 0) {
1824                 /* reserve few credits to prevent an assertion in JBD
1825                  * our debugging mechanism will be able to detected
1826                  * overuse. this can help to debug single-update
1827                  * transactions */
1828                 oh->ot_credits += 10;
1829                 if (unlikely(osd_param_is_not_sane(dev, th)))
1830                         oh->ot_credits = osd_transaction_size(dev);
1831         }
1832
1833         /*
1834          * XXX temporary stuff. Some abstraction layer should
1835          * be used.
1836          */
1837         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1838         osd_th_started(oh);
1839         if (!IS_ERR(jh)) {
1840                 oh->ot_handle = jh;
1841                 LASSERT(oti->oti_txns == 0);
1842
1843                 lu_device_get(&d->dd_lu_dev);
1844                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1845                               "osd-tx", th);
1846                 oti->oti_txns++;
1847                 rc = 0;
1848         } else {
1849                 rc = PTR_ERR(jh);
1850         }
1851 out:
1852         RETURN(rc);
1853 }
1854
1855 static int osd_seq_exists(const struct lu_env *env,
1856                           struct osd_device *osd, u64 seq)
1857 {
1858         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1859         struct seq_server_site  *ss = osd_seq_site(osd);
1860         int                     rc;
1861         ENTRY;
1862
1863         LASSERT(ss != NULL);
1864         LASSERT(ss->ss_server_fld != NULL);
1865
1866         rc = osd_fld_lookup(env, osd, seq, range);
1867         if (rc != 0) {
1868                 if (rc != -ENOENT)
1869                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1870                                osd_name(osd), seq, rc);
1871                 RETURN(0);
1872         }
1873
1874         RETURN(ss->ss_node_id == range->lsr_index);
1875 }
1876
1877 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1878 {
1879         struct dt_txn_commit_cb *dcb;
1880         struct dt_txn_commit_cb *tmp;
1881
1882         /* call per-transaction stop callbacks if any */
1883         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1884                                  dcb_linkage) {
1885                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1886                          "commit callback entry: magic=%x name='%s'\n",
1887                          dcb->dcb_magic, dcb->dcb_name);
1888                 list_del_init(&dcb->dcb_linkage);
1889                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1890         }
1891 }
1892
1893 /*
1894  * Concurrency: shouldn't matter.
1895  */
1896 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1897                           struct thandle *th)
1898 {
1899         int                     rc = 0, remove_agents = 0;
1900         struct osd_thandle     *oh;
1901         struct osd_thread_info *oti = osd_oti_get(env);
1902         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1903         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1904         struct qsd_instance    *qsd = osd->od_quota_slave;
1905         struct lquota_trans    *qtrans;
1906         ENTRY;
1907
1908         oh = container_of0(th, struct osd_thandle, ot_super);
1909
1910         remove_agents = oh->ot_remove_agents;
1911
1912         qtrans = oh->ot_quota_trans;
1913         oh->ot_quota_trans = NULL;
1914
1915         if (oh->ot_handle != NULL) {
1916                 int rc2;
1917                 handle_t *hdl = oh->ot_handle;
1918
1919                 /*
1920                  * add commit callback
1921                  * notice we don't do this in osd_trans_start()
1922                  * as underlying transaction can change during truncate
1923                  */
1924                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1925                                          &oh->ot_jcb);
1926
1927                 LASSERT(oti->oti_txns == 1);
1928                 oti->oti_txns--;
1929
1930                 rc = dt_txn_hook_stop(env, th);
1931                 if (rc != 0)
1932                         CERROR("%s: failed in transaction hook: rc = %d\n",
1933                                osd_name(osd), rc);
1934
1935                 osd_trans_stop_cb(oh, rc);
1936                 /* hook functions might modify th_sync */
1937                 hdl->h_sync = th->th_sync;
1938
1939                 oh->ot_handle = NULL;
1940                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1941                 if (rc2 != 0)
1942                         CERROR("%s: failed to stop transaction: rc = %d\n",
1943                                osd_name(osd), rc2);
1944                 if (!rc)
1945                         rc = rc2;
1946         } else {
1947                 osd_trans_stop_cb(oh, th->th_result);
1948                 OBD_FREE_PTR(oh);
1949         }
1950
1951         /* inform the quota slave device that the transaction is stopping */
1952         qsd_op_end(env, qsd, qtrans);
1953
1954         /* as we want IO to journal and data IO be concurrent, we don't block
1955          * awaiting data IO completion in osd_do_bio(), instead we wait here
1956          * once transaction is submitted to the journal. all reqular requests
1957          * don't do direct IO (except read/write), thus this wait_event becomes
1958          * no-op for them.
1959          *
1960          * IMPORTANT: we have to wait till any IO submited by the thread is
1961          * completed otherwise iobuf may be corrupted by different request
1962          */
1963         wait_event(iobuf->dr_wait,
1964                        atomic_read(&iobuf->dr_numreqs) == 0);
1965         osd_fini_iobuf(osd, iobuf);
1966         if (!rc)
1967                 rc = iobuf->dr_error;
1968
1969         if (unlikely(remove_agents != 0))
1970                 osd_process_scheduled_agent_removals(env, osd);
1971
1972         oti->oti_ins_cache_depth--;
1973         /* reset OI cache for safety */
1974         if (oti->oti_ins_cache_depth == 0)
1975                 oti->oti_ins_cache_used = 0;
1976
1977         sb_end_write(osd_sb(osd));
1978
1979         RETURN(rc);
1980 }
1981
1982 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1983 {
1984         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1985                                                ot_super);
1986
1987         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1988         LASSERT(&dcb->dcb_func != NULL);
1989         if (dcb->dcb_flags & DCB_TRANS_STOP)
1990                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1991         else
1992                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1993
1994         return 0;
1995 }
1996
1997 /*
1998  * Called just before object is freed. Releases all resources except for
1999  * object itself (that is released by osd_object_free()).
2000  *
2001  * Concurrency: no concurrent access is possible that late in object
2002  * life-cycle.
2003  */
2004 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2005 {
2006         struct osd_object *obj   = osd_obj(l);
2007         struct inode      *inode = obj->oo_inode;
2008
2009         LINVRNT(osd_invariant(obj));
2010
2011         /*
2012          * If object is unlinked remove fid->ino mapping from object index.
2013          */
2014
2015         osd_index_fini(obj);
2016         if (inode != NULL) {
2017                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
2018                 qid_t                    uid = i_uid_read(inode);
2019                 qid_t                    gid = i_gid_read(inode);
2020
2021                 obj->oo_inode = NULL;
2022                 iput(inode);
2023                 if (!obj->oo_header && qsd) {
2024                         struct osd_thread_info  *info = osd_oti_get(env);
2025                         struct lquota_id_info   *qi = &info->oti_qi;
2026
2027                         /* Release granted quota to master if necessary */
2028                         qi->lqi_id.qid_uid = uid;
2029                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2030
2031                         qi->lqi_id.qid_uid = gid;
2032                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2033
2034                         qi->lqi_id.qid_uid = i_projid_read(inode);
2035                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2036                 }
2037         }
2038 }
2039
2040 /*
2041  * Concurrency: ->loo_object_release() is called under site spin-lock.
2042  */
2043 static void osd_object_release(const struct lu_env *env,
2044                                struct lu_object *l)
2045 {
2046         struct osd_object *o = osd_obj(l);
2047         /* nobody should be releasing a non-destroyed object with nlink=0
2048          * the API allows this, but ldiskfs doesn't like and then report
2049          * this inode as deleted */
2050         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
2051                 LBUG();
2052 }
2053
2054 /*
2055  * Concurrency: shouldn't matter.
2056  */
2057 static int osd_object_print(const struct lu_env *env, void *cookie,
2058                             lu_printer_t p, const struct lu_object *l)
2059 {
2060         struct osd_object *o = osd_obj(l);
2061         struct iam_descr  *d;
2062
2063         if (o->oo_dir != NULL)
2064                 d = o->oo_dir->od_container.ic_descr;
2065         else
2066                 d = NULL;
2067         return (*p)(env, cookie,
2068                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2069                     o, o->oo_inode,
2070                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2071                     o->oo_inode ? o->oo_inode->i_generation : 0,
2072                     d ? d->id_ops->id_name : "plain");
2073 }
2074
2075 /*
2076  * Concurrency: shouldn't matter.
2077  */
2078 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2079                struct obd_statfs *sfs)
2080 {
2081         struct osd_device       *osd = osd_dt_dev(d);
2082         struct super_block      *sb = osd_sb(osd);
2083         struct kstatfs          *ksfs;
2084         __u64                    reserved;
2085         int                      result = 0;
2086
2087         if (unlikely(osd->od_mnt == NULL))
2088                 return -EINPROGRESS;
2089
2090         /* osd_lproc.c call this without env, allocate ksfs for that case */
2091         if (unlikely(env == NULL)) {
2092                 OBD_ALLOC_PTR(ksfs);
2093                 if (ksfs == NULL)
2094                         return -ENOMEM;
2095         } else {
2096                 ksfs = &osd_oti_get(env)->oti_ksfs;
2097         }
2098
2099         result = sb->s_op->statfs(sb->s_root, ksfs);
2100         if (result)
2101                 goto out;
2102
2103         statfs_pack(sfs, ksfs);
2104         if (unlikely(sb->s_flags & MS_RDONLY))
2105                 sfs->os_state |= OS_STATE_READONLY;
2106         if (ldiskfs_has_feature_extents(sb))
2107                 sfs->os_maxbytes = sb->s_maxbytes;
2108         else
2109                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2110
2111         /*
2112          * Reserve some space so to avoid fragmenting the filesystem too much.
2113          * Fragmentation not only impacts performance, but can also increase
2114          * metadata overhead significantly, causing grant calculation to be
2115          * wrong.
2116          *
2117          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2118          */
2119         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2120         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2121         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2122                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2123
2124         sfs->os_blocks -= reserved;
2125         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2126         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2127
2128 out:
2129         if (unlikely(env == NULL))
2130                 OBD_FREE_PTR(ksfs);
2131         return result;
2132 }
2133
2134 /**
2135  * Estimate space needed for file creations. We assume the largest filename
2136  * which is 2^64 - 1, hence a filename of 20 chars.
2137  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2138  */
2139 #ifdef __LDISKFS_DIR_REC_LEN
2140 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2141 #else
2142 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2143 #endif
2144
2145 /*
2146  * Concurrency: doesn't access mutable data.
2147  */
2148 static void osd_conf_get(const struct lu_env *env,
2149                          const struct dt_device *dev,
2150                          struct dt_device_param *param)
2151 {
2152         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2153         int                ea_overhead;
2154
2155         /*
2156          * XXX should be taken from not-yet-existing fs abstraction layer.
2157          */
2158         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2159         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2160         param->ddp_symlink_max  = sb->s_blocksize;
2161         param->ddp_mount_type     = LDD_MT_LDISKFS;
2162         if (ldiskfs_has_feature_extents(sb))
2163                 param->ddp_maxbytes = sb->s_maxbytes;
2164         else
2165                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2166         /* inode are statically allocated, so per-inode space consumption
2167          * is the space consumed by the directory entry */
2168         param->ddp_inodespace     = PER_OBJ_USAGE;
2169         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2170          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2171          * maximum length to not under count the actual number of extents
2172          * needed for writing a file. */
2173         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2174         /* worst-case extent insertion metadata overhead */
2175         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2176         param->ddp_mntopts      = 0;
2177         if (test_opt(sb, XATTR_USER))
2178                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2179         if (test_opt(sb, POSIX_ACL))
2180                 param->ddp_mntopts |= MNTOPT_ACL;
2181
2182         /* LOD might calculate the max stripe count based on max_ea_size,
2183          * so we need take account in the overhead as well,
2184          * xattr_header + magic + xattr_entry_head */
2185         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2186                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2187
2188 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2189         if (ldiskfs_has_feature_ea_inode(sb))
2190                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2191                                                                 ea_overhead;
2192         else
2193 #endif
2194                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2195
2196         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2197                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2198
2199         /* Preferred RPC size for efficient disk IO.  4MB shows good
2200          * all-around performance for ldiskfs, but use bigalloc chunk size
2201          * by default if larger. */
2202 #if defined(LDISKFS_CLUSTER_SIZE)
2203         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2204                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2205         else
2206 #endif
2207                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2208 }
2209
2210 /*
2211  * Concurrency: shouldn't matter.
2212  */
2213 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2214 {
2215         int rc;
2216
2217         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2218
2219         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2220
2221         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2222                rc);
2223
2224         return rc;
2225 }
2226
2227 /**
2228  * Start commit for OSD device.
2229  *
2230  * An implementation of dt_commit_async method for OSD device.
2231  * Asychronously starts underlayng fs sync and thereby a transaction
2232  * commit.
2233  *
2234  * \param env environment
2235  * \param d dt device
2236  *
2237  * \see dt_device_operations
2238  */
2239 static int osd_commit_async(const struct lu_env *env,
2240                             struct dt_device *d)
2241 {
2242         struct super_block *s = osd_sb(osd_dt_dev(d));
2243         ENTRY;
2244
2245         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2246         RETURN(s->s_op->sync_fs(s, 0));
2247 }
2248
2249 /* Our own copy of the set readonly functions if present, or NU if not. */
2250 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2251 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2252 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2253
2254 /*
2255  * Concurrency: shouldn't matter.
2256  */
2257 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2258 {
2259         struct super_block *sb = osd_sb(osd_dt_dev(d));
2260         struct block_device *dev = sb->s_bdev;
2261         int rc = -EOPNOTSUPP;
2262         ENTRY;
2263
2264         if (priv_dev_set_rdonly) {
2265                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2266
2267                 rc = 0;
2268                 CERROR("*** setting %s read-only ***\n",
2269                        osd_dt_dev(d)->od_svname);
2270
2271                 if (sb->s_op->freeze_fs) {
2272                         rc = sb->s_op->freeze_fs(sb);
2273                         if (rc)
2274                                 goto out;
2275                 }
2276
2277                 if (jdev && (jdev != dev)) {
2278                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2279                                (long)jdev);
2280                         priv_dev_set_rdonly(jdev);
2281                 }
2282                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2283                 priv_dev_set_rdonly(dev);
2284
2285                 if (sb->s_op->unfreeze_fs)
2286                         sb->s_op->unfreeze_fs(sb);
2287         }
2288
2289 out:
2290         if (rc)
2291                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2292                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2293
2294         RETURN(rc);
2295 }
2296
2297 /**
2298  * Note: we do not count into QUOTA here.
2299  * If we mount with --data_journal we may need more.
2300  */
2301 const int osd_dto_credits_noquota[DTO_NR] = {
2302         /**
2303          * Insert.
2304          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2305          * SINGLEDATA_TRANS_BLOCKS(8)
2306          * XXX Note: maybe iam need more, since iam have more level than
2307          *           EXT3 htree.
2308          */
2309         [DTO_INDEX_INSERT]  = 16,
2310         /**
2311          * Delete
2312          * just modify a single entry, probably merge few within a block
2313          */
2314         [DTO_INDEX_DELETE]  = 1,
2315         /**
2316          * Used for OI scrub
2317          */
2318         [DTO_INDEX_UPDATE]  = 16,
2319         /**
2320          * 4(inode, inode bits, groups, GDT)
2321          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2322          */
2323         [DTO_OBJECT_CREATE] = 4,
2324         /**
2325          * 4(inode, inode bits, groups, GDT)
2326          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2327          */
2328         [DTO_OBJECT_DELETE] = 4,
2329         /**
2330          * Attr set credits (inode)
2331          */
2332         [DTO_ATTR_SET_BASE] = 1,
2333         /**
2334          * Xattr set. The same as xattr of EXT3.
2335          * DATA_TRANS_BLOCKS(14)
2336          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2337          * are also counted in. Do not know why?
2338          */
2339         [DTO_XATTR_SET]     = 14,
2340         /**
2341          * credits for inode change during write.
2342          */
2343         [DTO_WRITE_BASE]    = 3,
2344         /**
2345          * credits for single block write.
2346          */
2347         [DTO_WRITE_BLOCK]   = 14,
2348         /**
2349          * Attr set credits for chown.
2350          * This is extra credits for setattr, and it is null without quota
2351          */
2352         [DTO_ATTR_SET_CHOWN] = 0
2353 };
2354
2355 static const struct dt_device_operations osd_dt_ops = {
2356         .dt_root_get       = osd_root_get,
2357         .dt_statfs         = osd_statfs,
2358         .dt_trans_create   = osd_trans_create,
2359         .dt_trans_start    = osd_trans_start,
2360         .dt_trans_stop     = osd_trans_stop,
2361         .dt_trans_cb_add   = osd_trans_cb_add,
2362         .dt_conf_get       = osd_conf_get,
2363         .dt_sync           = osd_sync,
2364         .dt_ro             = osd_ro,
2365         .dt_commit_async   = osd_commit_async,
2366 };
2367
2368 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2369                           unsigned role)
2370 {
2371         struct osd_object *obj = osd_dt_obj(dt);
2372         struct osd_thread_info *oti = osd_oti_get(env);
2373
2374         LINVRNT(osd_invariant(obj));
2375
2376         LASSERT(obj->oo_owner != env);
2377         down_read_nested(&obj->oo_sem, role);
2378
2379         LASSERT(obj->oo_owner == NULL);
2380         oti->oti_r_locks++;
2381 }
2382
2383 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2384                            unsigned role)
2385 {
2386         struct osd_object *obj = osd_dt_obj(dt);
2387         struct osd_thread_info *oti = osd_oti_get(env);
2388
2389         LINVRNT(osd_invariant(obj));
2390
2391         LASSERT(obj->oo_owner != env);
2392         down_write_nested(&obj->oo_sem, role);
2393
2394         LASSERT(obj->oo_owner == NULL);
2395         obj->oo_owner = env;
2396         oti->oti_w_locks++;
2397 }
2398
2399 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2400 {
2401         struct osd_object *obj = osd_dt_obj(dt);
2402         struct osd_thread_info *oti = osd_oti_get(env);
2403
2404         LINVRNT(osd_invariant(obj));
2405
2406         LASSERT(oti->oti_r_locks > 0);
2407         oti->oti_r_locks--;
2408         up_read(&obj->oo_sem);
2409 }
2410
2411 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2412 {
2413         struct osd_object *obj = osd_dt_obj(dt);
2414         struct osd_thread_info *oti = osd_oti_get(env);
2415
2416         LINVRNT(osd_invariant(obj));
2417
2418         LASSERT(obj->oo_owner == env);
2419         LASSERT(oti->oti_w_locks > 0);
2420         oti->oti_w_locks--;
2421         obj->oo_owner = NULL;
2422         up_write(&obj->oo_sem);
2423 }
2424
2425 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2426 {
2427         struct osd_object *obj = osd_dt_obj(dt);
2428
2429         LINVRNT(osd_invariant(obj));
2430
2431         return obj->oo_owner == env;
2432 }
2433
2434 static struct timespec *osd_inode_time(const struct lu_env *env,
2435                                        struct inode *inode, __u64 seconds)
2436 {
2437         struct osd_thread_info  *oti = osd_oti_get(env);
2438         struct timespec         *t   = &oti->oti_time;
2439
2440         t->tv_sec = seconds;
2441         t->tv_nsec = 0;
2442         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2443         return t;
2444 }
2445
2446 static void osd_inode_getattr(const struct lu_env *env,
2447                               struct inode *inode, struct lu_attr *attr)
2448 {
2449         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2450                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2451                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2452                            LA_BLKSIZE | LA_TYPE;
2453
2454         attr->la_atime   = LTIME_S(inode->i_atime);
2455         attr->la_mtime   = LTIME_S(inode->i_mtime);
2456         attr->la_ctime   = LTIME_S(inode->i_ctime);
2457         attr->la_mode    = inode->i_mode;
2458         attr->la_size    = i_size_read(inode);
2459         attr->la_blocks  = inode->i_blocks;
2460         attr->la_uid     = i_uid_read(inode);
2461         attr->la_gid     = i_gid_read(inode);
2462         attr->la_projid  = i_projid_read(inode);
2463         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2464         attr->la_nlink   = inode->i_nlink;
2465         attr->la_rdev    = inode->i_rdev;
2466         attr->la_blksize = 1 << inode->i_blkbits;
2467         attr->la_blkbits = inode->i_blkbits;
2468         /*
2469          * Ext4 did not transfer inherit flags from raw inode
2470          * to inode flags, and ext4 internally test raw inode
2471          * @i_flags directly. Instead of patching ext4, we do it here.
2472          */
2473         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2474                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2475 }
2476
2477 static int osd_attr_get(const struct lu_env *env,
2478                         struct dt_object *dt,
2479                         struct lu_attr *attr)
2480 {
2481         struct osd_object *obj = osd_dt_obj(dt);
2482
2483         if (unlikely(!dt_object_exists(dt)))
2484                 return -ENOENT;
2485         if (unlikely(obj->oo_destroyed))
2486                 return -ENOENT;
2487
2488         LASSERT(!dt_object_remote(dt));
2489         LINVRNT(osd_invariant(obj));
2490
2491         spin_lock(&obj->oo_guard);
2492         osd_inode_getattr(env, obj->oo_inode, attr);
2493         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2494                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2495         spin_unlock(&obj->oo_guard);
2496
2497         return 0;
2498 }
2499
2500 static int osd_declare_attr_qid(const struct lu_env *env,
2501                                 struct osd_object *obj,
2502                                 struct osd_thandle *oh, long long bspace,
2503                                 qid_t old_id, qid_t new_id, bool enforce,
2504                                 unsigned type, bool ignore_edquot)
2505 {
2506         int rc;
2507         struct osd_thread_info *info = osd_oti_get(env);
2508         struct lquota_id_info  *qi = &info->oti_qi;
2509
2510         qi->lqi_type = type;
2511         /* inode accounting */
2512         qi->lqi_is_blk = false;
2513
2514         /* one more inode for the new id ... */
2515         qi->lqi_id.qid_uid = new_id;
2516         qi->lqi_space      = 1;
2517         /* Reserve credits for the new id */
2518         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2519         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2520                 rc = 0;
2521         if (rc)
2522                 RETURN(rc);
2523
2524         /* and one less inode for the current id */
2525         qi->lqi_id.qid_uid = old_id;
2526         qi->lqi_space      = -1;
2527         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2528         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2529                 rc = 0;
2530         if (rc)
2531                 RETURN(rc);
2532
2533         /* block accounting */
2534         qi->lqi_is_blk = true;
2535
2536         /* more blocks for the new id ... */
2537         qi->lqi_id.qid_uid = new_id;
2538         qi->lqi_space      = bspace;
2539         /*
2540          * Credits for the new uid has been reserved, re-use "obj"
2541          * to save credit reservation.
2542          */
2543         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2544         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2545                 rc = 0;
2546         if (rc)
2547                 RETURN(rc);
2548
2549         /* and finally less blocks for the current uid */
2550         qi->lqi_id.qid_uid = old_id;
2551         qi->lqi_space      = -bspace;
2552         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2553         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2554                 rc = 0;
2555
2556         RETURN(rc);
2557 }
2558
2559 static int osd_declare_attr_set(const struct lu_env *env,
2560                                 struct dt_object *dt,
2561                                 const struct lu_attr *attr,
2562                                 struct thandle *handle)
2563 {
2564         struct osd_thandle     *oh;
2565         struct osd_object      *obj;
2566         qid_t                   uid;
2567         qid_t                   gid;
2568         long long               bspace;
2569         int                     rc = 0;
2570         bool                    enforce;
2571         ENTRY;
2572
2573         LASSERT(dt != NULL);
2574         LASSERT(handle != NULL);
2575
2576         obj = osd_dt_obj(dt);
2577         LASSERT(osd_invariant(obj));
2578
2579         oh = container_of0(handle, struct osd_thandle, ot_super);
2580         LASSERT(oh->ot_handle == NULL);
2581
2582         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2583                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2584
2585         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2586                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2587
2588         if (attr == NULL || obj->oo_inode == NULL)
2589                 RETURN(rc);
2590
2591         bspace   = obj->oo_inode->i_blocks << 9;
2592         bspace   = toqb(bspace);
2593
2594         /* Changing ownership is always preformed by super user, it should not
2595          * fail with EDQUOT unless required explicitly.
2596          *
2597          * We still need to call the osd_declare_qid() to calculate the journal
2598          * credits for updating quota accounting files and to trigger quota
2599          * space adjustment once the operation is completed.*/
2600         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2601                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2602
2603                 if (!ignore_edquot)
2604                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2605                                "(the quota space is %lld)\n",
2606                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2607                                attr->la_gid, bspace);
2608
2609                 /* USERQUOTA */
2610                 uid = i_uid_read(obj->oo_inode);
2611                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2612                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2613                                           attr->la_uid, enforce, USRQUOTA,
2614                                           true);
2615                 if (rc)
2616                         RETURN(rc);
2617
2618                 gid = i_gid_read(obj->oo_inode);
2619                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2620                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2621                                           i_gid_read(obj->oo_inode),
2622                                           attr->la_gid, enforce, GRPQUOTA,
2623                                           ignore_edquot);
2624                 if (rc)
2625                         RETURN(rc);
2626
2627         }
2628 #ifdef HAVE_PROJECT_QUOTA
2629         if (attr->la_valid & LA_PROJID) {
2630                 __u32 projid = i_projid_read(obj->oo_inode);
2631                 enforce = (attr->la_valid & LA_PROJID) &&
2632                                         (attr->la_projid != projid);
2633                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2634                                           (qid_t)projid, (qid_t)attr->la_projid,
2635                                           enforce, PRJQUOTA, true);
2636                 if (rc)
2637                         RETURN(rc);
2638         }
2639 #endif
2640         RETURN(rc);
2641 }
2642
2643 static int osd_inode_setattr(const struct lu_env *env,
2644                              struct inode *inode, const struct lu_attr *attr)
2645 {
2646         __u64 bits = attr->la_valid;
2647
2648         /* Only allow set size for regular file */
2649         if (!S_ISREG(inode->i_mode))
2650                 bits &= ~(LA_SIZE | LA_BLOCKS);
2651
2652         if (bits == 0)
2653                 return 0;
2654
2655         if (bits & LA_ATIME)
2656                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2657         if (bits & LA_CTIME)
2658                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2659         if (bits & LA_MTIME)
2660                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2661         if (bits & LA_SIZE) {
2662                 spin_lock(&inode->i_lock);
2663                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2664                 i_size_write(inode, attr->la_size);
2665                 spin_unlock(&inode->i_lock);
2666         }
2667
2668         /* OSD should not change "i_blocks" which is used by quota.
2669          * "i_blocks" should be changed by ldiskfs only. */
2670         if (bits & LA_MODE)
2671                 inode->i_mode = (inode->i_mode & S_IFMT) |
2672                                 (attr->la_mode & ~S_IFMT);
2673         if (bits & LA_UID)
2674                 i_uid_write(inode, attr->la_uid);
2675         if (bits & LA_GID)
2676                 i_gid_write(inode, attr->la_gid);
2677         if (bits & LA_PROJID)
2678                 i_projid_write(inode, attr->la_projid);
2679         if (bits & LA_NLINK)
2680                 set_nlink(inode, attr->la_nlink);
2681         if (bits & LA_RDEV)
2682                 inode->i_rdev = attr->la_rdev;
2683
2684         if (bits & LA_FLAGS) {
2685                 /* always keep S_NOCMTIME */
2686                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2687                                  S_NOCMTIME;
2688                 /*
2689                  * Ext4 did not transfer inherit flags from
2690                  * @inode->i_flags to raw inode i_flags when writing
2691                  * flags, we do it explictly here.
2692                  */
2693                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2694                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2695                 else
2696                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2697         }
2698         return 0;
2699 }
2700
2701 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2702 {
2703         int rc;
2704
2705         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2706             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2707                 struct iattr    iattr;
2708
2709                 ll_vfs_dq_init(inode);
2710                 iattr.ia_valid = 0;
2711                 if (attr->la_valid & LA_UID)
2712                         iattr.ia_valid |= ATTR_UID;
2713                 if (attr->la_valid & LA_GID)
2714                         iattr.ia_valid |= ATTR_GID;
2715                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2716                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2717
2718                 rc = ll_vfs_dq_transfer(inode, &iattr);
2719                 if (rc) {
2720                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2721                                "enforcement enabled on the ldiskfs "
2722                                "filesystem?\n", inode->i_sb->s_id, rc);
2723                         return rc;
2724                 }
2725         }
2726
2727         /* Handle project id transfer here properly */
2728         if (attr->la_valid & LA_PROJID &&
2729             attr->la_projid != i_projid_read(inode)) {
2730 #ifdef HAVE_PROJECT_QUOTA
2731                 rc = ldiskfs_transfer_project(inode, attr->la_projid);
2732 #else
2733                 rc = -ENOTSUPP;
2734 #endif
2735                 if (rc) {
2736                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2737                                "enforcement enabled on the ldiskfs "
2738                                "filesystem?\n", inode->i_sb->s_id, rc);
2739                         return rc;
2740                 }
2741         }
2742         return 0;
2743 }
2744
2745 static int osd_attr_set(const struct lu_env *env,
2746                         struct dt_object *dt,
2747                         const struct lu_attr *attr,
2748                         struct thandle *handle)
2749 {
2750         struct osd_object *obj = osd_dt_obj(dt);
2751         struct inode      *inode;
2752         int rc;
2753
2754         if (!dt_object_exists(dt))
2755                 return -ENOENT;
2756
2757         LASSERT(handle != NULL);
2758         LASSERT(!dt_object_remote(dt));
2759         LASSERT(osd_invariant(obj));
2760
2761         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2762
2763         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2764             !osd_obj2dev(obj)->od_is_ost) {
2765                 struct osd_thread_info  *oti  = osd_oti_get(env);
2766                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2767                 struct lu_fid           *fid1 = &oti->oti_fid;
2768                 struct osd_inode_id     *id   = &oti->oti_id;
2769                 struct iam_path_descr   *ipd;
2770                 struct iam_container    *bag;
2771                 struct osd_thandle      *oh;
2772                 int                      rc;
2773
2774                 fid_cpu_to_be(fid1, fid0);
2775                 memset(id, 1, sizeof(*id));
2776                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2777                                   fid0)->oi_dir.od_container;
2778                 ipd = osd_idx_ipd_get(env, bag);
2779                 if (unlikely(ipd == NULL))
2780                         RETURN(-ENOMEM);
2781
2782                 oh = container_of0(handle, struct osd_thandle, ot_super);
2783                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2784                                 (const struct iam_rec *)id, ipd);
2785                 osd_ipd_put(env, bag, ipd);
2786                 return(rc > 0 ? 0 : rc);
2787         }
2788
2789         inode = obj->oo_inode;
2790
2791         rc = osd_quota_transfer(inode, attr);
2792         if (rc)
2793                 return rc;
2794
2795         spin_lock(&obj->oo_guard);
2796         rc = osd_inode_setattr(env, inode, attr);
2797         spin_unlock(&obj->oo_guard);
2798         if (rc != 0)
2799                 GOTO(out, rc);
2800
2801         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2802
2803         if (!(attr->la_valid & LA_FLAGS))
2804                 GOTO(out, rc);
2805
2806         /* Let's check if there are extra flags need to be set into LMA */
2807         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2808                 struct osd_thread_info *info = osd_oti_get(env);
2809                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2810
2811                 LASSERT(!obj->oo_pfid_in_lma);
2812
2813                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2814                                  &info->oti_ost_attrs);
2815                 if (rc)
2816                         GOTO(out, rc);
2817
2818                 lma->lma_incompat |=
2819                         lustre_to_lma_flags(attr->la_flags);
2820                 lustre_lma_swab(lma);
2821                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2822                                      lma, sizeof(*lma), XATTR_REPLACE);
2823                 if (rc != 0) {
2824                         struct osd_device *osd = osd_obj2dev(obj);
2825
2826                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2827                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2828                               lma->lma_incompat, rc);
2829                 } else {
2830                         obj->oo_lma_flags =
2831                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2832                 }
2833                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2834         }
2835 out:
2836         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2837
2838         return rc;
2839 }
2840
2841 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2842                                            struct osd_object *obj,
2843                                            const char *name, const int namelen)
2844 {
2845         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2846 }
2847
2848 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2849                       umode_t mode, struct dt_allocation_hint *hint,
2850                       struct thandle *th)
2851 {
2852         int result;
2853         struct osd_device  *osd = osd_obj2dev(obj);
2854         struct osd_thandle *oth;
2855         struct dt_object   *parent = NULL;
2856         struct inode       *inode;
2857
2858         LINVRNT(osd_invariant(obj));
2859         LASSERT(obj->oo_inode == NULL);
2860         LASSERT(obj->oo_hl_head == NULL);
2861
2862         if (S_ISDIR(mode) && ldiskfs_pdo) {
2863                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2864                 if (obj->oo_hl_head == NULL)
2865                         return -ENOMEM;
2866         }
2867
2868         oth = container_of(th, struct osd_thandle, ot_super);
2869         LASSERT(oth->ot_handle->h_transaction != NULL);
2870
2871         if (hint != NULL && hint->dah_parent != NULL &&
2872             !dt_object_remote(hint->dah_parent))
2873                 parent = hint->dah_parent;
2874
2875         inode = ldiskfs_create_inode(oth->ot_handle,
2876                                      parent ? osd_dt_obj(parent)->oo_inode :
2877                                               osd_sb(osd)->s_root->d_inode,
2878                                      mode);
2879         if (!IS_ERR(inode)) {
2880                 /* Do not update file c/mtime in ldiskfs. */
2881                 inode->i_flags |= S_NOCMTIME;
2882
2883                 /* For new created object, it must be consistent,
2884                  * and it is unnecessary to scrub against it. */
2885                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2886
2887                 obj->oo_inode = inode;
2888                 result = 0;
2889         } else {
2890                 if (obj->oo_hl_head != NULL) {
2891                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2892                         obj->oo_hl_head = NULL;
2893                 }
2894                 result = PTR_ERR(inode);
2895         }
2896         LINVRNT(osd_invariant(obj));
2897         return result;
2898 }
2899
2900 enum {
2901         OSD_NAME_LEN = 255
2902 };
2903
2904 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2905                      struct lu_attr *attr,
2906                      struct dt_allocation_hint *hint,
2907                      struct dt_object_format *dof,
2908                      struct thandle *th)
2909 {
2910         int result;
2911         struct osd_thandle *oth;
2912         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2913
2914         LASSERT(S_ISDIR(attr->la_mode));
2915
2916         oth = container_of(th, struct osd_thandle, ot_super);
2917         LASSERT(oth->ot_handle->h_transaction != NULL);
2918         result = osd_mkfile(info, obj, mode, hint, th);
2919
2920         return result;
2921 }
2922
2923 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2924                         struct lu_attr *attr,
2925                         struct dt_allocation_hint *hint,
2926                         struct dt_object_format *dof,
2927                         struct thandle *th)
2928 {
2929         int result;
2930         struct osd_thandle *oth;
2931         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2932
2933         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2934
2935         LASSERT(S_ISREG(attr->la_mode));
2936
2937         oth = container_of(th, struct osd_thandle, ot_super);
2938         LASSERT(oth->ot_handle->h_transaction != NULL);
2939
2940         result = osd_mkfile(info, obj, mode, hint, th);
2941         if (result == 0) {
2942                 LASSERT(obj->oo_inode != NULL);
2943                 if (feat->dif_flags & DT_IND_VARKEY)
2944                         result = iam_lvar_create(obj->oo_inode,
2945                                                  feat->dif_keysize_max,
2946                                                  feat->dif_ptrsize,
2947                                                  feat->dif_recsize_max,
2948                                                  oth->ot_handle);
2949                 else
2950                         result = iam_lfix_create(obj->oo_inode,
2951                                                  feat->dif_keysize_max,
2952                                                  feat->dif_ptrsize,
2953                                                  feat->dif_recsize_max,
2954                                                  oth->ot_handle);
2955
2956         }
2957         return result;
2958 }
2959
2960 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2961                      struct lu_attr *attr,
2962                      struct dt_allocation_hint *hint,
2963                      struct dt_object_format *dof,
2964                      struct thandle *th)
2965 {
2966         LASSERT(S_ISREG(attr->la_mode));
2967         return osd_mkfile(info, obj, (attr->la_mode &
2968                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2969 }
2970
2971 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2972                      struct lu_attr *attr,
2973                      struct dt_allocation_hint *hint,
2974                      struct dt_object_format *dof,
2975                      struct thandle *th)
2976 {
2977         LASSERT(S_ISLNK(attr->la_mode));
2978         return osd_mkfile(info, obj, (attr->la_mode &
2979                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2980 }
2981
2982 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2983                      struct lu_attr *attr,
2984                      struct dt_allocation_hint *hint,
2985                      struct dt_object_format *dof,
2986                      struct thandle *th)
2987 {
2988         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2989         int result;
2990
2991         LINVRNT(osd_invariant(obj));
2992         LASSERT(obj->oo_inode == NULL);
2993         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2994                 S_ISFIFO(mode) || S_ISSOCK(mode));
2995
2996         result = osd_mkfile(info, obj, mode, hint, th);
2997         if (result == 0) {
2998                 LASSERT(obj->oo_inode != NULL);
2999                 /*
3000                  * This inode should be marked dirty for i_rdev.  Currently
3001                  * that is done in the osd_attr_init().
3002                  */
3003                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3004                                    attr->la_rdev);
3005         }
3006         LINVRNT(osd_invariant(obj));
3007         return result;
3008 }
3009
3010 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3011                               struct lu_attr *,
3012                               struct dt_allocation_hint *hint,
3013                               struct dt_object_format *dof,
3014                               struct thandle *);
3015
3016 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3017 {
3018         osd_obj_type_f result;
3019
3020         switch (type) {
3021         case DFT_DIR:
3022                 result = osd_mkdir;
3023                 break;
3024         case DFT_REGULAR:
3025                 result = osd_mkreg;
3026                 break;
3027         case DFT_SYM:
3028                 result = osd_mksym;
3029                 break;
3030         case DFT_NODE:
3031                 result = osd_mknod;
3032                 break;
3033         case DFT_INDEX:
3034                 result = osd_mk_index;
3035                 break;
3036
3037         default:
3038                 LBUG();
3039                 break;
3040         }
3041         return result;
3042 }
3043
3044
3045 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3046                         struct dt_object *parent, struct dt_object *child,
3047                         umode_t child_mode)
3048 {
3049         LASSERT(ah);
3050
3051         ah->dah_parent = parent;
3052         ah->dah_mode = child_mode;
3053
3054         if (parent != NULL && !dt_object_remote(parent)) {
3055                 /* will help to find FID->ino at dt_insert("..") */
3056                 struct osd_object *pobj = osd_dt_obj(parent);
3057                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3058         }
3059 }
3060
3061 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3062                           struct lu_attr *attr, struct dt_object_format *dof)
3063 {
3064         struct inode   *inode = obj->oo_inode;
3065         __u64           valid = attr->la_valid;
3066         int             result;
3067
3068         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3069
3070         if (dof->dof_type != DFT_NODE)
3071                 attr->la_valid &= ~LA_RDEV;
3072         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3073                 attr->la_valid &= ~LA_ATIME;
3074         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3075                 attr->la_valid &= ~LA_CTIME;
3076         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3077                 attr->la_valid &= ~LA_MTIME;
3078
3079         result = osd_quota_transfer(inode, attr);
3080         if (result)
3081                 return;
3082
3083         if (attr->la_valid != 0) {
3084                 result = osd_inode_setattr(info->oti_env, inode, attr);
3085                 /*
3086                  * The osd_inode_setattr() should always succeed here.  The
3087                  * only error that could be returned is EDQUOT when we are
3088                  * trying to change the UID or GID of the inode. However, this
3089                  * should not happen since quota enforcement is no longer
3090                  * enabled on ldiskfs (lquota takes care of it).
3091                  */
3092                 LASSERTF(result == 0, "%d\n", result);
3093                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3094         }
3095
3096         attr->la_valid = valid;
3097 }
3098
3099 /**
3100  * Helper function for osd_create()
3101  *
3102  * \retval 0, on success
3103  */
3104 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3105                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3106                         struct dt_object_format *dof, struct thandle *th)
3107 {
3108         int     result;
3109         __u32   umask;
3110
3111         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3112
3113         /* we drop umask so that permissions we pass are not affected */
3114         umask = current->fs->umask;
3115         current->fs->umask = 0;
3116
3117         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3118                                                   th);
3119         if (likely(obj->oo_inode != NULL)) {
3120                 LASSERT(obj->oo_inode->i_state & I_NEW);
3121
3122                 /* Unlock the inode before attr initialization to avoid
3123                  * unnecessary dqget operations. LU-6378 */
3124                 unlock_new_inode(obj->oo_inode);
3125         }
3126
3127         if (likely(result == 0)) {
3128                 osd_attr_init(info, obj, attr, dof);
3129                 osd_object_init0(obj);
3130         }
3131
3132         /* restore previous umask value */
3133         current->fs->umask = umask;
3134
3135         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3136
3137         return result;
3138 }
3139
3140 /**
3141  * Helper function for osd_create()
3142  *
3143  * \retval 0, on success
3144  */
3145 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3146                            const struct lu_fid *fid, struct thandle *th)
3147 {
3148         struct osd_thread_info *info = osd_oti_get(env);
3149         struct osd_inode_id    *id   = &info->oti_id;
3150         struct osd_device      *osd  = osd_obj2dev(obj);
3151         struct osd_thandle     *oh;
3152         int                     rc;
3153
3154         LASSERT(obj->oo_inode != NULL);
3155
3156         oh = container_of0(th, struct osd_thandle, ot_super);
3157         LASSERT(oh->ot_handle);
3158         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3159
3160         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3161         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3162                            OI_CHECK_FLD, NULL);
3163         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3164
3165         return rc;
3166 }
3167
3168 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3169                    u64 seq, struct lu_seq_range *range)
3170 {
3171         struct seq_server_site  *ss = osd_seq_site(osd);
3172
3173         if (fid_seq_is_idif(seq)) {
3174                 fld_range_set_ost(range);
3175                 range->lsr_index = idif_ost_idx(seq);
3176                 return 0;
3177         }
3178
3179         if (!fid_seq_in_fldb(seq)) {
3180                 fld_range_set_mdt(range);
3181                 if (ss != NULL)
3182                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3183                          * at all */
3184                         range->lsr_index = ss->ss_node_id;
3185                 return 0;
3186         }
3187
3188         LASSERT(ss != NULL);
3189         fld_range_set_any(range);
3190         /* OSD will only do local fld lookup */
3191         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3192 }
3193
3194 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3195                               struct lu_attr *attr,
3196                               struct dt_allocation_hint *hint,
3197                               struct dt_object_format *dof,
3198                               struct thandle *handle)
3199 {
3200         struct osd_thandle *oh;
3201         int rc;
3202         ENTRY;
3203
3204         LASSERT(handle != NULL);
3205
3206         oh = container_of0(handle, struct osd_thandle, ot_super);
3207         LASSERT(oh->ot_handle == NULL);
3208
3209         /* EA object consumes more credits than regular object: osd_mk_index
3210          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3211          * leaf_node, could involves the block, block bitmap, groups, GDT
3212          * change for each block, so add 4 * 2 credits in that case. */
3213         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3214                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3215                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3216         /* Reuse idle OI block may cause additional one OI block
3217          * to be changed. */
3218         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3219                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3220
3221         if (!attr)
3222                 RETURN(0);
3223
3224         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3225                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3226                                    NULL, OSD_QID_INODE);
3227         if (rc != 0)
3228                 RETURN(rc);
3229
3230         /* will help to find FID->ino mapping at dt_insert() */
3231         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3232                                    osd_dt_obj(dt));
3233
3234         RETURN(rc);
3235 }
3236
3237 /**
3238  * Called to destroy on-disk representation of the object
3239  *
3240  * Concurrency: must be locked
3241  */
3242 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3243                                struct thandle *th)
3244 {
3245         struct osd_object  *obj = osd_dt_obj(dt);
3246         struct inode       *inode = obj->oo_inode;
3247         struct osd_thandle *oh;
3248         int                 rc;
3249         ENTRY;
3250
3251         if (inode == NULL)
3252                 RETURN(-ENOENT);
3253
3254         oh = container_of0(th, struct osd_thandle, ot_super);
3255         LASSERT(oh->ot_handle == NULL);
3256
3257         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3258                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3259
3260         /* For removing agent entry */
3261         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3262                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3263
3264         /* Recycle idle OI leaf may cause additional three OI blocks
3265          * to be changed. */
3266         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3267                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3268                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3269         /* one less inode */
3270         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3271                                    i_projid_read(inode), -1, oh, obj, NULL,
3272                                    OSD_QID_INODE);
3273         if (rc)
3274                 RETURN(rc);
3275         /* data to be truncated */
3276         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3277                                    i_projid_read(inode), 0, oh, obj, NULL,
3278                                    OSD_QID_BLK);
3279         if (rc)
3280                 RETURN(rc);
3281
3282         /* will help to find FID->ino when this object is being
3283          * added to PENDING/ */
3284         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3285
3286         RETURN(rc);
3287 }
3288
3289 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3290                        struct thandle *th)
3291 {
3292         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3293         struct osd_object *obj = osd_dt_obj(dt);
3294         struct inode *inode = obj->oo_inode;
3295         struct osd_device *osd = osd_obj2dev(obj);
3296         struct osd_thandle *oh;
3297         int result;
3298         ENTRY;
3299
3300         oh = container_of0(th, struct osd_thandle, ot_super);
3301         LASSERT(oh->ot_handle);
3302         LASSERT(inode);
3303         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3304
3305         if (unlikely(fid_is_acct(fid)))
3306                 RETURN(-EPERM);
3307
3308         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3309                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3310                 if (result != 0)
3311                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3312                                osd_name(osd), PFID(fid), result);
3313         }
3314
3315         if (S_ISDIR(inode->i_mode)) {
3316                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3317                         inode->i_nlink == 2);
3318
3319                 spin_lock(&obj->oo_guard);
3320                 clear_nlink(inode);
3321                 spin_unlock(&obj->oo_guard);
3322                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3323         }
3324
3325         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3326
3327         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3328
3329         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3330                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3331                                        oh->ot_handle, OI_CHECK_FLD);
3332
3333         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3334         /* XXX: add to ext3 orphan list */
3335         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3336
3337         /* not needed in the cache anymore */
3338         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3339         obj->oo_destroyed = 1;
3340
3341         RETURN(0);
3342 }
3343
3344 /**
3345  * Put the fid into lustre_mdt_attrs, and then place the structure
3346  * inode's ea. This fid should not be altered during the life time
3347  * of the inode.
3348  *
3349  * \retval +ve, on success
3350  * \retval -ve, on error
3351  *
3352  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3353  */
3354 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3355                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3356 {
3357         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3358         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3359         int rc;
3360         ENTRY;
3361
3362         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3363                 RETURN(0);
3364
3365         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3366                 rc = -ENOMEM;
3367
3368         lustre_loa_init(loa, fid, compat, incompat);
3369         lustre_loa_swab(loa, false);
3370
3371         /* For the OST device with 256 bytes inode size by default,
3372          * the PFID EA will be stored together with LMA EA to avoid
3373          * performance trouble. Otherwise the PFID EA can be stored
3374          * independently. LU-8998 */
3375         if ((compat & LMAC_FID_ON_OST) &&
3376             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3377                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3378                                      sizeof(*loa), XATTR_CREATE);
3379         else
3380                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3381                                      sizeof(*lma), XATTR_CREATE);
3382         /* LMA may already exist, but we need to check that all the
3383          * desired compat/incompat flags have been added. */
3384         if (unlikely(rc == -EEXIST)) {
3385                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3386                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3387                 if (rc < 0)
3388                         RETURN(rc);
3389
3390                 if (rc < sizeof(*lma))
3391                         RETURN(-EINVAL);
3392
3393                 lustre_loa_swab(loa, true);
3394                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3395                     ((compat == 0 && incompat == 0) ||
3396                      (!(~lma->lma_compat & compat) &&
3397                       !(~lma->lma_incompat & incompat))))
3398                         RETURN(0);
3399
3400                 lma->lma_self_fid = *fid;
3401                 lma->lma_compat |= compat;
3402                 lma->lma_incompat |= incompat;
3403                 if (rc == sizeof(*lma)) {
3404                         lustre_lma_swab(lma);
3405                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3406                                              sizeof(*lma), XATTR_REPLACE);
3407                 } else {
3408                         lustre_loa_swab(loa, false);
3409                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3410                                              sizeof(*loa), XATTR_REPLACE);
3411                 }
3412         }
3413
3414         RETURN(rc);
3415 }
3416
3417 /**
3418  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3419  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3420  * To have compatilibility with 1.8 ldiskfs driver we need to have
3421  * magic number at start of fid data.
3422  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3423  * its inmemory API.
3424  */
3425 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3426                                          const struct lu_fid *fid)
3427 {
3428         if (!fid_is_namespace_visible(fid) ||
3429             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3430                 param->edp_magic = 0;
3431                 return;
3432         }
3433
3434         param->edp_magic = LDISKFS_LUFID_MAGIC;
3435         param->edp_len =  sizeof(struct lu_fid) + 1;
3436         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3437 }
3438
3439 /**
3440  * Try to read the fid from inode ea into dt_rec.
3441  *
3442  * \param fid object fid.
3443  *
3444  * \retval 0 on success
3445  */
3446 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3447                           __u32 ino, struct lu_fid *fid,
3448                           struct osd_inode_id *id)
3449 {
3450         struct osd_thread_info *info  = osd_oti_get(env);
3451         struct inode           *inode;
3452         ENTRY;
3453
3454         osd_id_gen(id, ino, OSD_OII_NOGEN);
3455         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3456         if (IS_ERR(inode))
3457                 RETURN(PTR_ERR(inode));
3458
3459         iput(inode);
3460         RETURN(0);
3461 }
3462
3463 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3464                                         struct inode *dir,
3465                                         struct inode *parent_dir,
3466                                         const struct lu_fid *dot_fid,
3467                                         const struct lu_fid *dot_dot_fid,
3468                                         struct osd_thandle *oth)
3469 {
3470         struct ldiskfs_dentry_param *dot_ldp;
3471         struct ldiskfs_dentry_param *dot_dot_ldp;
3472         __u32 saved_nlink = dir->i_nlink;
3473         int rc;
3474
3475         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3476         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3477
3478         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3479         dot_ldp->edp_magic = 0;
3480
3481         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3482                                     dir, dot_ldp, dot_dot_ldp);
3483         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3484          * the subseqent ref_add() will increase the dir->i_nlink
3485          * as 3. That is incorrect for new created directory.
3486          *
3487          * It looks like hack, because we want to make the OSD API
3488          * to be order-independent for new created directory object
3489          * between dt_insert(..) and ref_add() operations.
3490          *
3491          * Here, we only restore the in-RAM dir-inode's nlink attr,
3492          * becuase if the nlink attr is not 2, then there will be
3493          * ref_add() called following the dt_insert(..), such call
3494          * will make both the in-RAM and on-disk dir-inode's nlink
3495          * attr to be set as 2. LU-7447 */
3496         set_nlink(dir, saved_nlink);
3497         return rc;
3498 }
3499
3500 /**
3501  * Create an local agent inode for remote entry
3502  */
3503 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3504                                                   struct osd_device *osd,
3505                                                   struct osd_object *pobj,
3506                                                   const struct lu_fid *fid,
3507                                                   __u32 type,
3508                                                   struct thandle *th)
3509 {
3510         struct osd_thread_info  *info = osd_oti_get(env);
3511         struct inode            *local;
3512         struct osd_thandle      *oh;
3513         int                     rc;
3514         ENTRY;
3515
3516         LASSERT(th);
3517         oh = container_of(th, struct osd_thandle, ot_super);
3518         LASSERT(oh->ot_handle->h_transaction != NULL);
3519
3520         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3521         if (IS_ERR(local)) {
3522                 CERROR("%s: create local error %d\n", osd_name(osd),
3523                        (int)PTR_ERR(local));
3524                 RETURN(local);
3525         }
3526
3527         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3528          * correct gid on remote file, not agent here */
3529         local->i_gid = current_fsgid();
3530         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3531         unlock_new_inode(local);
3532
3533         /* Agent inode should not have project ID */
3534 #ifdef  HAVE_PROJECT_QUOTA
3535         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3536             i_projid_read(pobj->oo_inode) != 0) {
3537                 rc = ldiskfs_transfer_project(local, 0);
3538                 if (rc) {
3539                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3540                                "quota enforcement enabled on the ldiskfs "
3541                                "filesystem?\n", local->i_sb->s_id, rc);
3542                         RETURN(ERR_PTR(rc));
3543                 }
3544         }
3545 #endif
3546         /* Set special LMA flag for local agent inode */
3547         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3548         if (rc != 0) {
3549                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3550                        osd_name(osd), PFID(fid), rc);
3551                 RETURN(ERR_PTR(rc));
3552         }
3553
3554         if (!S_ISDIR(type))
3555                 RETURN(local);
3556
3557         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3558                                          lu_object_fid(&pobj->oo_dt.do_lu),
3559                                          fid, oh);
3560         if (rc != 0) {
3561                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3562                         osd_name(osd), PFID(fid), rc);
3563                 RETURN(ERR_PTR(rc));
3564         }
3565
3566         RETURN(local);
3567 }
3568
3569 /**
3570  * when direntry is deleted, we have to take care of possible agent inode
3571  * referenced by that. unfortunately we can't do this at that point:
3572  * iget() within a running transaction leads to deadlock and we better do
3573  * not call that every delete declaration to save performance. so we put
3574  * a potention agent inode on a list and process that once the transaction
3575  * is over. Notice it's not any worse in terms of real orphans as regular
3576  * object destroy doesn't put inodes on the on-disk orphan list. this should
3577  * be addressed separately
3578  */
3579 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3580                                             struct osd_thandle *oh,
3581                                             __u32 ino)
3582 {
3583         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3584         struct osd_obj_orphan *oor;
3585
3586         OBD_ALLOC_PTR(oor);
3587         if (oor == NULL)
3588                 return -ENOMEM;
3589
3590         oor->oor_ino = ino;
3591         oor->oor_env = (struct lu_env *)env;
3592         spin_lock(&osd->od_osfs_lock);
3593         list_add(&oor->oor_list, &osd->od_orphan_list);
3594         spin_unlock(&osd->od_osfs_lock);
3595
3596         oh->ot_remove_agents = 1;
3597
3598         return 0;
3599
3600 }
3601
3602 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3603                                                 struct osd_device *osd)
3604 {
3605         struct osd_thread_info *info = osd_oti_get(env);
3606         struct osd_obj_orphan *oor, *tmp;
3607         struct osd_inode_id id;
3608         struct list_head list;
3609         struct inode *inode;
3610         struct lu_fid fid;
3611         handle_t *jh;
3612         __u32 ino;
3613
3614         INIT_LIST_HEAD(&list);
3615
3616         spin_lock(&osd->od_osfs_lock);
3617         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3618                 if (oor->oor_env == env) {
3619                         list_del(&oor->oor_list);
3620                         list_add(&oor->oor_list, &list);
3621                 }
3622         }
3623         spin_unlock(&osd->od_osfs_lock);
3624
3625         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3626
3627                 ino = oor->oor_ino;
3628
3629                 list_del(&oor->oor_list);
3630                 OBD_FREE_PTR(oor);
3631
3632                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3633                 inode = osd_iget_fid(info, osd, &id, &fid);
3634                 if (IS_ERR(inode))
3635                         continue;
3636
3637                 if (!osd_remote_fid(env, osd, &fid)) {
3638                         iput(inode);
3639                         continue;
3640                 }
3641
3642                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3643                 clear_nlink(inode);
3644                 mark_inode_dirty(inode);
3645                 ldiskfs_journal_stop(jh);
3646                 iput(inode);
3647         }
3648
3649         return 0;
3650 }
3651
3652 /**
3653  * OSD layer object create function for OST objects (b=11826).
3654  *
3655  * The FID is inserted into inode xattr here.
3656  *
3657  * \retval   0, on success
3658  * \retval -ve, on error
3659  */
3660 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3661                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3662                       struct dt_object_format *dof, struct thandle *th)
3663 {
3664         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3665         struct osd_object       *obj    = osd_dt_obj(dt);
3666         struct osd_thread_info  *info   = osd_oti_get(env);
3667         int                      result, on_ost = 0;
3668
3669         ENTRY;
3670
3671         if (dt_object_exists(dt))
3672                 RETURN(-EEXIST);
3673
3674         LINVRNT(osd_invariant(obj));
3675         LASSERT(!dt_object_remote(dt));
3676         LASSERT(osd_is_write_locked(env, obj));
3677         LASSERT(th != NULL);
3678
3679         if (unlikely(fid_is_acct(fid)))
3680                 /* Quota files can't be created from the kernel any more,
3681                  * 'tune2fs -O quota' will take care of creating them */
3682                 RETURN(-EPERM);
3683
3684         result = __osd_create(info, obj, attr, hint, dof, th);
3685         if (result == 0) {
3686                 if (fid_is_idif(fid) &&
3687                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3688                         struct lu_fid *tfid = &info->oti_fid;
3689                         struct ost_id *oi   = &info->oti_ostid;
3690
3691                         fid_to_ostid(fid, oi);
3692                         ostid_to_fid(tfid, oi, 0);
3693                         on_ost = 1;
3694                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3695                                                 LMAC_FID_ON_OST, 0);
3696                 } else {
3697                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3698                                                fid, OI_CHECK_FLD);
3699                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3700                                                 on_ost ? LMAC_FID_ON_OST : 0,
3701                                                 0);
3702                 }
3703                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3704                         obj->oo_dt.do_body_ops = &osd_body_ops;
3705         }
3706
3707         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3708                 result = __osd_oi_insert(env, obj, fid, th);
3709
3710         /* a small optimization - dt_insert() isn't usually applied
3711          * to OST objects, so we don't need to cache OI mapping for
3712          * OST objects */
3713         if (result == 0 && on_ost == 0) {
3714                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3715                 result = osd_idc_find_and_init(env, osd, obj);
3716                 LASSERT(result == 0);
3717         }
3718
3719         LASSERT(ergo(result == 0,
3720                      dt_object_exists(dt) && !dt_object_remote(dt)));
3721         LINVRNT(osd_invariant(obj));
3722         RETURN(result);
3723 }
3724
3725 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3726                                struct thandle *handle)
3727 {
3728         struct osd_thandle *oh;
3729
3730         /* it's possible that object doesn't exist yet */
3731         LASSERT(handle != NULL);
3732
3733         oh = container_of0(handle, struct osd_thandle, ot_super);
3734         LASSERT(oh->ot_handle == NULL);
3735
3736         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3737                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3738
3739         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3740
3741         return 0;
3742 }
3743
3744 /*
3745  * Concurrency: @dt is write locked.
3746  */
3747 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3748                        struct thandle *th)
3749 {
3750         struct osd_object  *obj = osd_dt_obj(dt);
3751         struct inode       *inode = obj->oo_inode;
3752         struct osd_thandle *oh;
3753         int                 rc = 0;
3754
3755         if (!dt_object_exists(dt) || obj->oo_destroyed)
3756                 return -ENOENT;
3757
3758         LINVRNT(osd_invariant(obj));
3759         LASSERT(!dt_object_remote(dt));
3760         LASSERT(osd_is_write_locked(env, obj));
3761         LASSERT(th != NULL);
3762
3763         oh = container_of0(th, struct osd_thandle, ot_super);
3764         LASSERT(oh->ot_handle != NULL);
3765
3766         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3767
3768         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3769                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3770         /*
3771          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3772          * (65000) subdirectories by storing "1" in i_nlink if the link count
3773          * would otherwise overflow. Directory tranversal tools understand
3774          * that (st_nlink == 1) indicates that the filesystem dose not track
3775          * hard links count on the directory, and will not abort subdirectory
3776          * scanning early once (st_nlink - 2) subdirs have been found.
3777          *
3778          * This also has to properly handle the case of inodes with nlink == 0
3779          * in case they are being linked into the PENDING directory
3780          */
3781         spin_lock(&obj->oo_guard);
3782         if (unlikely(inode->i_nlink == 0))
3783                 /* inc_nlink from 0 may cause WARN_ON */
3784                 set_nlink(inode, 1);
3785         else {
3786                 ldiskfs_inc_count(oh->ot_handle, inode);
3787                 if (!S_ISDIR(inode->i_mode))
3788                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3789         }
3790         spin_unlock(&obj->oo_guard);
3791
3792         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3793         LINVRNT(osd_invariant(obj));
3794
3795         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3796
3797         return rc;
3798 }
3799
3800 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3801                                struct thandle *handle)
3802 {
3803         struct osd_thandle *oh;
3804
3805         if (!dt_object_exists(dt))
3806                 return -ENOENT;
3807
3808         LASSERT(!dt_object_remote(dt));
3809         LASSERT(handle != NULL);
3810
3811         oh = container_of0(handle, struct osd_thandle, ot_super);
3812         LASSERT(oh->ot_handle == NULL);
3813
3814         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3815                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3816
3817         return 0;
3818 }
3819
3820 /*
3821  * Concurrency: @dt is write locked.
3822  */
3823 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3824                        struct thandle *th)
3825 {
3826         struct osd_object       *obj = osd_dt_obj(dt);
3827         struct inode            *inode = obj->oo_inode;
3828         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3829         struct osd_thandle      *oh;
3830
3831         if (!dt_object_exists(dt))
3832                 return -ENOENT;
3833
3834         LINVRNT(osd_invariant(obj));
3835         LASSERT(!dt_object_remote(dt));
3836         LASSERT(osd_is_write_locked(env, obj));
3837         LASSERT(th != NULL);
3838
3839         oh = container_of0(th, struct osd_thandle, ot_super);
3840         LASSERT(oh->ot_handle != NULL);
3841
3842         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3843
3844         spin_lock(&obj->oo_guard);
3845         /* That can be result of upgrade from old Lustre version and
3846          * applied only to local files.  Just skip this ref_del call.
3847          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3848         if (inode->i_nlink == 0) {
3849                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3850                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3851                              ", maybe an upgraded file? (LU-3915)\n",
3852                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3853                 spin_unlock(&obj->oo_guard);
3854                 return 0;
3855         }
3856
3857         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3858                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3859
3860         ldiskfs_dec_count(oh->ot_handle, inode);
3861         spin_unlock(&obj->oo_guard);
3862
3863         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3864         LINVRNT(osd_invariant(obj));
3865
3866         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3867
3868         return 0;
3869 }
3870
3871 /*
3872  * Concurrency: @dt is read locked.
3873  */
3874 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3875                          struct lu_buf *buf, const char *name)
3876 {
3877         struct osd_object      *obj    = osd_dt_obj(dt);
3878         struct inode           *inode  = obj->oo_inode;
3879         struct osd_thread_info *info   = osd_oti_get(env);
3880         struct dentry          *dentry = &info->oti_obj_dentry;
3881         bool                    cache_xattr = false;
3882         int                     rc;
3883
3884         LASSERT(buf);
3885
3886         /* version get is not real XATTR but uses xattr API */
3887         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3888                 dt_obj_version_t *ver = buf->lb_buf;
3889
3890                 /* for version we are just using xattr API but change inode
3891                  * field instead */
3892                 if (buf->lb_len == 0)
3893                         return sizeof(dt_obj_version_t);
3894
3895                 if (buf->lb_len < sizeof(dt_obj_version_t))
3896                         return -ERANGE;
3897
3898                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3899                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3900
3901                 *ver = LDISKFS_I(inode)->i_fs_version;
3902
3903                 return sizeof(dt_obj_version_t);
3904         }
3905
3906         if (!dt_object_exists(dt))
3907                 return -ENOENT;
3908
3909         LASSERT(!dt_object_remote(dt));
3910         LASSERT(inode->i_op != NULL);
3911 #ifdef HAVE_IOP_XATTR
3912         LASSERT(inode->i_op->getxattr != NULL);
3913 #endif
3914
3915         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3916             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3917                 cache_xattr = true;
3918
3919         if (cache_xattr) {
3920                 rc = osd_oxc_get(obj, name, buf);
3921                 if (rc != -ENOENT)
3922                         return rc;
3923         }
3924
3925         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
3926                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3927                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3928                 struct filter_fid *ff;
3929                 struct ost_layout *ol;
3930
3931                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3932                 if (rc)
3933                         return rc;
3934
3935                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
3936
3937                 rc = sizeof(*ff);
3938                 if (buf->lb_len == 0 || !buf->lb_buf)
3939                         return rc;
3940
3941                 if (buf->lb_len < rc)
3942                         return -ERANGE;
3943
3944                 ff = buf->lb_buf;
3945                 ol = &ff->ff_layout;
3946                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3947                                                   PFID_STRIPE_IDX_BITS);
3948                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3949                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3950                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3951                 if (lma->lma_compat & LMAC_COMP_INFO) {
3952                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3953                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3954                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3955                 } else {
3956                         ol->ol_comp_start = 0;
3957                         ol->ol_comp_end = 0;
3958                         ol->ol_comp_id = 0;
3959                 }
3960         } else {
3961                 rc = __osd_xattr_get(inode, dentry, name,
3962                                      buf->lb_buf, buf->lb_len);
3963         }
3964
3965         if (cache_xattr) {
3966                 if (rc == -ENOENT || rc == -ENODATA)
3967                         osd_oxc_add(obj, name, NULL, 0);
3968                 else if (rc > 0 && buf->lb_buf != NULL)
3969                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3970         }
3971
3972         return rc;
3973 }
3974
3975 static int osd_declare_xattr_set(const struct lu_env *env,
3976                                  struct dt_object *dt,
3977                                  const struct lu_buf *buf, const char *name,
3978                                  int fl, struct thandle *handle)
3979 {
3980         struct osd_thandle *oh;
3981         int credits = 0;
3982         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3983
3984         LASSERT(handle != NULL);
3985
3986         oh = container_of0(handle, struct osd_thandle, ot_super);
3987         LASSERT(oh->ot_handle == NULL);
3988
3989         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3990                 /* For non-upgrading case, the LMA is set first and
3991                  * usually fit inode. But for upgrade case, the LMA
3992                  * may be in another separated EA block. */
3993                 if (dt_object_exists(dt)) {
3994                         if (fl == LU_XATTR_REPLACE)
3995                                 credits = 1;
3996                         else
3997                                 goto upgrade;
3998                 }
3999         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4000                 credits = 1;
4001         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4002                 /* We may need to delete the old PFID EA. */
4003                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4004                 if (fl == LU_XATTR_REPLACE)
4005                         credits += 1;
4006                 else
4007                         goto upgrade;
4008         } else {
4009                 /* If some name entry resides on remote MDT, then will create
4010                  * agent entry under remote parent. On the other hand, if the
4011                  * remote entry will be removed, then related agent entry may
4012                  * need to be removed from the remote parent. So there may be
4013                  * kinds of cases, let's declare enough credits. The credits
4014                  * for create agent entry is enough for remove case. */
4015                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4016                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4017                         if (dt_object_exists(dt))
4018                                 credits += 1; /* For updating LMA */
4019                 }
4020
4021 upgrade:
4022                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4023
4024                 if (buf != NULL) {
4025                         ssize_t buflen;
4026
4027                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4028                                 /* learn xattr size from osd_xattr_get if
4029                                    attribute has not been read yet */
4030                                 buflen = __osd_xattr_get(
4031                                     osd_dt_obj(dt)->oo_inode,
4032                                     &osd_oti_get(env)->oti_obj_dentry,
4033                                     name, NULL, 0);
4034                                 if (buflen < 0)
4035                                         buflen = 0;
4036                         } else {
4037                                 buflen = buf->lb_len;
4038                         }
4039
4040                         if (buflen > sb->s_blocksize) {
4041                                 credits += osd_calc_bkmap_credits(
4042                                     sb, NULL, 0, -1,
4043                                     (buflen + sb->s_blocksize - 1) >>
4044                                     sb->s_blocksize_bits);
4045                         }
4046                 }
4047                 /*
4048                  * xattr set may involve inode quota change, reserve credits for
4049                  * dquot_initialize()
4050                  */
4051                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4052         }
4053
4054         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4055
4056         return 0;
4057 }
4058
4059 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4060                               const struct lu_buf *buf, int fl,
4061                               struct thandle *handle)
4062 {
4063         struct osd_thread_info *info = osd_oti_get(env);
4064         struct dentry *dentry = &info->oti_obj_dentry;
4065         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4066         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4067         struct inode *inode = obj->oo_inode;
4068         struct filter_fid *ff = buf->lb_buf;
4069         struct ost_layout *ol = &ff->ff_layout;
4070         int flags = XATTR_REPLACE;
4071         int rc;
4072         ENTRY;
4073
4074         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4075                 RETURN(-EINVAL);
4076
4077         rc = osd_get_lma(info, inode, dentry, loa);
4078         if (rc == -ENODATA) {
4079                 /* Usually for upgarding from old device */
4080                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4081                                 LMAC_FID_ON_OST, 0);
4082                 flags = XATTR_CREATE;
4083         } else if (rc) {
4084                 RETURN(rc);
4085         }
4086
4087         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4088                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4089                         RETURN(-EEXIST);
4090
4091                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4092                         /* Separate PFID EA from LMA */
4093                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4094                         lustre_lma_swab(lma);
4095                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4096                                              sizeof(*lma), XATTR_REPLACE);
4097                         if (!rc) {
4098                                 obj->oo_pfid_in_lma = 0;
4099                                 rc = LU_XATTR_CREATE;
4100                         }
4101
4102                         RETURN(rc);
4103                 }
4104         } else {
4105                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4106                         RETURN(fl);
4107
4108                 /* Old client does not send stripe information,
4109                  * then store the PFID EA on disk separatedly. */
4110                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4111                              ol->ol_stripe_size == 0))
4112                         RETURN(fl);
4113
4114                 /* Remove old PFID EA entry firstly. */
4115                 ll_vfs_dq_init(inode);
4116                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4117                 if (rc == -ENODATA) {
4118                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4119                                 RETURN(rc);
4120                 } else if (rc) {
4121                         RETURN(rc);
4122                 }
4123         }
4124
4125         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4126         if (likely(ol->ol_stripe_size != 0)) {
4127                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4128                                              PFID_STRIPE_IDX_BITS;
4129                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4130                 lma->lma_compat |= LMAC_STRIPE_INFO;
4131                 if (ol->ol_comp_id != 0) {
4132                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4133                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4134                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4135                         lma->lma_compat |= LMAC_COMP_INFO;
4136                 }
4137         }
4138
4139         lustre_loa_swab(loa, false);
4140
4141         /* Store the PFID EA inside LMA. */
4142         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4143                              flags);
4144         if (!rc)
4145                 obj->oo_pfid_in_lma = 1;
4146
4147         RETURN(rc);
4148 }
4149
4150 /*
4151  * In DNE environment, the object (in spite of regular file or directory)
4152  * and its name entry may reside on different MDTs. Under such case, we will
4153  * create an agent entry on the MDT where the object resides. The agent entry
4154  * references the object locally, that makes the object to be visible to the
4155  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4156  * such as 'tar' can handle the object properly.
4157  *
4158  * We handle the agent entry during set linkEA that is the common interface
4159  * for both regular file and directroy, can handle kinds of cases, such as
4160  * create/link/unlink/rename, and so on.
4161  *
4162  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4163  *
4164  * XXX: There are two known issues:
4165  * 1. For one object, we will create at most one agent entry even if there
4166  *    may be more than one cross-MDTs hard links on the object. So the local
4167  *    e2fsck may claim that the object's nlink is larger than the name entries
4168  *    that reference such inode. And in further, the e2fsck will fix the nlink
4169  *    attribute to match the local references. Then it will cause the object's
4170  *    nlink attribute to be inconsistent with the global references. it is bad
4171  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4172  *    other hand, the global namespace LFSCK can repair the object's attribute
4173  *    according to the linkEA.
4174  * 2. There may be too many hard links on the object as to its linkEA overflow,
4175  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4176  *    case happened, then at this point, we do not know whether there are some
4177  *    cross-MDTs reference. But there are local references, it guarantees that
4178  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4179  */
4180 static int osd_xattr_handle_linkea(const struct lu_env *env,
4181                                    struct osd_device *osd,
4182                                    struct osd_object *obj,
4183                                    const struct lu_buf *buf,
4184                                    struct thandle *handle)
4185 {
4186         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4187         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4188         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4189         struct lu_name tmpname;
4190         struct osd_thandle *oh;
4191         int rc;
4192         bool remote = false;
4193         ENTRY;
4194
4195         oh = container_of0(handle, struct osd_thandle, ot_super);
4196         LASSERT(oh->ot_handle != NULL);
4197
4198         rc = linkea_init_with_rec(&ldata);
4199         if (!rc) {
4200                 linkea_first_entry(&ldata);
4201                 while (ldata.ld_lee != NULL && !remote) {
4202                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4203                                             &tmpname, tfid);
4204                         if (osd_remote_fid(env, osd, tfid) > 0)
4205                                 remote = true;
4206                         else
4207                                 linkea_next_entry(&ldata);
4208                 }
4209         } else if (rc == -ENODATA) {
4210                 rc = 0;
4211         } else {
4212                 RETURN(rc);
4213         }
4214
4215         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4216                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4217                 if (rc)
4218                         CERROR("%s: failed to remove agent entry for "DFID
4219                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4220         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4221                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4222                 if (rc)
4223                         CERROR("%s: failed to create agent entry for "DFID
4224                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4225         }
4226
4227         RETURN(rc);
4228 }
4229
4230 /*
4231  * Concurrency: @dt is write locked.
4232  */
4233 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4234                          const struct lu_buf *buf, const char *name, int fl,
4235                          struct thandle *handle)
4236 {
4237         struct osd_object *obj = osd_dt_obj(dt);
4238         struct osd_device *osd = osd_obj2dev(obj);
4239         struct inode *inode = obj->oo_inode;
4240         struct osd_thread_info *info = osd_oti_get(env);
4241         int fs_flags = 0;
4242         int len;
4243         int rc;
4244         ENTRY;
4245
4246         LASSERT(handle);
4247         LASSERT(buf);
4248
4249         /* version set is not real XATTR */
4250         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4251                 dt_obj_version_t *version = buf->lb_buf;
4252
4253                 /* for version we are just using xattr API but change inode
4254                  * field instead */
4255                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4256
4257                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4258                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4259
4260                 LDISKFS_I(inode)->i_fs_version = *version;
4261                 /* Version is set after all inode operations are finished,
4262                  * so we should mark it dirty here */
4263                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4264
4265                 RETURN(0);
4266         }
4267
4268         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4269                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4270
4271         len = buf->lb_len;
4272         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4273
4274         /* For the OST device with 256 bytes inode size by default,
4275          * the PFID EA will be stored together with LMA EA to avoid
4276          * performance trouble. Otherwise the PFID EA can be stored
4277          * independently. LU-8998 */
4278         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4279             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4280                 LASSERT(buf->lb_buf);
4281
4282                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4283                 if (fl <= 0)
4284                         RETURN(fl);
4285         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4286                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4287                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4288
4289                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4290                 if (rc)
4291                         RETURN(rc);
4292
4293                 lma->lma_incompat |= LMAI_STRIPED;
4294                 lustre_lma_swab(lma);
4295                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4296                                      sizeof(*lma), XATTR_REPLACE);
4297                 if (rc != 0)
4298                         RETURN(rc);
4299         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4300                 LASSERT(!osd->od_is_ost);
4301
4302                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4303                 if (rc)
4304                         RETURN(rc);
4305         }
4306
4307         if (fl & LU_XATTR_REPLACE)
4308                 fs_flags |= XATTR_REPLACE;
4309
4310         if (fl & LU_XATTR_CREATE)
4311                 fs_flags |= XATTR_CREATE;
4312
4313         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4314         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4315
4316         if (rc == 0 &&
4317             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4318              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4319                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4320
4321         return rc;
4322 }
4323
4324 /*
4325  * Concurrency: @dt is read locked.
4326  */
4327 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4328                           const struct lu_buf *buf)
4329 {
4330         struct osd_object      *obj    = osd_dt_obj(dt);
4331         struct inode           *inode  = obj->oo_inode;
4332         struct osd_thread_info *info   = osd_oti_get(env);
4333         struct dentry          *dentry = &info->oti_obj_dentry;
4334
4335         if (!dt_object_exists(dt))
4336                 return -ENOENT;
4337
4338         LASSERT(!dt_object_remote(dt));
4339         LASSERT(inode->i_op != NULL);
4340         LASSERT(inode->i_op->listxattr != NULL);
4341
4342         dentry->d_inode = inode;
4343         dentry->d_sb = inode->i_sb;
4344         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4345 }
4346
4347 static int osd_declare_xattr_del(const struct lu_env *env,
4348                                  struct dt_object *dt, const char *name,
4349                                  struct thandle *handle)
4350 {
4351         struct osd_thandle *oh;
4352         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4353
4354         LASSERT(!dt_object_remote(dt));
4355         LASSERT(handle != NULL);
4356
4357         oh = container_of0(handle, struct osd_thandle, ot_super);
4358         LASSERT(oh->ot_handle == NULL);
4359
4360         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4361                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4362         /*
4363          * xattr del may involve inode quota change, reserve credits for
4364          * dquot_initialize()
4365          */
4366         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4367
4368         return 0;
4369 }
4370
4371 /*
4372  * Concurrency: @dt is write locked.
4373  */
4374 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4375                          const char *name, struct thandle *handle)
4376 {
4377         struct osd_object      *obj    = osd_dt_obj(dt);
4378         struct inode           *inode  = obj->oo_inode;
4379         struct osd_thread_info *info   = osd_oti_get(env);
4380         struct dentry          *dentry = &info->oti_obj_dentry;
4381         int                     rc;
4382
4383         if (!dt_object_exists(dt))
4384                 return -ENOENT;
4385
4386         LASSERT(!dt_object_remote(dt));
4387         LASSERT(inode->i_op != NULL);
4388         LASSERT(handle != NULL);
4389 #ifdef HAVE_IOP_XATTR
4390         LASSERT(inode->i_op->removexattr != NULL);
4391 #endif
4392
4393         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4394
4395         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4396                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4397
4398                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4399                                  &info->oti_ost_attrs);
4400                 if (!rc) {
4401                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4402
4403                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4404                         lustre_lma_swab(lma);
4405                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4406                                              sizeof(*lma), XATTR_REPLACE);
4407                         if (!rc)
4408                                 obj->oo_pfid_in_lma = 0;
4409                 }
4410         } else {
4411                 ll_vfs_dq_init(inode);
4412                 dentry->d_inode = inode;
4413                 dentry->d_sb = inode->i_sb;
4414                 rc = osd_removexattr(dentry, inode, name);
4415         }
4416
4417         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4418
4419         if (rc == 0 &&
4420             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4421              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4422                 osd_oxc_del(obj, name);
4423
4424         return rc;
4425 }
4426
4427 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4428                            __u64 start, __u64 end)
4429 {
4430         struct osd_object       *obj    = osd_dt_obj(dt);
4431         struct inode            *inode  = obj->oo_inode;
4432         struct osd_thread_info  *info   = osd_oti_get(env);
4433         struct dentry           *dentry = &info->oti_obj_dentry;
4434         struct file             *file   = &info->oti_file;
4435         int                     rc;
4436
4437         ENTRY;
4438
4439         dentry->d_inode = inode;
4440         dentry->d_sb = inode->i_sb;
4441         file->f_path.dentry = dentry;
4442         file->f_mapping = inode->i_mapping;
4443         file->f_op = inode->i_fop;
4444         set_file_inode(file, inode);
4445
4446         rc = ll_vfs_fsync_range(file, start, end, 0);
4447
4448         RETURN(rc);
4449 }
4450
4451 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4452 {
4453         return 0;
4454 }
4455
4456 /*
4457  * Index operations.
4458  */
4459
4460 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4461                            const struct dt_index_features *feat)
4462 {
4463         struct iam_descr *descr;
4464
4465         if (osd_object_is_root(o))
4466                 return feat == &dt_directory_features;
4467
4468         LASSERT(o->oo_dir != NULL);
4469
4470         descr = o->oo_dir->od_container.ic_descr;
4471         if (feat == &dt_directory_features) {
4472                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4473                         return 1;
4474                 else
4475                         return 0;
4476         } else {
4477                 return
4478                         feat->dif_keysize_min <= descr->id_key_size &&
4479                         descr->id_key_size <= feat->dif_keysize_max &&
4480                         feat->dif_recsize_min <= descr->id_rec_size &&
4481                         descr->id_rec_size <= feat->dif_recsize_max &&
4482                         !(feat->dif_flags & (DT_IND_VARKEY |
4483                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4484                         ergo(feat->dif_flags & DT_IND_UPDATE,
4485                              1 /* XXX check that object (and file system) is
4486                                 * writable */);
4487         }
4488 }
4489
4490 static int osd_iam_container_init(const struct lu_env *env,
4491                                   struct osd_object *obj,
4492                                   struct osd_directory *dir)
4493 {
4494         struct iam_container *bag = &dir->od_container;
4495         int result;
4496
4497         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4498         if (result != 0)
4499                 return result;
4500
4501         result = iam_container_setup(bag);
4502         if (result == 0)
4503                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4504         else
4505                 iam_container_fini(bag);
4506
4507         return result;
4508 }
4509
4510
4511 /*
4512  * Concurrency: no external locking is necessary.
4513  */
4514 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4515                          const struct dt_index_features *feat)
4516 {
4517         int                      result;
4518         int                      skip_iam = 0;
4519         struct osd_object       *obj = osd_dt_obj(dt);
4520
4521         LINVRNT(osd_invariant(obj));
4522
4523         if (osd_object_is_root(obj)) {
4524                 dt->do_index_ops = &osd_index_ea_ops;
4525                 result = 0;
4526         } else if (feat == &dt_directory_features) {
4527                 dt->do_index_ops = &osd_index_ea_ops;
4528                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4529                         result = 0;
4530                 else
4531                         result = -ENOTDIR;
4532                 skip_iam = 1;
4533         } else if (unlikely(feat == &dt_otable_features)) {
4534                 dt->do_index_ops = &osd_otable_ops;
4535                 return 0;
4536         } else if (unlikely(feat == &dt_acct_features)) {
4537                 dt->do_index_ops = &osd_acct_index_ops;
4538                 result = 0;
4539                 skip_iam = 1;
4540         } else if (!osd_has_index(obj)) {
4541                 struct osd_directory *dir;
4542                 struct osd_device *osd = osd_obj2dev(obj);
4543                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4544
4545                 OBD_ALLOC_PTR(dir);
4546                 if (dir) {
4547
4548                         spin_lock(&obj->oo_guard);
4549                         if (obj->oo_dir == NULL)
4550                                 obj->oo_dir = dir;
4551                         else
4552                                 /*
4553                                  * Concurrent thread allocated container data.
4554                                  */
4555                                 OBD_FREE_PTR(dir);
4556                         spin_unlock(&obj->oo_guard);
4557                         /*
4558                          * Now, that we have container data, serialize its
4559                          * initialization.
4560                          */
4561                         down_write(&obj->oo_ext_idx_sem);
4562                         /*
4563                          * recheck under lock.
4564                          */
4565
4566                         if (osd_has_index(obj)) {
4567                                 result = 0;
4568                                 goto unlock;
4569                         }
4570
4571                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4572                         if (result || feat == &dt_lfsck_namespace_features ||
4573                             feat == &dt_lfsck_layout_orphan_features ||
4574                             feat == &dt_lfsck_layout_dangling_features)
4575                                 goto unlock;
4576
4577                         result = osd_index_register(osd, fid,
4578                                                     feat->dif_keysize_max,
4579                                                     feat->dif_recsize_max);
4580                         if (result < 0)
4581                                 CWARN("%s: failed to register index "
4582                                       DFID": rc = %d\n",
4583                                       osd_name(osd), PFID(fid), result);
4584                         else if (result > 0)
4585                                 result = 0;
4586                         else
4587                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4588                                        " (%d/%d) registered\n",
4589                                        osd_name(osd), PFID(fid),
4590                                        (int)feat->dif_keysize_max,
4591                                        (int)feat->dif_recsize_max);
4592
4593 unlock:
4594                         up_write(&obj->oo_ext_idx_sem);
4595                 } else {
4596                         result = -ENOMEM;
4597                 }
4598         } else {
4599                 result = 0;
4600         }
4601
4602         if (result == 0 && skip_iam == 0) {
4603                 if (!osd_iam_index_probe(env, obj, feat))
4604                         result = -ENOTDIR;
4605         }
4606         LINVRNT(osd_invariant(obj));
4607
4608         return result;
4609 }
4610
4611 static int osd_otable_it_attr_get(const struct lu_env *env,
4612                                  struct dt_object *dt,
4613                                  struct lu_attr *attr)
4614 {
4615         attr->la_valid = 0;
4616         return 0;
4617 }
4618
4619 static const struct dt_object_operations osd_obj_ops = {
4620         .do_read_lock           = osd_read_lock,
4621         .do_write_lock          = osd_write_lock,
4622         .do_read_unlock         = osd_read_unlock,
4623         .do_write_unlock        = osd_write_unlock,
4624         .do_write_locked        = osd_write_locked,
4625         .do_attr_get            = osd_attr_get,
4626         .do_declare_attr_set    = osd_declare_attr_set,
4627         .do_attr_set            = osd_attr_set,
4628         .do_ah_init             = osd_ah_init,
4629         .do_declare_create      = osd_declare_create,
4630         .do_create              = osd_create,
4631         .do_declare_destroy     = osd_declare_destroy,
4632         .do_destroy             = osd_destroy,
4633         .do_index_try           = osd_index_try,
4634         .do_declare_ref_add     = osd_declare_ref_add,
4635         .do_ref_add             = osd_ref_add,
4636         .do_declare_ref_del     = osd_declare_ref_del,
4637         .do_ref_del             = osd_ref_del,
4638         .do_xattr_get           = osd_xattr_get,
4639         .do_declare_xattr_set   = osd_declare_xattr_set,
4640         .do_xattr_set           = osd_xattr_set,
4641         .do_declare_xattr_del   = osd_declare_xattr_del,
4642         .do_xattr_del           = osd_xattr_del,
4643         .do_xattr_list          = osd_xattr_list,
4644         .do_object_sync         = osd_object_sync,
4645         .do_invalidate          = osd_invalidate,
4646 };
4647
4648 static const struct dt_object_operations osd_obj_otable_it_ops = {
4649         .do_attr_get    = osd_otable_it_attr_get,
4650         .do_index_try   = osd_index_try,
4651 };
4652
4653 static int osd_index_declare_iam_delete(const struct lu_env *env,
4654                                         struct dt_object *dt,
4655                                         const struct dt_key *key,
4656                                         struct thandle *handle)
4657 {
4658         struct osd_thandle    *oh;
4659
4660         oh = container_of0(handle, struct osd_thandle, ot_super);
4661         LASSERT(oh->ot_handle == NULL);
4662
4663         /* Recycle  may cause additional three blocks to be changed. */
4664         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4665                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4666
4667         return 0;
4668 }
4669
4670 /**
4671  *      delete a (key, value) pair from index \a dt specified by \a key
4672  *
4673  *      \param  dt      osd index object
4674  *      \param  key     key for index
4675  *      \param  rec     record reference
4676  *      \param  handle  transaction handler
4677  *
4678  *      \retval  0  success
4679  *      \retval -ve   failure
4680  */
4681 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4682                                 const struct dt_key *key,
4683                                 struct thandle *handle)
4684 {
4685         struct osd_thread_info *oti = osd_oti_get(env);
4686         struct osd_object      *obj = osd_dt_obj(dt);
4687         struct osd_thandle     *oh;
4688         struct iam_path_descr  *ipd;
4689         struct iam_container   *bag = &obj->oo_dir->od_container;
4690         int                     rc;
4691         ENTRY;
4692
4693         if (!dt_object_exists(dt))
4694                 RETURN(-ENOENT);
4695
4696         LINVRNT(osd_invariant(obj));
4697         LASSERT(!dt_object_remote(dt));
4698         LASSERT(bag->ic_object == obj->oo_inode);
4699         LASSERT(handle != NULL);
4700
4701         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4702
4703         ipd = osd_idx_ipd_get(env, bag);
4704         if (unlikely(ipd == NULL))
4705                 RETURN(-ENOMEM);
4706
4707         oh = container_of0(handle, struct osd_thandle, ot_super);
4708         LASSERT(oh->ot_handle != NULL);
4709         LASSERT(oh->ot_handle->h_transaction != NULL);
4710
4711         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4712                 /* swab quota uid/gid provided by caller */
4713                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4714                 key = (const struct dt_key *)&oti->oti_quota_id;
4715         }
4716
4717         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4718         osd_ipd_put(env, bag, ipd);
4719         LINVRNT(osd_invariant(obj));
4720         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4721         RETURN(rc);
4722 }
4723
4724 static int osd_index_declare_ea_delete(const struct lu_env *env,
4725                                        struct dt_object *dt,
4726                                        const struct dt_key *key,
4727                                        struct thandle *handle)
4728 {
4729         struct osd_thandle *oh;
4730         struct inode       *inode;
4731         int                 rc, credits;
4732         ENTRY;
4733
4734         LASSERT(!dt_object_remote(dt));
4735         LASSERT(handle != NULL);
4736
4737         oh = container_of0(handle, struct osd_thandle, ot_super);
4738         LASSERT(oh->ot_handle == NULL);
4739
4740         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4741         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4742
4743         inode = osd_dt_obj(dt)->oo_inode;
4744         if (inode == NULL)
4745                 RETURN(-ENOENT);
4746
4747         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4748                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4749                                    NULL, OSD_QID_BLK);
4750         RETURN(rc);
4751 }
4752
4753 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4754                                           struct dt_rec *fid)
4755 {
4756         struct osd_fid_pack *rec;
4757         int                  rc = -ENODATA;
4758
4759         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4760                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4761                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4762                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4763                         rc = -EINVAL;
4764         }
4765         return rc;
4766 }
4767
4768 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4769                           const struct lu_fid *fid)
4770 {
4771         struct seq_server_site  *ss = osd_seq_site(osd);
4772         ENTRY;
4773
4774         /* FID seqs not in FLDB, must be local seq */
4775         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4776                 RETURN(0);
4777
4778         /* If FLD is not being initialized yet, it only happens during the
4779          * initialization, likely during mgs initialization, and we assume
4780          * this is local FID. */
4781         if (ss == NULL || ss->ss_server_fld == NULL)
4782                 RETURN(0);
4783
4784         /* Only check the local FLDB here */
4785         if (osd_seq_exists(env, osd, fid_seq(fid)))
4786                 RETURN(0);
4787
4788         RETURN(1);
4789 }
4790
4791 static void osd_take_care_of_agent(const struct lu_env *env,
4792                                    struct osd_device *osd,
4793                                    struct osd_thandle *oh,
4794                                    struct ldiskfs_dir_entry_2 *de)
4795 {
4796         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4797         struct osd_idmap_cache *idc;
4798         int rc, schedule = 0;
4799
4800         LASSERT(de != NULL);
4801
4802         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4803         if (likely(rc == 0)) {
4804                 idc = osd_idc_find_or_init(env, osd, fid);
4805                 if (IS_ERR(idc) || idc->oic_remote)
4806                         schedule = 1;
4807         } else if (rc == -ENODATA) {
4808                 /* can't get FID, postpone to the end of the
4809                  * transaction when iget() is safe */
4810                 schedule = 1;
4811         } else {
4812                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
4813         }
4814         if (schedule)
4815                 osd_schedule_agent_inode_removal(env, oh,
4816                                                  le32_to_cpu(de->inode));
4817 }
4818
4819 /**
4820  * Index delete function for interoperability mode (b11826).
4821  * It will remove the directory entry added by osd_index_ea_insert().
4822  * This entry is needed to maintain name->fid mapping.
4823  *
4824  * \param key,  key i.e. file entry to be deleted
4825  *
4826  * \retval   0, on success
4827  * \retval -ve, on error
4828  */
4829 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4830                                const struct dt_key *key, struct thandle *handle)
4831 {
4832         struct osd_object          *obj = osd_dt_obj(dt);
4833         struct inode               *dir = obj->oo_inode;
4834         struct dentry              *dentry;
4835         struct osd_thandle         *oh;
4836         struct ldiskfs_dir_entry_2 *de = NULL;
4837         struct buffer_head         *bh;
4838         struct htree_lock          *hlock = NULL;
4839         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4840         int                        rc;
4841         ENTRY;
4842
4843         if (!dt_object_exists(dt))
4844                 RETURN(-ENOENT);
4845
4846         LINVRNT(osd_invariant(obj));
4847         LASSERT(!dt_object_remote(dt));
4848         LASSERT(handle != NULL);
4849
4850         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4851
4852         oh = container_of(handle, struct osd_thandle, ot_super);
4853         LASSERT(oh->ot_handle != NULL);
4854         LASSERT(oh->ot_handle->h_transaction != NULL);
4855
4856         ll_vfs_dq_init(dir);
4857         dentry = osd_child_dentry_get(env, obj,
4858                                       (char *)key, strlen((char *)key));
4859
4860         if (obj->oo_hl_head != NULL) {
4861                 hlock = osd_oti_get(env)->oti_hlock;
4862                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4863                                    dir, LDISKFS_HLOCK_DEL);
4864         } else {
4865                 down_write(&obj->oo_ext_idx_sem);
4866         }
4867
4868         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4869         if (!IS_ERR(bh)) {
4870                 /* If this is not the ".." entry, it might be a remote DNE
4871                  * entry and  we need to check if the FID is for a remote
4872                  * MDT.  If the FID is  not in the directory entry (e.g.
4873                  * upgraded 1.8 filesystem without dirdata enabled) then
4874                  * we need to get the FID from the LMA. For a remote directory
4875                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4876                  * case.
4877                  *
4878                  * Delete the entry before the agent inode in order to
4879                  * simplify error handling.  At worst an error after deleting
4880                  * the entry first might leak the agent inode afterward. The
4881                  * reverse would need filesystem abort in case of error deleting
4882                  * the entry after the agent had been removed, or leave a
4883                  * dangling entry pointing at a random inode. */
4884                 if (strcmp((char *)key, dotdot) != 0)
4885                         osd_take_care_of_agent(env, osd, oh, de);
4886                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4887                 brelse(bh);
4888         } else {
4889                 rc = PTR_ERR(bh);
4890         }
4891         if (hlock != NULL)
4892                 ldiskfs_htree_unlock(hlock);
4893         else
4894                 up_write(&obj->oo_ext_idx_sem);
4895
4896         GOTO(out, rc);
4897
4898 out:
4899         LASSERT(osd_invariant(obj));
4900         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4901         RETURN(rc);
4902 }
4903
4904 /**
4905  *      Lookup index for \a key and copy record to \a rec.
4906  *
4907  *      \param  dt      osd index object
4908  *      \param  key     key for index
4909  *      \param  rec     record reference
4910  *
4911  *      \retval  +ve  success : exact mach
4912  *      \retval  0    return record with key not greater than \a key
4913  *      \retval -ve   failure
4914  */
4915 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4916                                 struct dt_rec *rec, const struct dt_key *key)
4917 {
4918         struct osd_object      *obj = osd_dt_obj(dt);
4919         struct iam_path_descr  *ipd;
4920         struct iam_container   *bag = &obj->oo_dir->od_container;
4921         struct osd_thread_info *oti = osd_oti_get(env);
4922         struct iam_iterator    *it = &oti->oti_idx_it;
4923         struct iam_rec         *iam_rec;
4924         int                     rc;
4925         ENTRY;
4926
4927         if (!dt_object_exists(dt))
4928                 RETURN(-ENOENT);
4929
4930         LASSERT(osd_invariant(obj));
4931         LASSERT(!dt_object_remote(dt));
4932         LASSERT(bag->ic_object == obj->oo_inode);
4933
4934         ipd = osd_idx_ipd_get(env, bag);
4935         if (IS_ERR(ipd))
4936                 RETURN(-ENOMEM);
4937
4938         /* got ipd now we can start iterator. */
4939         iam_it_init(it, bag, 0, ipd);
4940
4941         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4942                 /* swab quota uid/gid provided by caller */
4943                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4944                 key = (const struct dt_key *)&oti->oti_quota_id;
4945         }
4946
4947         rc = iam_it_get(it, (struct iam_key *)key);
4948         if (rc >= 0) {
4949                 if (S_ISDIR(obj->oo_inode->i_mode))
4950                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4951                 else
4952                         iam_rec = (struct iam_rec *) rec;
4953
4954                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4955
4956                 if (S_ISDIR(obj->oo_inode->i_mode))
4957                         osd_fid_unpack((struct lu_fid *) rec,
4958                                        (struct osd_fid_pack *)iam_rec);
4959                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4960                         osd_quota_unpack(obj, rec);
4961         }
4962
4963         iam_it_put(it);
4964         iam_it_fini(it);
4965         osd_ipd_put(env, bag, ipd);
4966
4967         LINVRNT(osd_invariant(obj));
4968
4969         RETURN(rc);
4970 }
4971
4972 static int osd_index_declare_iam_insert(const struct lu_env *env,
4973                                         struct dt_object *dt,
4974                                         const struct dt_rec *rec,
4975                                         const struct dt_key *key,
4976                                         struct thandle *handle)
4977 {
4978         struct osd_thandle *oh;
4979
4980         LASSERT(handle != NULL);
4981
4982         oh = container_of0(handle, struct osd_thandle, ot_super);
4983         LASSERT(oh->ot_handle == NULL);
4984
4985         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4986                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4987
4988         return 0;
4989 }
4990
4991 /**
4992  *      Inserts (key, value) pair in \a dt index object.
4993  *
4994  *      \param  dt      osd index object
4995  *      \param  key     key for index
4996  *      \param  rec     record reference
4997  *      \param  th      transaction handler
4998  *
4999  *      \retval  0  success
5000  *      \retval -ve failure
5001  */
5002 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5003                                 const struct dt_rec *rec,
5004                                 const struct dt_key *key, struct thandle *th,
5005                                 int ignore_quota)
5006 {
5007         struct osd_object     *obj = osd_dt_obj(dt);
5008         struct iam_path_descr *ipd;
5009         struct osd_thandle    *oh;
5010         struct iam_container  *bag;
5011         struct osd_thread_info *oti = osd_oti_get(env);
5012         struct iam_rec         *iam_rec;
5013         int                     rc;
5014         ENTRY;
5015
5016         if (!dt_object_exists(dt))
5017                 RETURN(-ENOENT);
5018
5019         LINVRNT(osd_invariant(obj));
5020         LASSERT(!dt_object_remote(dt));
5021
5022         bag = &obj->oo_dir->od_container;
5023         LASSERT(bag->ic_object == obj->oo_inode);
5024         LASSERT(th != NULL);
5025
5026         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5027
5028         ipd = osd_idx_ipd_get(env, bag);
5029         if (unlikely(ipd == NULL))
5030                 RETURN(-ENOMEM);
5031
5032         oh = container_of0(th, struct osd_thandle, ot_super);
5033         LASSERT(oh->ot_handle != NULL);
5034         LASSERT(oh->ot_handle->h_transaction != NULL);
5035         if (S_ISDIR(obj->oo_inode->i_mode)) {
5036                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5037                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
5038         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5039                 /* pack quota uid/gid */
5040                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5041                 key = (const struct dt_key *)&oti->oti_quota_id;
5042                 /* pack quota record */
5043                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5044                 iam_rec = (struct iam_rec *)rec;
5045         } else {
5046                 iam_rec = (struct iam_rec *)rec;
5047         }
5048
5049         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5050                         iam_rec, ipd);
5051         osd_ipd_put(env, bag, ipd);
5052         LINVRNT(osd_invariant(obj));
5053         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5054         RETURN(rc);
5055 }
5056
5057 /**
5058  * Calls ldiskfs_add_entry() to add directory entry
5059  * into the directory. This is required for
5060  * interoperability mode (b11826)
5061  *
5062  * \retval   0, on success
5063  * \retval -ve, on error
5064  */
5065 static int __osd_ea_add_rec(struct osd_thread_info *info,
5066                             struct osd_object *pobj, struct inode  *cinode,
5067                             const char *name, const struct lu_fid *fid,
5068                             struct htree_lock *hlock, struct thandle *th)
5069 {
5070         struct ldiskfs_dentry_param *ldp;
5071         struct dentry               *child;
5072         struct osd_thandle          *oth;
5073         int                          rc;
5074
5075         oth = container_of(th, struct osd_thandle, ot_super);
5076         LASSERT(oth->ot_handle != NULL);
5077         LASSERT(oth->ot_handle->h_transaction != NULL);
5078         LASSERT(pobj->oo_inode);
5079
5080         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5081         if (unlikely(pobj->oo_inode ==
5082                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5083                 ldp->edp_magic = 0;
5084         else
5085                 osd_get_ldiskfs_dirent_param(ldp, fid);
5086         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5087         child->d_fsdata = (void *)ldp;
5088         ll_vfs_dq_init(pobj->oo_inode);
5089         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5090                                    child, cinode, hlock);
5091         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5092                 struct ldiskfs_dir_entry_2      *de;
5093                 struct buffer_head              *bh;
5094                 int                              rc1;
5095
5096                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5097                                             NULL, hlock);
5098                 if (!IS_ERR(bh)) {
5099                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5100                                                                bh);
5101                         if (rc1 == 0) {
5102                                 if (S_ISDIR(cinode->i_mode))
5103                                         de->file_type = LDISKFS_DIRENT_LUFID |
5104                                                         LDISKFS_FT_REG_FILE;
5105                                 else
5106                                         de->file_type = LDISKFS_DIRENT_LUFID |
5107                                                         LDISKFS_FT_DIR;
5108                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5109                                                               NULL, bh);
5110                         }
5111                         brelse(bh);
5112                 }
5113         }
5114
5115         RETURN(rc);
5116 }
5117
5118 /**
5119  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5120  * into the directory.Also sets flags into osd object to
5121  * indicate dot and dotdot are created. This is required for
5122  * interoperability mode (b11826)
5123  *
5124  * \param dir   directory for dot and dotdot fixup.
5125  * \param obj   child object for linking
5126  *
5127  * \retval   0, on success
5128  * \retval -ve, on error
5129  */
5130 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5131                               struct osd_object *dir,
5132                               struct inode *parent_dir, const char *name,
5133                               const struct lu_fid *dot_fid,
5134                               const struct lu_fid *dot_dot_fid,
5135                               struct thandle *th)
5136 {
5137         struct inode                *inode = dir->oo_inode;
5138         struct osd_thandle          *oth;
5139         int result = 0;
5140
5141         oth = container_of(th, struct osd_thandle, ot_super);
5142         LASSERT(oth->ot_handle->h_transaction != NULL);
5143         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5144
5145         if (strcmp(name, dot) == 0) {
5146                 if (dir->oo_compat_dot_created) {
5147                         result = -EEXIST;
5148                 } else {
5149                         LASSERT(inode->i_ino == parent_dir->i_ino);
5150                         dir->oo_compat_dot_created = 1;
5151                         result = 0;
5152                 }
5153         } else if (strcmp(name, dotdot) == 0) {
5154                 if (!dir->oo_compat_dot_created)
5155                         return -EINVAL;
5156                 /* in case of rename, dotdot is already created */
5157                 if (dir->oo_compat_dotdot_created) {
5158                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5159                                                 dot_dot_fid, NULL, th);
5160                 }
5161
5162                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5163                         struct lu_fid tfid = *dot_dot_fid;
5164
5165                         tfid.f_oid--;
5166                         result = osd_add_dot_dotdot_internal(info,
5167                                         dir->oo_inode, parent_dir, dot_fid,
5168                                         &tfid, oth);
5169                 } else {
5170                         result = osd_add_dot_dotdot_internal(info,
5171                                         dir->oo_inode, parent_dir, dot_fid,
5172                                         dot_dot_fid, oth);
5173                 }
5174
5175                 if (result == 0)
5176                         dir->oo_compat_dotdot_created = 1;
5177         }
5178
5179         return result;
5180 }
5181
5182
5183 /**
5184  * It will call the appropriate osd_add* function and return the
5185  * value, return by respective functions.
5186  */
5187 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5188                           struct inode *cinode, const char *name,
5189                           const struct lu_fid *fid, struct thandle *th)
5190 {
5191         struct osd_thread_info *info   = osd_oti_get(env);
5192         struct htree_lock      *hlock;
5193         int                     rc;
5194
5195         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5196
5197         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
5198                                                    name[2] =='\0'))) {
5199                 if (hlock != NULL) {
5200                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5201                                            pobj->oo_inode, 0);
5202                 } else {
5203                         down_write(&pobj->oo_ext_idx_sem);
5204                 }
5205
5206                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5207                                         lu_object_fid(&pobj->oo_dt.do_lu),
5208                                         fid, th);
5209         } else {
5210                 if (hlock != NULL) {
5211                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5212                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5213                 } else {
5214                         down_write(&pobj->oo_ext_idx_sem);
5215                 }
5216
5217                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5218                         struct lu_fid *tfid = &info->oti_fid;
5219
5220                         *tfid = *fid;
5221                         tfid->f_ver = ~0;
5222                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5223                                               tfid, hlock, th);
5224                 } else {
5225                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5226                                               hlock, th);
5227                 }
5228         }
5229         if (hlock != NULL)
5230                 ldiskfs_htree_unlock(hlock);
5231         else
5232                 up_write(&pobj->oo_ext_idx_sem);
5233
5234         return rc;
5235 }
5236
5237 static int
5238 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5239                       struct osd_idmap_cache *oic)
5240 {
5241         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5242         struct lu_fid *fid = &oic->oic_fid;
5243         struct osd_inode_id *id = &oic->oic_lid;
5244         struct inode *inode = NULL;
5245         int once = 0;
5246         bool insert;
5247         int rc;
5248         ENTRY;
5249
5250         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5251                 RETURN(0);
5252
5253         if (thread_is_running(&scrub->os_thread) &&
5254             scrub->os_pos_current > id->oii_ino)
5255                 RETURN(0);
5256
5257         if (dev->od_auto_scrub_interval == AS_NEVER ||
5258             ktime_get_real_seconds() <
5259             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5260                 RETURN(0);
5261
5262 again:
5263         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5264         if (rc == -ENOENT) {
5265                 __u32 gen = id->oii_gen;
5266
5267                 insert = true;
5268                 if (inode != NULL)
5269                         goto trigger;
5270
5271                 inode = osd_iget(oti, dev, id);
5272                 /* The inode has been removed (by race maybe). */
5273                 if (IS_ERR(inode)) {
5274                         rc = PTR_ERR(inode);
5275
5276                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5277                 }
5278
5279                 /* The OI mapping is lost. */
5280                 if (gen != OSD_OII_NOGEN)
5281                         goto trigger;
5282
5283                 /* The inode may has been reused by others, we do not know,
5284                  * leave it to be handled by subsequent osd_fid_lookup(). */
5285                 GOTO(out, rc = 0);
5286         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5287                 GOTO(out, rc);
5288         }
5289
5290         insert = false;
5291
5292 trigger:
5293         if (thread_is_running(&scrub->os_thread)) {
5294                 if (inode == NULL) {
5295                         inode = osd_iget(oti, dev, id);
5296                         /* The inode has been removed (by race maybe). */
5297                         if (IS_ERR(inode)) {
5298                                 rc = PTR_ERR(inode);
5299
5300                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5301                         }
5302                 }
5303
5304                 rc = osd_oii_insert(dev, oic, insert);
5305                 /* There is race condition between osd_oi_lookup and OI scrub.
5306                  * The OI scrub finished just after osd_oi_lookup() failure.
5307                  * Under such case, it is unnecessary to trigger OI scrub again,
5308                  * but try to call osd_oi_lookup() again. */
5309                 if (unlikely(rc == -EAGAIN))
5310                         goto again;
5311
5312                 if (!S_ISDIR(inode->i_mode))
5313                         rc = 0;
5314                 else
5315                         rc = osd_check_lmv(oti, dev, inode, oic);
5316
5317                 GOTO(out, rc);
5318         }
5319
5320         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5321                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5322                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5323                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5324                        "%s: trigger partial OI scrub for RPC inconsistency "
5325                        "checking FID "DFID": rc = %d\n",
5326                        osd_dev2name(dev), PFID(fid), rc);
5327                 if (rc == 0 || rc == -EALREADY)
5328                         goto again;
5329         }
5330
5331         GOTO(out, rc);
5332
5333 out:
5334         if (inode)
5335                 iput(inode);
5336
5337         RETURN(rc);
5338 }
5339
5340 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5341                                struct osd_device *dev,
5342                                struct osd_idmap_cache *oic,
5343                                struct lu_fid *fid, __u32 ino)
5344 {
5345         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5346         struct inode *inode;
5347         int rc;
5348
5349         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5350         inode = osd_iget(oti, dev, &oic->oic_lid);
5351         if (IS_ERR(inode)) {
5352                 fid_zero(&oic->oic_fid);
5353                 return PTR_ERR(inode);
5354         }
5355
5356         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5357         iput(inode);
5358         if (rc != 0)
5359                 fid_zero(&oic->oic_fid);
5360         else
5361                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5362         return rc;
5363 }
5364
5365 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5366                       struct osd_inode_id *id, const struct lu_fid *fid)
5367 {
5368         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5369                id->oii_ino, id->oii_gen, info);
5370         info->oti_cache.oic_lid = *id;
5371         info->oti_cache.oic_fid = *fid;
5372         info->oti_cache.oic_dev = osd;
5373 }
5374
5375 /**
5376  * Get parent FID from the linkEA.
5377  *
5378  * For a directory which parent resides on remote MDT, to satisfy the
5379  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5380  * the other hand, to make the lookup(..) on the directory can return
5381  * the real parent FID, we append the real parent FID after its ".."
5382  * name entry in the /REMOTE_PARENT_DIR.
5383  *
5384  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5385  * backup. So after the restore, we cannot get the right parent FID from
5386  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5387  * we have stored the real parent FID in the directory object's linkEA,
5388  * we can parse the linkEA for the real parent FID.
5389  *
5390  * \param[in] env       pointer to the thread context
5391  * \param[in] obj       pointer to the object to be handled
5392  * \param[out]fid       pointer to the buffer to hold the parent FID
5393  *
5394  * \retval              0 for getting the real parent FID successfully
5395  * \retval              negative error number on failure
5396  */
5397 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5398                                     struct osd_object *obj,
5399                                     struct lu_fid *fid)
5400 {
5401         struct osd_thread_info  *oti    = osd_oti_get(env);
5402         struct lu_buf           *buf    = &oti->oti_big_buf;
5403         struct dentry           *dentry = &oti->oti_obj_dentry;
5404         struct inode            *inode  = obj->oo_inode;
5405         struct linkea_data       ldata  = { NULL };
5406         int                      rc;
5407         ENTRY;
5408
5409         fid_zero(fid);
5410         if (!S_ISDIR(inode->i_mode))
5411                 RETURN(-EIO);
5412
5413 again:
5414         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5415                              buf->lb_buf, buf->lb_len);
5416         if (rc == -ERANGE) {
5417                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5418                                      NULL, 0);
5419                 if (rc > 0) {
5420                         lu_buf_realloc(buf, rc);
5421                         if (buf->lb_buf == NULL)
5422                                 RETURN(-ENOMEM);
5423
5424                         goto again;
5425                 }
5426         }
5427
5428         if (unlikely(rc == 0))
5429                 RETURN(-ENODATA);
5430
5431         if (rc < 0)
5432                 RETURN(rc);
5433
5434         if (unlikely(buf->lb_buf == NULL)) {
5435                 lu_buf_realloc(buf, rc);
5436                 if (buf->lb_buf == NULL)
5437                         RETURN(-ENOMEM);
5438
5439                 goto again;
5440         }
5441
5442         ldata.ld_buf = buf;
5443         rc = linkea_init_with_rec(&ldata);
5444         if (!rc) {
5445                 linkea_first_entry(&ldata);
5446                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5447         }
5448
5449         RETURN(rc);
5450 }
5451
5452 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5453                                     struct inode *inode,
5454                                     const struct lu_fid *pfid,
5455                                     const char *name, const int namelen)
5456 {
5457         struct osd_thread_info *oti = osd_oti_get(env);
5458         struct lu_buf *buf = &oti->oti_big_buf;
5459         struct dentry *dentry = &oti->oti_obj_dentry;
5460         struct linkea_data ldata = { NULL };
5461         struct lu_name cname = { .ln_name = name,
5462                                  .ln_namelen = namelen };
5463         int rc;
5464         ENTRY;
5465
5466 again:
5467         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5468                              buf->lb_buf, buf->lb_len);
5469         if (rc == -ERANGE)
5470                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5471
5472         if (rc < 0)
5473                 RETURN(rc);
5474
5475         if (unlikely(rc == 0))
5476                 RETURN(-ENODATA);
5477
5478         if (buf->lb_len < rc) {
5479                 lu_buf_realloc(buf, rc);
5480                 if (buf->lb_buf == NULL)
5481                         RETURN(-ENOMEM);
5482
5483                 goto again;
5484         }
5485
5486         ldata.ld_buf = buf;
5487         rc = linkea_init_with_rec(&ldata);
5488         if (!rc)
5489                 rc = linkea_links_find(&ldata, &cname, pfid);
5490
5491         RETURN(rc);
5492 }
5493
5494 /**
5495  * Calls ->lookup() to find dentry. From dentry get inode and
5496  * read inode's ea to get fid. This is required for  interoperability
5497  * mode (b11826)
5498  *
5499  * \retval   0, on success
5500  * \retval -ve, on error
5501  */
5502 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5503                              struct dt_rec *rec, const struct dt_key *key)
5504 {
5505         struct inode                    *dir    = obj->oo_inode;
5506         struct dentry                   *dentry;
5507         struct ldiskfs_dir_entry_2      *de;
5508         struct buffer_head              *bh;
5509         struct lu_fid                   *fid = (struct lu_fid *) rec;
5510         struct htree_lock               *hlock = NULL;
5511         int                             ino;
5512         int                             rc;
5513         ENTRY;
5514
5515         LASSERT(dir->i_op != NULL);
5516         LASSERT(dir->i_op->lookup != NULL);
5517
5518         dentry = osd_child_dentry_get(env, obj,
5519                                       (char *)key, strlen((char *)key));
5520
5521         if (obj->oo_hl_head != NULL) {
5522                 hlock = osd_oti_get(env)->oti_hlock;
5523                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5524                                    dir, LDISKFS_HLOCK_LOOKUP);
5525         } else {
5526                 down_read(&obj->oo_ext_idx_sem);
5527         }
5528
5529         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5530         if (!IS_ERR(bh)) {
5531                 struct osd_thread_info *oti = osd_oti_get(env);
5532                 struct osd_inode_id *id = &oti->oti_id;
5533                 struct osd_idmap_cache *oic = &oti->oti_cache;
5534                 struct osd_device *dev = osd_obj2dev(obj);
5535
5536                 ino = le32_to_cpu(de->inode);
5537                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5538                         brelse(bh);
5539                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5540                         GOTO(out, rc);
5541                 }
5542
5543                 rc = osd_get_fid_from_dentry(de, rec);
5544
5545                 /* done with de, release bh */
5546                 brelse(bh);
5547                 if (rc != 0) {
5548                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5549                                 const char *name = (const char *)key;
5550
5551                                 /* If the parent is on remote MDT, and there
5552                                  * is no FID-in-dirent, then we have to get
5553                                  * the parent FID from the linkEA.  */
5554                                 if (likely(strlen(name) == 2 &&
5555                                            name[0] == '.' && name[1] == '.'))
5556                                         rc = osd_get_pfid_from_linkea(env, obj,
5557                                                                       fid);
5558                         } else {
5559                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5560                         }
5561                 } else {
5562                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5563                 }
5564
5565                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5566                         fid_zero(&oic->oic_fid);
5567
5568                         GOTO(out, rc);
5569                 }
5570
5571                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5572                 rc = osd_consistency_check(oti, dev, oic);
5573                 if (rc == -ENOENT)
5574                         fid_zero(&oic->oic_fid);
5575                 else
5576                         /* Other error should not affect lookup result. */
5577                         rc = 0;
5578         } else {
5579                 rc = PTR_ERR(bh);
5580         }
5581
5582         GOTO(out, rc);
5583
5584 out:
5585         if (hlock != NULL)
5586                 ldiskfs_htree_unlock(hlock);
5587         else
5588                 up_read(&obj->oo_ext_idx_sem);
5589         return rc;
5590 }
5591
5592 static int osd_index_declare_ea_insert(const struct lu_env *env,
5593                                        struct dt_object *dt,
5594                                        const struct dt_rec *rec,
5595                                        const struct dt_key *key,
5596                                        struct thandle *handle)
5597 {
5598         struct osd_thandle      *oh;
5599         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5600         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5601         const struct lu_fid     *fid    = rec1->rec_fid;
5602         int                      credits, rc = 0;
5603         struct osd_idmap_cache  *idc;
5604         ENTRY;
5605
5606         LASSERT(!dt_object_remote(dt));
5607         LASSERT(handle != NULL);
5608         LASSERT(fid != NULL);
5609         LASSERT(rec1->rec_type != 0);
5610
5611         oh = container_of0(handle, struct osd_thandle, ot_super);
5612         LASSERT(oh->ot_handle == NULL);
5613
5614         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5615
5616         /* we can't call iget() while a transactions is running
5617          * (this can lead to a deadlock), but we need to know
5618          * inum and object type. so we find this information at
5619          * declaration and cache in per-thread info */
5620         idc = osd_idc_find_or_init(env, osd, fid);
5621         if (IS_ERR(idc))
5622                 RETURN(PTR_ERR(idc));
5623         if (idc->oic_remote) {
5624                 /* a reference to remote inode is represented by an
5625                  * agent inode which we have to create */
5626                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5627                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5628         }
5629
5630         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5631
5632         if (osd_dt_obj(dt)->oo_inode != NULL) {
5633                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5634
5635                 /* We ignore block quota on meta pool (MDTs), so needn't
5636                  * calculate how many blocks will be consumed by this index
5637                  * insert */
5638                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5639                                            i_gid_read(inode),
5640                                            i_projid_read(inode), 0,
5641                                            oh, osd_dt_obj(dt), NULL,
5642                                            OSD_QID_BLK);
5643                 if (rc)
5644                         RETURN(rc);
5645
5646 #ifdef HAVE_PROJECT_QUOTA
5647                 /* Reserve credits for local agent inode to transfer
5648                  * to 0, quota enforcement is ignored in this case.
5649                  */
5650                 if (idc->oic_remote &&
5651                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5652                     i_projid_read(inode) != 0)
5653                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5654                                                   0, i_projid_read(inode),
5655                                                   0, false, PRJQUOTA, true);
5656 #endif
5657         }
5658
5659
5660         RETURN(rc);
5661 }
5662
5663 /**
5664  * Index add function for interoperability mode (b11826).
5665  * It will add the directory entry.This entry is needed to
5666  * maintain name->fid mapping.
5667  *
5668  * \param key it is key i.e. file entry to be inserted
5669  * \param rec it is value of given key i.e. fid
5670  *
5671  * \retval   0, on success
5672  * \retval -ve, on error
5673  */
5674 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5675                                const struct dt_rec *rec,
5676                                const struct dt_key *key, struct thandle *th,
5677                                int ignore_quota)
5678 {
5679         struct osd_object       *obj = osd_dt_obj(dt);
5680         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5681         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5682         const struct lu_fid     *fid    = rec1->rec_fid;
5683         const char              *name = (const char *)key;
5684         struct osd_thread_info  *oti   = osd_oti_get(env);
5685         struct inode            *child_inode = NULL;
5686         struct osd_idmap_cache  *idc;
5687         int                     rc;
5688         ENTRY;
5689
5690         if (!dt_object_exists(dt))
5691                 RETURN(-ENOENT);
5692
5693         LASSERT(osd_invariant(obj));
5694         LASSERT(!dt_object_remote(dt));
5695         LASSERT(th != NULL);
5696
5697         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5698
5699         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5700
5701         idc = osd_idc_find(env, osd, fid);
5702         if (unlikely(idc == NULL)) {
5703                 /* this dt_insert() wasn't declared properly, so
5704                  * FID is missing in OI cache. we better do not
5705                  * lookup FID in FLDB/OI and don't risk to deadlock,
5706                  * but in some special cases (lfsck testing, etc)
5707                  * it's much simpler than fixing a caller */
5708                 CERROR("%s: "DFID" wasn't declared for insert\n",
5709                        osd_name(osd), PFID(fid));
5710                 dump_stack();
5711                 idc = osd_idc_find_or_init(env, osd, fid);
5712                 if (IS_ERR(idc))
5713                         RETURN(PTR_ERR(idc));
5714         }
5715
5716         if (idc->oic_remote) {
5717                 /* Insert remote entry */
5718                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5719                         child_inode =
5720                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
5721                 } else {
5722                         child_inode = osd_create_local_agent_inode(env, osd,
5723                                         obj, fid, rec1->rec_type & S_IFMT, th);
5724                         if (IS_ERR(child_inode))
5725                                 RETURN(PTR_ERR(child_inode));
5726                 }
5727         } else {
5728                 /* Insert local entry */
5729                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5730                         /* for a reason OI cache wasn't filled properly */
5731                         CERROR("%s: OIC for "DFID" isn't filled\n",
5732                                osd_name(osd), PFID(fid));
5733                         RETURN(-EINVAL);
5734                 }
5735                 child_inode = oti->oti_inode;
5736                 if (unlikely(child_inode == NULL)) {
5737                         struct ldiskfs_inode_info *lii;
5738                         OBD_ALLOC_PTR(lii);
5739                         if (lii == NULL)
5740                                 RETURN(-ENOMEM);
5741                         child_inode = oti->oti_inode = &lii->vfs_inode;
5742                 }
5743                 child_inode->i_sb = osd_sb(osd);
5744                 child_inode->i_ino = idc->oic_lid.oii_ino;
5745                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5746         }
5747
5748         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5749
5750         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5751                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5752
5753         if (child_inode && child_inode != oti->oti_inode)
5754                 iput(child_inode);
5755         LASSERT(osd_invariant(obj));
5756         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5757         RETURN(rc);
5758 }
5759
5760 /**
5761  *  Initialize osd Iterator for given osd index object.
5762  *
5763  *  \param  dt      osd index object
5764  */
5765
5766 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5767                                      struct dt_object *dt,
5768                                      __u32 unused)
5769 {
5770         struct osd_it_iam      *it;
5771         struct osd_object      *obj = osd_dt_obj(dt);
5772         struct lu_object       *lo  = &dt->do_lu;
5773         struct iam_path_descr  *ipd;
5774         struct iam_container   *bag = &obj->oo_dir->od_container;
5775
5776         if (!dt_object_exists(dt))
5777                 return ERR_PTR(-ENOENT);
5778
5779         OBD_ALLOC_PTR(it);
5780         if (it == NULL)
5781                 return ERR_PTR(-ENOMEM);
5782
5783         ipd = osd_it_ipd_get(env, bag);
5784         if (likely(ipd != NULL)) {
5785                 it->oi_obj = obj;
5786                 it->oi_ipd = ipd;
5787                 lu_object_get(lo);
5788                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5789                 return (struct dt_it *)it;
5790         } else {
5791                 OBD_FREE_PTR(it);
5792                 return ERR_PTR(-ENOMEM);
5793         }
5794 }
5795
5796 /**
5797  * free given Iterator.
5798  */
5799 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5800 {
5801         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5802         struct osd_object       *obj = it->oi_obj;
5803
5804         iam_it_fini(&it->oi_it);
5805         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5806         osd_object_put(env, obj);
5807         OBD_FREE_PTR(it);
5808 }
5809
5810 /**
5811  *  Move Iterator to record specified by \a key
5812  *
5813  *  \param  di      osd iterator
5814  *  \param  key     key for index
5815  *
5816  *  \retval +ve  di points to record with least key not larger than key
5817  *  \retval  0   di points to exact matched key
5818  *  \retval -ve  failure
5819  */
5820
5821 static int osd_it_iam_get(const struct lu_env *env,
5822                           struct dt_it *di, const struct dt_key *key)
5823 {
5824         struct osd_thread_info  *oti = osd_oti_get(env);
5825         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5826
5827         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5828                 /* swab quota uid/gid */
5829                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5830                 key = (struct dt_key *)&oti->oti_quota_id;
5831         }
5832
5833         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5834 }
5835
5836 /**
5837  *  Release Iterator
5838  *
5839  *  \param  di      osd iterator
5840  */
5841 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5842 {
5843         struct osd_it_iam *it = (struct osd_it_iam *)di;
5844
5845         iam_it_put(&it->oi_it);
5846 }
5847
5848 /**
5849  *  Move iterator by one record
5850  *
5851  *  \param  di      osd iterator
5852  *
5853  *  \retval +1   end of container reached
5854  *  \retval  0   success
5855  *  \retval -ve  failure
5856  */
5857
5858 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5859 {
5860         struct osd_it_iam *it = (struct osd_it_iam *)di;
5861
5862         return iam_it_next(&it->oi_it);
5863 }
5864
5865 /**
5866  * Return pointer to the key under iterator.
5867  */
5868
5869 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5870                                  const struct dt_it *di)
5871 {
5872         struct osd_thread_info *oti = osd_oti_get(env);
5873         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5874         struct osd_object      *obj = it->oi_obj;
5875         struct dt_key          *key;
5876
5877         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5878
5879         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5880                 /* swab quota uid/gid */
5881                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5882                 key = (struct dt_key *)&oti->oti_quota_id;
5883         }
5884
5885         return key;
5886 }
5887
5888 /**
5889  * Return size of key under iterator (in bytes)
5890  */
5891
5892 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5893 {
5894         struct osd_it_iam *it = (struct osd_it_iam *)di;
5895
5896         return iam_it_key_size(&it->oi_it);
5897 }
5898
5899 static inline void
5900 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5901 {
5902         /* check if file type is required */
5903         if (ent->lde_attrs & LUDA_TYPE) {
5904                 struct luda_type *lt;
5905                 int align = sizeof(*lt) - 1;
5906
5907                 len = (len + align) & ~align;
5908                 lt = (struct luda_type *)(ent->lde_name + len);
5909                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5910         }
5911
5912         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5913 }
5914
5915 /**
5916  * build lu direct from backend fs dirent.
5917  */
5918
5919 static inline void
5920 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5921                    char *name, __u16 namelen, __u16 type, __u32 attr)
5922 {
5923         ent->lde_attrs = attr | LUDA_FID;
5924         fid_cpu_to_le(&ent->lde_fid, fid);
5925
5926         ent->lde_hash = cpu_to_le64(offset);
5927         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5928
5929         strncpy(ent->lde_name, name, namelen);
5930         ent->lde_name[namelen] = '\0';
5931         ent->lde_namelen = cpu_to_le16(namelen);
5932
5933         /* append lustre attributes */
5934         osd_it_append_attrs(ent, namelen, type);
5935 }
5936
5937 /**
5938  * Return pointer to the record under iterator.
5939  */
5940 static int osd_it_iam_rec(const struct lu_env *env,
5941                           const struct dt_it *di,
5942                           struct dt_rec *dtrec, __u32 attr)
5943 {
5944         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5945         struct osd_thread_info *info = osd_oti_get(env);
5946         ENTRY;
5947
5948         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5949                 const struct osd_fid_pack *rec;
5950                 struct lu_fid             *fid = &info->oti_fid;
5951                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5952                 char                      *name;
5953                 int                        namelen;
5954                 __u64                      hash;
5955                 int                        rc;
5956
5957                 name = (char *)iam_it_key_get(&it->oi_it);
5958                 if (IS_ERR(name))
5959                         RETURN(PTR_ERR(name));
5960
5961                 namelen = iam_it_key_size(&it->oi_it);
5962
5963                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5964                 if (IS_ERR(rec))
5965                         RETURN(PTR_ERR(rec));
5966
5967                 rc = osd_fid_unpack(fid, rec);
5968                 if (rc)
5969                         RETURN(rc);
5970
5971                 hash = iam_it_store(&it->oi_it);
5972
5973                 /* IAM does not store object type in IAM index (dir) */
5974                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5975                                    0, LUDA_FID);
5976         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5977                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5978                            (struct iam_rec *)dtrec);
5979                 osd_quota_unpack(it->oi_obj, dtrec);
5980         } else {
5981                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5982                            (struct iam_rec *)dtrec);
5983         }
5984
5985         RETURN(0);
5986 }
5987
5988 /**
5989  * Returns cookie for current Iterator position.
5990  */
5991 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5992 {
5993         struct osd_it_iam *it = (struct osd_it_iam *)di;
5994
5995         return iam_it_store(&it->oi_it);
5996 }
5997
5998 /**
5999  * Restore iterator from cookie.
6000  *
6001  * \param  di      osd iterator
6002  * \param  hash    Iterator location cookie
6003  *
6004  * \retval +ve  di points to record with least key not larger than key.
6005  * \retval  0   di points to exact matched key
6006  * \retval -ve  failure
6007  */
6008
6009 static int osd_it_iam_load(const struct lu_env *env,
6010                            const struct dt_it *di, __u64 hash)
6011 {
6012         struct osd_it_iam *it = (struct osd_it_iam *)di;
6013
6014         return iam_it_load(&it->oi_it, hash);
6015 }
6016
6017 static const struct dt_index_operations osd_index_iam_ops = {
6018         .dio_lookup         = osd_index_iam_lookup,
6019         .dio_declare_insert = osd_index_declare_iam_insert,
6020         .dio_insert         = osd_index_iam_insert,
6021         .dio_declare_delete = osd_index_declare_iam_delete,
6022         .dio_delete         = osd_index_iam_delete,
6023         .dio_it     = {
6024                 .init     = osd_it_iam_init,
6025                 .fini     = osd_it_iam_fini,
6026                 .get      = osd_it_iam_get,
6027                 .put      = osd_it_iam_put,
6028                 .next     = osd_it_iam_next,
6029                 .key      = osd_it_iam_key,
6030                 .key_size = osd_it_iam_key_size,
6031                 .rec      = osd_it_iam_rec,
6032                 .store    = osd_it_iam_store,
6033                 .load     = osd_it_iam_load
6034         }
6035 };
6036
6037
6038 /**
6039  * Creates or initializes iterator context.
6040  *
6041  * \retval struct osd_it_ea, iterator structure on success
6042  *
6043  */
6044 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6045                                     struct dt_object *dt,
6046                                     __u32 attr)
6047 {
6048         struct osd_object       *obj  = osd_dt_obj(dt);
6049         struct osd_thread_info  *info = osd_oti_get(env);
6050         struct osd_it_ea        *oie;
6051         struct file             *file;
6052         struct lu_object        *lo   = &dt->do_lu;
6053         struct dentry           *obj_dentry;
6054         ENTRY;
6055
6056         if (!dt_object_exists(dt) || obj->oo_destroyed)
6057                 RETURN(ERR_PTR(-ENOENT));
6058
6059         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6060         if (oie == NULL)
6061                 RETURN(ERR_PTR(-ENOMEM));
6062         obj_dentry = &oie->oie_dentry;
6063
6064         obj_dentry->d_inode = obj->oo_inode;
6065         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6066         obj_dentry->d_name.hash = 0;
6067
6068         oie->oie_rd_dirent       = 0;
6069         oie->oie_it_dirent       = 0;
6070         oie->oie_dirent          = NULL;
6071         if (unlikely(!info->oti_it_ea_buf_used)) {
6072                 oie->oie_buf = info->oti_it_ea_buf;
6073                 info->oti_it_ea_buf_used = 1;
6074         } else {
6075                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6076                 if (oie->oie_buf == NULL)
6077                         RETURN(ERR_PTR(-ENOMEM));
6078         }
6079         oie->oie_obj             = obj;
6080
6081         file = &oie->oie_file;
6082
6083         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6084         if (attr & LUDA_64BITHASH)
6085                 file->f_mode    = FMODE_64BITHASH;
6086         else
6087                 file->f_mode    = FMODE_32BITHASH;
6088         file->f_path.dentry     = obj_dentry;
6089         file->f_mapping         = obj->oo_inode->i_mapping;
6090         file->f_op              = obj->oo_inode->i_fop;
6091         set_file_inode(file, obj->oo_inode);
6092
6093         lu_object_get(lo);
6094         RETURN((struct dt_it *) oie);
6095 }
6096
6097 /**
6098  * Destroy or finishes iterator context.
6099  *
6100  * \param di iterator structure to be destroyed
6101  */
6102 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6103 {
6104         struct osd_thread_info  *info = osd_oti_get(env);
6105         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
6106         struct osd_object       *obj    = oie->oie_obj;
6107         struct inode            *inode  = obj->oo_inode;
6108
6109         ENTRY;
6110         oie->oie_file.f_op->release(inode, &oie->oie_file);
6111         osd_object_put(env, obj);
6112         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6113                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6114         else
6115                 info->oti_it_ea_buf_used = 0;
6116         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6117         EXIT;
6118 }
6119
6120 /**
6121  * It position the iterator at given key, so that next lookup continues from
6122  * that key Or it is similar to dio_it->load() but based on a key,
6123  * rather than file position.
6124  *
6125  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6126  * to the beginning.
6127  *
6128  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6129  */
6130 static int osd_it_ea_get(const struct lu_env *env,
6131                          struct dt_it *di, const struct dt_key *key)
6132 {
6133         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
6134
6135         ENTRY;
6136         LASSERT(((const char *)key)[0] == '\0');
6137         it->oie_file.f_pos      = 0;
6138         it->oie_rd_dirent       = 0;
6139         it->oie_it_dirent       = 0;
6140         it->oie_dirent          = NULL;
6141
6142         RETURN(+1);
6143 }
6144
6145 /**
6146  * Does nothing
6147  */
6148 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6149 {
6150 }
6151
6152 struct osd_filldir_cbs {
6153 #ifdef HAVE_DIR_CONTEXT
6154         struct dir_context ctx;
6155 #endif
6156         struct osd_it_ea  *it;
6157 };
6158 /**
6159  * It is called internally by ->readdir(). It fills the
6160  * iterator's in-memory data structure with required
6161  * information i.e. name, namelen, rec_size etc.
6162  *
6163  * \param buf in which information to be filled in.
6164  * \param name name of the file in given dir
6165  *
6166  * \retval 0 on success
6167  * \retval 1 on buffer full
6168  */
6169 #ifdef HAVE_FILLDIR_USE_CTX
6170 static int osd_ldiskfs_filldir(struct dir_context *buf,
6171 #else
6172 static int osd_ldiskfs_filldir(void *buf,
6173 #endif
6174                                const char *name, int namelen,
6175                                loff_t offset, __u64 ino, unsigned d_type)
6176 {
6177         struct osd_it_ea        *it   =
6178                 ((struct osd_filldir_cbs *)buf)->it;
6179         struct osd_object       *obj  = it->oie_obj;
6180         struct osd_it_ea_dirent *ent  = it->oie_dirent;
6181         struct lu_fid           *fid  = &ent->oied_fid;
6182         struct osd_fid_pack     *rec;
6183         ENTRY;
6184
6185         /* this should never happen */
6186         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6187                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6188                 RETURN(-EIO);
6189         }
6190
6191         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
6192             OSD_IT_EA_BUFSIZE)
6193                 RETURN(1);
6194
6195         /* "." is just the object itself. */
6196         if (namelen == 1 && name[0] == '.') {
6197                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6198         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6199                 rec = (struct osd_fid_pack*) (name + namelen + 1);
6200                 if (osd_fid_unpack(fid, rec) != 0)
6201                         fid_zero(fid);
6202         } else {
6203                 fid_zero(fid);
6204         }
6205         d_type &= ~LDISKFS_DIRENT_LUFID;
6206
6207         /* NOT export local root. */
6208         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6209                 ino = obj->oo_inode->i_ino;
6210                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6211         }
6212
6213         ent->oied_ino     = ino;
6214         ent->oied_off     = offset;
6215         ent->oied_namelen = namelen;
6216         ent->oied_type    = d_type;
6217
6218         memcpy(ent->oied_name, name, namelen);
6219
6220         it->oie_rd_dirent++;
6221         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6222         RETURN(0);
6223 }
6224
6225 /**
6226  * Calls ->readdir() to load a directory entry at a time
6227  * and stored it in iterator's in-memory data structure.
6228  *
6229  * \param di iterator's in memory structure
6230  *
6231  * \retval   0 on success
6232  * \retval -ve on error
6233  * \retval +1 reach the end of entry
6234  */
6235 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6236                                const struct dt_it *di)
6237 {
6238         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
6239         struct osd_object  *obj   = it->oie_obj;
6240         struct inode       *inode = obj->oo_inode;
6241         struct htree_lock  *hlock = NULL;
6242         struct file        *filp  = &it->oie_file;
6243         int                 rc = 0;
6244         struct osd_filldir_cbs buf = {
6245 #ifdef HAVE_DIR_CONTEXT
6246                 .ctx.actor = osd_ldiskfs_filldir,
6247 #endif
6248                 .it = it
6249         };
6250
6251         ENTRY;
6252         it->oie_dirent = it->oie_buf;
6253         it->oie_rd_dirent = 0;
6254
6255         if (obj->oo_hl_head != NULL) {
6256                 hlock = osd_oti_get(env)->oti_hlock;
6257                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6258                                    inode, LDISKFS_HLOCK_READDIR);
6259         } else {
6260                 down_read(&obj->oo_ext_idx_sem);
6261         }
6262
6263 #ifdef HAVE_DIR_CONTEXT
6264         buf.ctx.pos = filp->f_pos;
6265         rc = inode->i_fop->iterate(filp, &buf.ctx);
6266         filp->f_pos = buf.ctx.pos;
6267 #else
6268         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6269 #endif
6270
6271         if (hlock != NULL)
6272                 ldiskfs_htree_unlock(hlock);
6273         else
6274                 up_read(&obj->oo_ext_idx_sem);
6275
6276         if (it->oie_rd_dirent == 0) {
6277                 /*If it does not get any dirent, it means it has been reached
6278                  *to the end of the dir */
6279                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6280                 if (rc == 0)
6281                         rc = 1;
6282         } else {
6283                 it->oie_dirent = it->oie_buf;
6284                 it->oie_it_dirent = 1;
6285         }
6286
6287         RETURN(rc);
6288 }
6289
6290 /**
6291  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6292  * to load a directory entry at a time and stored it in
6293  * iterator's in-memory data structure.
6294  *
6295  * \param di iterator's in memory structure
6296  *
6297  * \retval +ve iterator reached to end
6298  * \retval   0 iterator not reached to end
6299  * \retval -ve on error
6300  */
6301 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6302 {
6303         struct osd_it_ea *it = (struct osd_it_ea *)di;
6304         int rc;
6305
6306         ENTRY;
6307
6308         if (it->oie_it_dirent < it->oie_rd_dirent) {
6309                 it->oie_dirent =
6310                         (void *) it->oie_dirent +
6311                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6312                                        it->oie_dirent->oied_namelen);
6313                 it->oie_it_dirent++;
6314                 RETURN(0);
6315         } else {
6316                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6317                         rc = +1;
6318                 else
6319                         rc = osd_ldiskfs_it_fill(env, di);
6320         }
6321
6322         RETURN(rc);
6323 }
6324
6325 /**
6326  * Returns the key at current position from iterator's in memory structure.
6327  *
6328  * \param di iterator's in memory structure
6329  *
6330  * \retval key i.e. struct dt_key on success
6331  */
6332 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6333                                     const struct dt_it *di)
6334 {
6335         struct osd_it_ea *it = (struct osd_it_ea *)di;
6336
6337         return (struct dt_key *)it->oie_dirent->oied_name;
6338 }
6339
6340 /**
6341  * Returns the key's size at current position from iterator's in memory structure.
6342  *
6343  * \param di iterator's in memory structure
6344  *
6345  * \retval key_size i.e. struct dt_key on success
6346  */
6347 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6348 {
6349         struct osd_it_ea *it = (struct osd_it_ea *)di;
6350
6351         return it->oie_dirent->oied_namelen;
6352 }
6353
6354 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6355 {
6356         if (LDISKFS_DIR_REC_LEN(de) >=
6357             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6358                 return true;
6359
6360         return false;
6361 }
6362
6363 static inline bool
6364 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6365                      unsigned blocksize, bool dotdot)
6366 {
6367         if (dotdot)
6368                 return osd_dotdot_has_space(de);
6369
6370         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6371             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6372                 return true;
6373
6374         return false;
6375 }
6376
6377 static int
6378 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6379                     handle_t *jh, struct dentry *dentry,
6380                     const struct lu_fid *fid, struct buffer_head *bh,
6381                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6382                     bool dotdot)
6383 {
6384         struct inode                *dir        = dentry->d_parent->d_inode;
6385         struct inode                *inode      = dentry->d_inode;
6386         struct osd_fid_pack         *rec;
6387         struct ldiskfs_dentry_param *ldp;
6388         int                          namelen    = dentry->d_name.len;
6389         int                          rc;
6390         struct osd_thread_info     *info        = osd_oti_get(env);
6391         ENTRY;
6392
6393         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6394                 RETURN(0);
6395
6396         /* There is enough space to hold the FID-in-dirent. */
6397         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6398                 rc = ldiskfs_journal_get_write_access(jh, bh);
6399                 if (rc != 0)
6400                         RETURN(rc);
6401
6402                 de->name[namelen] = 0;
6403                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6404                 rec->fp_len = sizeof(struct lu_fid) + 1;
6405                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6406                 de->file_type |= LDISKFS_DIRENT_LUFID;
6407                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6408
6409                 RETURN(rc);
6410         }
6411
6412         LASSERT(!dotdot);
6413
6414         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6415         if (rc != 0)
6416                 RETURN(rc);
6417
6418         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6419         osd_get_ldiskfs_dirent_param(ldp, fid);
6420         dentry->d_fsdata = (void *)ldp;
6421         ll_vfs_dq_init(dir);
6422         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6423         /* It is too bad, we cannot reinsert the name entry back.
6424          * That means we lose it! */
6425         if (rc != 0)
6426                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6427                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6428                        osd_ino2name(inode),
6429                        dir->i_ino, dir->i_generation, namelen,
6430                        dentry->d_name.name, PFID(fid), rc);
6431
6432         RETURN(rc);
6433 }
6434
6435 static int
6436 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6437                         struct osd_it_ea *it, struct lu_fid *fid,
6438                         struct osd_inode_id *id, __u32 *attr)
6439 {
6440         struct osd_thread_info     *info        = osd_oti_get(env);
6441         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6442         struct osd_device          *dev         = osd_obj2dev(obj);
6443         struct super_block         *sb          = osd_sb(dev);
6444         const char                 *devname     = osd_name(dev);
6445         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6446         struct inode               *dir         = obj->oo_inode;
6447         struct htree_lock          *hlock       = NULL;
6448         struct buffer_head         *bh          = NULL;
6449         handle_t                   *jh          = NULL;
6450         struct ldiskfs_dir_entry_2 *de;
6451         struct dentry              *dentry;
6452         struct inode               *inode;
6453         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6454         int                         credits;
6455         int                         rc;
6456         bool                        dotdot      = false;
6457         bool                        dirty       = false;
6458         ENTRY;
6459
6460         if (ent->oied_name[0] == '.') {
6461                 if (ent->oied_namelen == 1)
6462                         RETURN(0);
6463
6464                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6465                         dotdot = true;
6466         }
6467
6468         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6469         inode = osd_iget(info, dev, id);
6470         if (IS_ERR(inode)) {
6471                 rc = PTR_ERR(inode);
6472                 if (rc == -ENOENT || rc == -ESTALE) {
6473                         /* Maybe dangling name entry, or
6474                          * corrupted directory entry. */
6475                         *attr |= LUDA_UNKNOWN;
6476                         rc = 0;
6477                 } else {
6478                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6479                                "check_repair, dir = %lu/%u, name = %.*s, "
6480                                "ino = %llu, rc = %d\n",
6481                                devname, dir->i_ino, dir->i_generation,
6482                                ent->oied_namelen, ent->oied_name,
6483                                ent->oied_ino, rc);
6484                 }
6485
6486                 RETURN(rc);
6487         }
6488
6489         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6490                                            ent->oied_namelen);
6491         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6492         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6493                 lma = NULL;
6494         else if (rc != 0)
6495                 GOTO(out, rc);
6496
6497         /* We need to ensure that the name entry is still valid.
6498          * Because it may be removed or renamed by other already.
6499          *
6500          * The unlink or rename operation will start journal before PDO lock,
6501          * so to avoid deadlock, here we need to start journal handle before
6502          * related PDO lock also. But because we do not know whether there
6503          * will be something to be repaired before PDO lock, we just start
6504          * journal without conditions.
6505          *
6506          * We may need to remove the name entry firstly, then insert back.
6507          * One credit is for user quota file update.
6508          * One credit is for group quota file update.
6509          * Two credits are for dirty inode. */
6510         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6511                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6512
6513         if (dev->od_dirent_journal != 0) {
6514
6515 again:
6516                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6517                 if (IS_ERR(jh)) {
6518                         rc = PTR_ERR(jh);
6519                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6520                                "check_repair, dir = %lu/%u, credits = %d, "
6521                                "name = %.*s, ino = %llu: rc = %d\n",
6522                                devname, dir->i_ino, dir->i_generation, credits,
6523                                ent->oied_namelen, ent->oied_name,
6524                                ent->oied_ino, rc);
6525
6526                         GOTO(out_inode, rc);
6527                 }
6528
6529                 if (obj->oo_hl_head != NULL) {
6530                         hlock = osd_oti_get(env)->oti_hlock;
6531                         /* "0" means exclusive lock for the whole directory.
6532                          * We need to prevent others access such name entry
6533                          * during the delete + insert. Neither HLOCK_ADD nor
6534                          * HLOCK_DEL cannot guarantee the atomicity. */
6535                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6536                 } else {
6537                         down_write(&obj->oo_ext_idx_sem);
6538                 }
6539         } else {
6540                 if (obj->oo_hl_head != NULL) {
6541                         hlock = osd_oti_get(env)->oti_hlock;
6542                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6543                                            LDISKFS_HLOCK_LOOKUP);
6544                 } else {
6545                         down_read(&obj->oo_ext_idx_sem);
6546                 }
6547         }
6548
6549         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6550         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6551                 *attr |= LUDA_IGNORE;
6552
6553                 GOTO(out, rc = 0);
6554         }
6555
6556         /* For dotdot entry, if there is not enough space to hold the
6557          * FID-in-dirent, just keep them there. It only happens when the
6558          * device upgraded from 1.8 or restored from MDT file-level backup.
6559          * For the whole directory, only dotdot entry have no FID-in-dirent
6560          * and needs to get FID from LMA when readdir, it will not affect the
6561          * performance much. */
6562         if (dotdot && !osd_dotdot_has_space(de)) {
6563                 *attr |= LUDA_UNKNOWN;
6564
6565                 GOTO(out, rc = 0);
6566         }
6567
6568         if (lma != NULL) {
6569                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6570                         GOTO(out, rc = 0);
6571
6572                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6573                         struct lu_fid *tfid = &lma->lma_self_fid;
6574
6575                         if (likely(dotdot &&
6576                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6577                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6578                                 /* It must be REMOTE_PARENT_DIR and as the
6579                                  * 'dotdot' entry of remote directory */
6580                                 *attr |= LUDA_IGNORE;
6581                         } else {
6582                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6583                                        "parent directory, but got %.*s under "
6584                                        "dir = %lu/%u with the FID "DFID"\n",
6585                                        devname, ent->oied_namelen,
6586                                        ent->oied_name, dir->i_ino,
6587                                        dir->i_generation, PFID(tfid));
6588
6589                                 *attr |= LUDA_UNKNOWN;
6590                         }
6591
6592                         GOTO(out, rc = 0);
6593                 }
6594         }
6595
6596         if (!fid_is_zero(fid)) {
6597                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6598                                               ent->oied_namelen);
6599                 if (rc == -ENOENT ||
6600                     (rc == -ENODATA &&
6601                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6602                         /* linkEA does not recognize the dirent entry,
6603                          * it may because the dirent entry corruption
6604                          * and points to other's inode. */
6605                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6606                                "recognize the dirent, dir = %lu/%u, "
6607                                " name = %.*s, ino = %llu, "
6608                                DFID": rc = %d\n", devname, dir->i_ino,
6609                                dir->i_generation, ent->oied_namelen,
6610                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6611                         *attr |= LUDA_UNKNOWN;
6612
6613                         GOTO(out, rc = 0);
6614                 }
6615
6616                 if (rc && rc != -ENODATA) {
6617                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6618                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6619                                DFID": rc = %d\n", devname, dir->i_ino,
6620                                dir->i_generation, ent->oied_namelen,
6621                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6622                         *attr |= LUDA_UNKNOWN;
6623
6624                         GOTO(out, rc = 0);
6625                 }
6626         }
6627
6628         if (lma != NULL) {
6629                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6630                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6631                 if (*attr & LUDA_VERIFY_DRYRUN) {
6632                         *fid = lma->lma_self_fid;
6633                         *attr |= LUDA_REPAIR;
6634
6635                         GOTO(out, rc = 0);
6636                 }
6637
6638                 if (jh == NULL) {
6639                         brelse(bh);
6640                         dev->od_dirent_journal = 1;
6641                         if (hlock != NULL) {
6642                                 ldiskfs_htree_unlock(hlock);
6643                                 hlock = NULL;
6644                         } else {
6645                                 up_read(&obj->oo_ext_idx_sem);
6646                         }
6647
6648                         goto again;
6649                 }
6650
6651                 *fid = lma->lma_self_fid;
6652                 dirty = true;
6653                 /* Update or append the FID-in-dirent. */
6654                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6655                                          bh, de, hlock, dotdot);
6656                 if (rc == 0)
6657                         *attr |= LUDA_REPAIR;
6658                 else
6659                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6660                                "the dirent, dir = %lu/%u, name = %.*s, "
6661                                "ino = %llu, "DFID": rc = %d\n",
6662                                devname, dir->i_ino, dir->i_generation,
6663                                ent->oied_namelen, ent->oied_name,
6664                                ent->oied_ino, PFID(fid), rc);
6665         } else {
6666                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6667                 if (*attr & LUDA_VERIFY_DRYRUN) {
6668                         if (fid_is_sane(fid)) {
6669                                 *attr |= LUDA_REPAIR;
6670                         } else if (dev->od_index == 0) {
6671                                 lu_igif_build(fid, inode->i_ino,
6672                                               inode->i_generation);
6673                                 *attr |= LUDA_UPGRADE;
6674                         }
6675
6676                         GOTO(out, rc = 0);
6677                 }
6678
6679                 if (jh == NULL) {
6680                         brelse(bh);
6681                         dev->od_dirent_journal = 1;
6682                         if (hlock != NULL) {
6683                                 ldiskfs_htree_unlock(hlock);
6684                                 hlock = NULL;
6685                         } else {
6686                                 up_read(&obj->oo_ext_idx_sem);
6687                         }
6688
6689                         goto again;
6690                 }
6691
6692                 dirty = true;
6693                 if (unlikely(fid_is_sane(fid))) {
6694                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6695                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6696                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6697                         if (rc == 0)
6698                                 *attr |= LUDA_REPAIR;
6699                         else
6700                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6701                                        "update dirent, dir = %lu/%u, "
6702                                        "name = %.*s, ino = %llu, "
6703                                        DFID": rc = %d\n",
6704                                        devname, dir->i_ino, dir->i_generation,
6705                                        ent->oied_namelen, ent->oied_name,
6706                                        ent->oied_ino, PFID(fid), rc);
6707                 } else if (dev->od_index == 0) {
6708                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6709                         /* It is probably IGIF object. Only aappend the
6710                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6711                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6712                                                  bh, de, hlock, dotdot);
6713                         if (rc == 0)
6714                                 *attr |= LUDA_UPGRADE;
6715                         else
6716                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6717                                        "after the dirent, dir = %lu/%u, "
6718                                        "name = %.*s, ino = %llu, "
6719                                        DFID": rc = %d\n",
6720                                        devname, dir->i_ino, dir->i_generation,
6721                                        ent->oied_namelen, ent->oied_name,
6722                                        ent->oied_ino, PFID(fid), rc);
6723                 }
6724         }
6725
6726         GOTO(out, rc);
6727
6728 out:
6729         if (!IS_ERR(bh))
6730                 brelse(bh);
6731         if (hlock != NULL) {
6732                 ldiskfs_htree_unlock(hlock);
6733         } else {
6734                 if (dev->od_dirent_journal != 0)
6735                         up_write(&obj->oo_ext_idx_sem);
6736                 else
6737                         up_read(&obj->oo_ext_idx_sem);
6738         }
6739
6740         if (jh != NULL)
6741                 ldiskfs_journal_stop(jh);
6742
6743 out_inode:
6744         iput(inode);
6745         if (rc >= 0 && !dirty)
6746                 dev->od_dirent_journal = 0;
6747
6748         return rc;
6749 }
6750
6751 /**
6752  * Returns the value at current position from iterator's in memory structure.
6753  *
6754  * \param di struct osd_it_ea, iterator's in memory structure
6755  * \param attr attr requested for dirent.
6756  * \param lde lustre dirent
6757  *
6758  * \retval   0 no error and \param lde has correct lustre dirent.
6759  * \retval -ve on error
6760  */
6761 static inline int osd_it_ea_rec(const struct lu_env *env,
6762                                 const struct dt_it *di,
6763                                 struct dt_rec *dtrec, __u32 attr)
6764 {
6765         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6766         struct osd_object      *obj   = it->oie_obj;
6767         struct osd_device      *dev   = osd_obj2dev(obj);
6768         struct osd_thread_info *oti   = osd_oti_get(env);
6769         struct osd_inode_id    *id    = &oti->oti_id;
6770         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6771         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6772         __u32                   ino   = it->oie_dirent->oied_ino;
6773         int                     rc    = 0;
6774         ENTRY;
6775
6776         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
6777
6778         if (attr & LUDA_VERIFY) {
6779                 if (unlikely(is_remote_parent_ino(dev, ino))) {
6780                         attr |= LUDA_IGNORE;
6781                         /* If the parent is on remote MDT, and there
6782                          * is no FID-in-dirent, then we have to get
6783                          * the parent FID from the linkEA.  */
6784                         if (!fid_is_sane(fid) &&
6785                             it->oie_dirent->oied_namelen == 2 &&
6786                             it->oie_dirent->oied_name[0] == '.' &&
6787                             it->oie_dirent->oied_name[1] == '.')
6788                                 osd_get_pfid_from_linkea(env, obj, fid);
6789                 } else {
6790                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6791                                                      &attr);
6792                 }
6793
6794                 if (!fid_is_sane(fid))
6795                         attr |= LUDA_UNKNOWN;
6796         } else {
6797                 attr &= ~LU_DIRENT_ATTRS_MASK;
6798                 if (!fid_is_sane(fid)) {
6799                         bool is_dotdot = false;
6800                         if (it->oie_dirent->oied_namelen == 2 &&
6801                             it->oie_dirent->oied_name[0] == '.' &&
6802                             it->oie_dirent->oied_name[1] == '.')
6803                                 is_dotdot = true;
6804                         /* If the parent is on remote MDT, and there
6805                          * is no FID-in-dirent, then we have to get
6806                          * the parent FID from the linkEA.  */
6807                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
6808                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6809                         } else {
6810                                 if (is_dotdot == false &&
6811                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6812                                         RETURN(-ENOENT);
6813
6814                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6815                         }
6816                 } else {
6817                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6818                 }
6819         }
6820
6821         /* Pack the entry anyway, at least the offset is right. */
6822         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6823                            it->oie_dirent->oied_name,
6824                            it->oie_dirent->oied_namelen,
6825                            it->oie_dirent->oied_type, attr);
6826
6827         if (rc < 0)
6828                 RETURN(rc);
6829
6830         if (osd_remote_fid(env, dev, fid))
6831                 RETURN(0);
6832
6833         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6834                 osd_add_oi_cache(oti, dev, id, fid);
6835
6836         RETURN(rc > 0 ? 0 : rc);
6837 }
6838
6839 /**
6840  * Returns the record size size at current position.
6841  *
6842  * This function will return record(lu_dirent) size in bytes.
6843  *
6844  * \param[in] env       execution environment
6845  * \param[in] di        iterator's in memory structure
6846  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6847  *                      calculate the lu_dirent size.
6848  *
6849  * \retval      record size(in bytes & in memory) of the current lu_dirent
6850  *              entry.
6851  */
6852 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6853                               __u32 attr)
6854 {
6855         struct osd_it_ea *it = (struct osd_it_ea *)di;
6856
6857         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6858 }
6859
6860 /**
6861  * Returns a cookie for current position of the iterator head, so that
6862  * user can use this cookie to load/start the iterator next time.
6863  *
6864  * \param di iterator's in memory structure
6865  *
6866  * \retval cookie for current position, on success
6867  */
6868 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6869 {
6870         struct osd_it_ea *it = (struct osd_it_ea *)di;
6871
6872         return it->oie_dirent->oied_off;
6873 }
6874
6875 /**
6876  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6877  * to load a directory entry at a time and stored it i inn,
6878  * in iterator's in-memory data structure.
6879  *
6880  * \param di struct osd_it_ea, iterator's in memory structure
6881  *
6882  * \retval +ve on success
6883  * \retval -ve on error
6884  */
6885 static int osd_it_ea_load(const struct lu_env *env,
6886                           const struct dt_it *di, __u64 hash)
6887 {
6888         struct osd_it_ea *it = (struct osd_it_ea *)di;
6889         int rc;
6890
6891         ENTRY;
6892         it->oie_file.f_pos = hash;
6893
6894         rc =  osd_ldiskfs_it_fill(env, di);
6895         if (rc > 0)
6896                 rc = -ENODATA;
6897
6898         if (rc == 0)
6899                 rc = +1;
6900
6901         RETURN(rc);
6902 }
6903
6904 /**
6905  * Index lookup function for interoperability mode (b11826).
6906  *
6907  * \param key,  key i.e. file name to be searched
6908  *
6909  * \retval +ve, on success
6910  * \retval -ve, on error
6911  */
6912 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6913                                struct dt_rec *rec, const struct dt_key *key)
6914 {
6915         struct osd_object *obj = osd_dt_obj(dt);
6916         int rc = 0;
6917
6918         ENTRY;
6919
6920         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6921         LINVRNT(osd_invariant(obj));
6922
6923         rc = osd_ea_lookup_rec(env, obj, rec, key);
6924         if (rc == 0)
6925                 rc = +1;
6926         RETURN(rc);
6927 }
6928
6929 /**
6930  * Index and Iterator operations for interoperability
6931  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6932  */
6933 static const struct dt_index_operations osd_index_ea_ops = {
6934         .dio_lookup         = osd_index_ea_lookup,
6935         .dio_declare_insert = osd_index_declare_ea_insert,
6936         .dio_insert         = osd_index_ea_insert,
6937         .dio_declare_delete = osd_index_declare_ea_delete,
6938         .dio_delete         = osd_index_ea_delete,
6939         .dio_it     = {
6940                 .init     = osd_it_ea_init,
6941                 .fini     = osd_it_ea_fini,
6942                 .get      = osd_it_ea_get,
6943                 .put      = osd_it_ea_put,
6944                 .next     = osd_it_ea_next,
6945                 .key      = osd_it_ea_key,
6946                 .key_size = osd_it_ea_key_size,
6947                 .rec      = osd_it_ea_rec,
6948                 .rec_size = osd_it_ea_rec_size,
6949                 .store    = osd_it_ea_store,
6950                 .load     = osd_it_ea_load
6951         }
6952 };
6953
6954 static void *osd_key_init(const struct lu_context *ctx,
6955                           struct lu_context_key *key)
6956 {
6957         struct osd_thread_info *info;
6958
6959         OBD_ALLOC_PTR(info);
6960         if (info == NULL)
6961                 return ERR_PTR(-ENOMEM);
6962
6963         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6964         if (info->oti_it_ea_buf == NULL)
6965                 goto out_free_info;
6966
6967         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6968
6969         info->oti_hlock = ldiskfs_htree_lock_alloc();
6970         if (info->oti_hlock == NULL)
6971                 goto out_free_ea;
6972
6973         return info;
6974
6975  out_free_ea:
6976         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6977  out_free_info:
6978         OBD_FREE_PTR(info);
6979         return ERR_PTR(-ENOMEM);
6980 }
6981
6982 static void osd_key_fini(const struct lu_context *ctx,
6983                          struct lu_context_key *key, void* data)
6984 {
6985         struct osd_thread_info *info = data;
6986         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6987         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6988
6989         if (info->oti_inode != NULL)
6990                 OBD_FREE_PTR(lli);
6991         if (info->oti_hlock != NULL)
6992                 ldiskfs_htree_lock_free(info->oti_hlock);
6993         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6994         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6995         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6996         lu_buf_free(&info->oti_big_buf);
6997         if (idc != NULL) {
6998                 LASSERT(info->oti_ins_cache_size > 0);
6999                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7000                 info->oti_ins_cache = NULL;
7001                 info->oti_ins_cache_size = 0;
7002         }
7003         OBD_FREE_PTR(info);
7004 }
7005
7006 static void osd_key_exit(const struct lu_context *ctx,
7007                          struct lu_context_key *key, void *data)
7008 {
7009         struct osd_thread_info *info = data;
7010
7011         LASSERT(info->oti_r_locks == 0);
7012         LASSERT(info->oti_w_locks == 0);
7013         LASSERT(info->oti_txns    == 0);
7014 }
7015
7016 /* type constructor/destructor: osd_type_init, osd_type_fini */
7017 LU_TYPE_INIT_FINI(osd, &osd_key);
7018
7019 struct lu_context_key osd_key = {
7020         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7021         .lct_init = osd_key_init,
7022         .lct_fini = osd_key_fini,
7023         .lct_exit = osd_key_exit
7024 };
7025
7026
7027 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7028                            const char *name, struct lu_device *next)
7029 {
7030         struct osd_device *osd = osd_dev(d);
7031
7032         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
7033             >= sizeof(osd->od_svname))
7034                 return -E2BIG;
7035         return osd_procfs_init(osd, name);
7036 }
7037
7038 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7039 {
7040         struct seq_server_site  *ss = osd_seq_site(osd);
7041         int                     rc;
7042         ENTRY;
7043
7044         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7045                 RETURN(0);
7046
7047         if (unlikely(ss == NULL))
7048                 RETURN(-ENODEV);
7049
7050         OBD_ALLOC_PTR(osd->od_cl_seq);
7051         if (osd->od_cl_seq == NULL)
7052                 RETURN(-ENOMEM);
7053
7054         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7055                              osd->od_svname, ss->ss_server_seq);
7056         if (rc != 0) {
7057                 OBD_FREE_PTR(osd->od_cl_seq);
7058                 osd->od_cl_seq = NULL;
7059                 RETURN(rc);
7060         }
7061
7062         if (ss->ss_node_id == 0) {
7063                 /* If the OSD on the sequence controller(MDT0), then allocate
7064                  * sequence here, otherwise allocate sequence after connected
7065                  * to MDT0 (see mdt_register_lwp_callback()). */
7066                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7067                                    &osd->od_cl_seq->lcs_space, env);
7068         }
7069
7070         RETURN(rc);
7071 }
7072
7073 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7074 {
7075         if (osd->od_cl_seq == NULL)
7076                 return;
7077
7078         seq_client_fini(osd->od_cl_seq);
7079         OBD_FREE_PTR(osd->od_cl_seq);
7080         osd->od_cl_seq = NULL;
7081 }
7082
7083 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7084 {
7085         ENTRY;
7086
7087         /* shutdown quota slave instance associated with the device */
7088         if (o->od_quota_slave != NULL) {
7089                 struct qsd_instance *qsd = o->od_quota_slave;
7090
7091                 o->od_quota_slave = NULL;
7092                 qsd_fini(env, qsd);
7093         }
7094
7095         osd_fid_fini(env, o);
7096         osd_scrub_cleanup(env, o);
7097
7098         RETURN(0);
7099 }
7100
7101 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7102 {
7103         ENTRY;
7104
7105         if (o->od_mnt != NULL) {
7106                 shrink_dcache_sb(osd_sb(o));
7107                 osd_sync(env, &o->od_dt_dev);
7108
7109                 mntput(o->od_mnt);
7110                 o->od_mnt = NULL;
7111         }
7112
7113         EXIT;
7114 }
7115
7116 static int osd_mount(const struct lu_env *env,
7117                      struct osd_device *o, struct lustre_cfg *cfg)
7118 {
7119         const char              *name  = lustre_cfg_string(cfg, 0);
7120         const char              *dev  = lustre_cfg_string(cfg, 1);
7121         const char              *opts;
7122         unsigned long            page, s_flags, lmd_flags = 0;
7123         struct page             *__page;
7124         struct file_system_type *type;
7125         char                    *options = NULL;
7126         char                    *str;
7127         struct osd_thread_info  *info = osd_oti_get(env);
7128         struct lu_fid           *fid = &info->oti_fid;
7129         struct inode            *inode;
7130         int                      rc = 0, force_over_512tb = 0;
7131         ENTRY;
7132
7133         if (o->od_mnt != NULL)
7134                 RETURN(0);
7135
7136         if (strlen(dev) >= sizeof(o->od_mntdev))
7137                 RETURN(-E2BIG);
7138         strcpy(o->od_mntdev, dev);
7139
7140         str = lustre_cfg_string(cfg, 2);
7141         s_flags = simple_strtoul(str, NULL, 0);
7142         str = strstr(str, ":");
7143         if (str)
7144                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7145         opts = lustre_cfg_string(cfg, 3);
7146 #ifdef __BIG_ENDIAN
7147         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7148                 CERROR("%s: device %s extents feature is not guaranteed to "
7149                        "work on big-endian systems. Use \"bigendian_extents\" "
7150                        "mount option to override.\n", name, dev);
7151                 RETURN(-EINVAL);
7152         }
7153 #endif
7154 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7155         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7156                 CWARN("force_over_128tb option is deprecated. "
7157                       "Filesystems less than 512TB can be created without any "
7158                       "force options. Use force_over_512tb option for "
7159                       "filesystems greater than 512TB.\n");
7160         }
7161 #endif
7162 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7163         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7164                 CWARN("force_over_256tb option is deprecated. "
7165                       "Filesystems less than 512TB can be created without any "
7166                       "force options. Use force_over_512tb option for "
7167                       "filesystems greater than 512TB.\n");
7168         }
7169 #endif
7170
7171         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7172                 force_over_512tb = 1;
7173
7174         __page = alloc_page(GFP_KERNEL);
7175         if (__page == NULL)
7176                 GOTO(out, rc = -ENOMEM);
7177         page = (unsigned long)page_address(__page);
7178         options = (char *)page;
7179         *options = '\0';
7180         if (opts != NULL) {
7181                 /* strip out the options for back compatiblity */
7182                 static char *sout[] = {
7183                         "mballoc",
7184                         "iopen",
7185                         "noiopen",
7186                         "iopen_nopriv",
7187                         "extents",
7188                         "noextents",
7189                         /* strip out option we processed in osd */
7190                         "bigendian_extents",
7191                         "force_over_128tb",
7192                         "force_over_256tb",
7193                         "force_over_512tb",
7194                         NULL
7195                 };
7196                 strcat(options, opts);
7197                 for (rc = 0, str = options; sout[rc]; ) {
7198                         char *op = strstr(str, sout[rc]);
7199                         if (op == NULL) {
7200                                 rc++;
7201                                 str = options;
7202                                 continue;
7203                         }
7204                         if (op == options || *(op - 1) == ',') {
7205                                 str = op + strlen(sout[rc]);
7206                                 if (*str == ',' || *str == '\0') {
7207                                         *str == ',' ? str++ : str;
7208                                         memmove(op, str, strlen(str) + 1);
7209                                 }
7210                         }
7211                         for (str = op; *str != ',' && *str != '\0'; str++)
7212                                 ;
7213                 }
7214         } else {
7215                 strncat(options, "user_xattr,acl", 14);
7216         }
7217
7218         /* Glom up mount options */
7219         if (*options != '\0')
7220                 strcat(options, ",");
7221         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7222
7223         type = get_fs_type("ldiskfs");
7224         if (!type) {
7225                 CERROR("%s: cannot find ldiskfs module\n", name);
7226                 GOTO(out, rc = -ENODEV);
7227         }
7228
7229         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7230         module_put(type->owner);
7231
7232         if (IS_ERR(o->od_mnt)) {
7233                 rc = PTR_ERR(o->od_mnt);
7234                 o->od_mnt = NULL;
7235                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7236                 GOTO(out, rc);
7237         }
7238
7239         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7240                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7241                                  force_over_512tb == 0) {
7242                 CERROR("%s: device %s LDISKFS does not support filesystems "
7243                        "greater than 512TB and can cause data corruption. "
7244                        "Use \"force_over_512tb\" mount option to override.\n",
7245                        name, dev);
7246                 GOTO(out_mnt, rc = -EINVAL);
7247         }
7248
7249         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7250                 if (priv_dev_set_rdonly) {
7251                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7252                         o->od_dt_dev.dd_rdonly = 1;
7253                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7254                                       name);
7255                 } else {
7256                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7257                                       name);
7258
7259                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7260                 }
7261         } else if (priv_dev_check_rdonly &&
7262                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7263                 CERROR("%s: underlying device %s is marked as "
7264                        "read-only. Setup failed\n", name, dev);
7265
7266                 GOTO(out_mnt, rc = -EROFS);
7267         }
7268
7269         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7270                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7271                 GOTO(out_mnt, rc = -EINVAL);
7272         }
7273
7274 #ifdef LDISKFS_MOUNT_DIRDATA
7275         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7276                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7277         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7278                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7279                       "enabling the dirdata feature. If you do not want to "
7280                       "downgrade to Lustre-1.x again, you can enable it via "
7281                       "'tune2fs -O dirdata device'\n", name, dev);
7282 #endif
7283         inode = osd_sb(o)->s_root->d_inode;
7284         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7285         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7286         if (rc != 0) {
7287                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7288                 GOTO(out_mnt, rc);
7289         }
7290
7291         if (lmd_flags & LMD_FLG_NOSCRUB)
7292                 o->od_auto_scrub_interval = AS_NEVER;
7293
7294         GOTO(out, rc = 0);
7295
7296 out_mnt:
7297         mntput(o->od_mnt);
7298         o->od_mnt = NULL;
7299
7300 out:
7301         if (__page)
7302                 __free_page(__page);
7303
7304         return rc;
7305 }
7306
7307 static struct lu_device *osd_device_fini(const struct lu_env *env,
7308                                          struct lu_device *d)
7309 {
7310         struct osd_device *o = osd_dev(d);
7311         ENTRY;
7312
7313         osd_index_backup(env, o, false);
7314         osd_shutdown(env, o);
7315         osd_procfs_fini(o);
7316         osd_obj_map_fini(o);
7317         osd_umount(env, o);
7318
7319         RETURN(NULL);
7320 }
7321
7322 static int osd_device_init0(const struct lu_env *env,
7323                             struct osd_device *o,
7324                             struct lustre_cfg *cfg)
7325 {
7326         struct lu_device        *l = osd2lu_dev(o);
7327         struct osd_thread_info *info;
7328         int                     rc;
7329         int                     cplen = 0;
7330
7331         /* if the module was re-loaded, env can loose its keys */
7332         rc = lu_env_refill((struct lu_env *) env);
7333         if (rc)
7334                 GOTO(out, rc);
7335         info = osd_oti_get(env);
7336         LASSERT(info);
7337
7338         l->ld_ops = &osd_lu_ops;
7339         o->od_dt_dev.dd_ops = &osd_dt_ops;
7340
7341         spin_lock_init(&o->od_osfs_lock);
7342         mutex_init(&o->od_otable_mutex);
7343         INIT_LIST_HEAD(&o->od_orphan_list);
7344         INIT_LIST_HEAD(&o->od_index_backup_list);
7345         INIT_LIST_HEAD(&o->od_index_restore_list);
7346         spin_lock_init(&o->od_lock);
7347         o->od_index_backup_policy = LIBP_NONE;
7348
7349         o->od_read_cache = 1;
7350         o->od_writethrough_cache = 1;
7351         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7352         o->od_auto_scrub_interval = AS_DEFAULT;
7353
7354         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7355                         sizeof(o->od_svname));
7356         if (cplen >= sizeof(o->od_svname)) {
7357                 rc = -E2BIG;
7358                 GOTO(out, rc);
7359         }
7360
7361         o->od_index_backup_stop = 0;
7362         o->od_index = -1; /* -1 means index is invalid */
7363         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7364         if (rc == LDD_F_SV_TYPE_OST)
7365                 o->od_is_ost = 1;
7366
7367         o->od_full_scrub_ratio = OFSR_DEFAULT;
7368         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7369         rc = osd_mount(env, o, cfg);
7370         if (rc != 0)
7371                 GOTO(out, rc);
7372
7373         rc = osd_obj_map_init(env, o);
7374         if (rc != 0)
7375                 GOTO(out_mnt, rc);
7376
7377         rc = lu_site_init(&o->od_site, l);
7378         if (rc != 0)
7379                 GOTO(out_compat, rc);
7380         o->od_site.ls_bottom_dev = l;
7381
7382         rc = lu_site_init_finish(&o->od_site);
7383         if (rc != 0)
7384                 GOTO(out_site, rc);
7385
7386         INIT_LIST_HEAD(&o->od_ios_list);
7387         /* setup scrub, including OI files initialization */
7388         o->od_in_init = 1;
7389         rc = osd_scrub_setup(env, o);
7390         o->od_in_init = 0;
7391         if (rc < 0)
7392                 GOTO(out_site, rc);
7393
7394         rc = osd_procfs_init(o, o->od_svname);
7395         if (rc != 0) {
7396                 CERROR("%s: can't initialize procfs: rc = %d\n",
7397                        o->od_svname, rc);
7398                 GOTO(out_scrub, rc);
7399         }
7400
7401         LASSERT(l->ld_site->ls_linkage.next != NULL);
7402         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7403
7404         /* initialize quota slave instance */
7405         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7406                                      o->od_proc_entry);
7407         if (IS_ERR(o->od_quota_slave)) {
7408                 rc = PTR_ERR(o->od_quota_slave);
7409                 o->od_quota_slave = NULL;
7410                 GOTO(out_procfs, rc);
7411         }
7412
7413         RETURN(0);
7414
7415 out_procfs:
7416         osd_procfs_fini(o);
7417 out_scrub:
7418         osd_scrub_cleanup(env, o);
7419 out_site:
7420         lu_site_fini(&o->od_site);
7421 out_compat:
7422         osd_obj_map_fini(o);
7423 out_mnt:
7424         osd_umount(env, o);
7425 out:
7426         return rc;
7427 }
7428
7429 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7430                                           struct lu_device_type *t,
7431                                           struct lustre_cfg *cfg)
7432 {
7433         struct osd_device *o;
7434         int                rc;
7435
7436         OBD_ALLOC_PTR(o);
7437         if (o == NULL)
7438                 return ERR_PTR(-ENOMEM);
7439
7440         rc = dt_device_init(&o->od_dt_dev, t);
7441         if (rc == 0) {
7442                 /* Because the ctx might be revived in dt_device_init,
7443                  * refill the env here */
7444                 lu_env_refill((struct lu_env *)env);
7445                 rc = osd_device_init0(env, o, cfg);
7446                 if (rc)
7447                         dt_device_fini(&o->od_dt_dev);
7448         }
7449
7450         if (unlikely(rc != 0))
7451                 OBD_FREE_PTR(o);
7452
7453         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7454 }
7455
7456 static struct lu_device *osd_device_free(const struct lu_env *env,
7457                                          struct lu_device *d)
7458 {
7459         struct osd_device *o = osd_dev(d);
7460         ENTRY;
7461
7462         /* XXX: make osd top device in order to release reference */
7463         d->ld_site->ls_top_dev = d;
7464         lu_site_purge(env, d->ld_site, -1);
7465         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7466                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7467                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7468         }
7469         lu_site_fini(&o->od_site);
7470         dt_device_fini(&o->od_dt_dev);
7471         OBD_FREE_PTR(o);
7472         RETURN(NULL);
7473 }
7474
7475 static int osd_process_config(const struct lu_env *env,
7476                               struct lu_device *d, struct lustre_cfg *cfg)
7477 {
7478         struct osd_device               *o = osd_dev(d);
7479         int                             rc;
7480         ENTRY;
7481
7482         switch (cfg->lcfg_command) {
7483         case LCFG_SETUP:
7484                 rc = osd_mount(env, o, cfg);
7485                 break;
7486         case LCFG_CLEANUP:
7487                 /* For the case LCFG_PRE_CLEANUP is not called in advance,
7488                  * that may happend if hit failure during mount process. */
7489                 osd_index_backup(env, o, false);
7490                 lu_dev_del_linkage(d->ld_site, d);
7491                 rc = osd_shutdown(env, o);
7492                 break;
7493         case LCFG_PARAM:
7494                 LASSERT(&o->od_dt_dev);
7495                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7496                                               cfg, &o->od_dt_dev);
7497                 if (rc > 0 || rc == -ENOSYS) {
7498                         rc = class_process_proc_param(PARAM_OST,
7499                                                       lprocfs_osd_obd_vars,
7500                                                       cfg, &o->od_dt_dev);
7501                         if (rc > 0)
7502                                 rc = 0;
7503                 }
7504                 break;
7505         case LCFG_PRE_CLEANUP:
7506                 osd_scrub_stop(o);
7507                 osd_index_backup(env, o,
7508                                  o->od_index_backup_policy != LIBP_NONE);
7509                 rc = 0;
7510                 break;
7511         default:
7512                 rc = -ENOSYS;
7513         }
7514
7515         RETURN(rc);
7516 }
7517
7518 static int osd_recovery_complete(const struct lu_env *env,
7519                                  struct lu_device *d)
7520 {
7521         struct osd_device       *osd = osd_dev(d);
7522         int                      rc = 0;
7523         ENTRY;
7524
7525         if (osd->od_quota_slave == NULL)
7526                 RETURN(0);
7527
7528         /* start qsd instance on recovery completion, this notifies the quota
7529          * slave code that we are about to process new requests now */
7530         rc = qsd_start(env, osd->od_quota_slave);
7531         RETURN(rc);
7532 }
7533
7534 /*
7535  * we use exports to track all osd users
7536  */
7537 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7538                            struct obd_device *obd, struct obd_uuid *cluuid,
7539                            struct obd_connect_data *data, void *localdata)
7540 {
7541         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7542         struct lustre_handle  conn;
7543         int                   rc;
7544         ENTRY;
7545
7546         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7547
7548         rc = class_connect(&conn, obd, cluuid);
7549         if (rc)
7550                 RETURN(rc);
7551
7552         *exp = class_conn2export(&conn);
7553
7554         spin_lock(&osd->od_osfs_lock);
7555         osd->od_connects++;
7556         spin_unlock(&osd->od_osfs_lock);
7557
7558         RETURN(0);
7559 }
7560
7561 /*
7562  * once last export (we don't count self-export) disappeared
7563  * osd can be released
7564  */
7565 static int osd_obd_disconnect(struct obd_export *exp)
7566 {
7567         struct obd_device *obd = exp->exp_obd;
7568         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7569         int                rc, release = 0;
7570         ENTRY;
7571
7572         /* Only disconnect the underlying layers on the final disconnect. */
7573         spin_lock(&osd->od_osfs_lock);
7574         osd->od_connects--;
7575         if (osd->od_connects == 0)
7576                 release = 1;
7577         spin_unlock(&osd->od_osfs_lock);
7578
7579         rc = class_disconnect(exp); /* bz 9811 */
7580
7581         if (rc == 0 && release)
7582                 class_manual_cleanup(obd);
7583         RETURN(rc);
7584 }
7585
7586 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7587                        struct lu_device *dev)
7588 {
7589         struct osd_device       *osd    = osd_dev(dev);
7590         struct lr_server_data   *lsd    =
7591                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7592         int                      result = 0;
7593         ENTRY;
7594
7595         if (osd->od_quota_slave != NULL) {
7596                 /* set up quota slave objects */
7597                 result = qsd_prepare(env, osd->od_quota_slave);
7598                 if (result != 0)
7599                         RETURN(result);
7600         }
7601
7602         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7603 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7604                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7605                         osd->od_index_in_idif = 1;
7606                 } else {
7607                         osd->od_index_in_idif = 0;
7608                         result = osd_register_proc_index_in_idif(osd);
7609                         if (result != 0)
7610                                 RETURN(result);
7611                 }
7612 #else
7613                 osd->od_index_in_idif = 1;
7614 #endif
7615         }
7616
7617         result = osd_fid_init(env, osd);
7618
7619         RETURN(result);
7620 }
7621
7622 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7623                          struct lu_fid *fid, struct md_op_data *op_data)
7624 {
7625         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7626
7627         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7628 }
7629
7630 static const struct lu_object_operations osd_lu_obj_ops = {
7631         .loo_object_init      = osd_object_init,
7632         .loo_object_delete    = osd_object_delete,
7633         .loo_object_release   = osd_object_release,
7634         .loo_object_free      = osd_object_free,
7635         .loo_object_print     = osd_object_print,
7636         .loo_object_invariant = osd_object_invariant
7637 };
7638
7639 const struct lu_device_operations osd_lu_ops = {
7640         .ldo_object_alloc      = osd_object_alloc,
7641         .ldo_process_config    = osd_process_config,
7642         .ldo_recovery_complete = osd_recovery_complete,
7643         .ldo_prepare           = osd_prepare,
7644 };
7645
7646 static const struct lu_device_type_operations osd_device_type_ops = {
7647         .ldto_init = osd_type_init,
7648         .ldto_fini = osd_type_fini,
7649
7650         .ldto_start = osd_type_start,
7651         .ldto_stop  = osd_type_stop,
7652
7653         .ldto_device_alloc = osd_device_alloc,
7654         .ldto_device_free  = osd_device_free,
7655
7656         .ldto_device_init    = osd_device_init,
7657         .ldto_device_fini    = osd_device_fini
7658 };
7659
7660 static struct lu_device_type osd_device_type = {
7661         .ldt_tags     = LU_DEVICE_DT,
7662         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7663         .ldt_ops      = &osd_device_type_ops,
7664         .ldt_ctx_tags = LCT_LOCAL,
7665 };
7666
7667 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7668 {
7669         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7670         struct super_block *sb = osd_sb(osd);
7671
7672         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7673 }
7674
7675 /*
7676  * lprocfs legacy support.
7677  */
7678 static struct obd_ops osd_obd_device_ops = {
7679         .o_owner = THIS_MODULE,
7680         .o_connect      = osd_obd_connect,
7681         .o_disconnect   = osd_obd_disconnect,
7682         .o_fid_alloc    = osd_fid_alloc,
7683         .o_health_check = osd_health_check,
7684 };
7685
7686 static int __init osd_init(void)
7687 {
7688         int rc;
7689
7690         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7691 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7692         /* please, try to keep osd_thread_info smaller than a page */
7693         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7694 #endif
7695
7696         osd_oi_mod_init();
7697
7698         rc = lu_kmem_init(ldiskfs_caches);
7699         if (rc)
7700                 return rc;
7701
7702 #ifdef CONFIG_KALLSYMS
7703         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7704         priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7705         /* Clear readonly is unused at this time */
7706         /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7707 #endif
7708
7709         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7710                                  lprocfs_osd_module_vars,
7711                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7712         if (rc)
7713                 lu_kmem_fini(ldiskfs_caches);
7714         return rc;
7715 }
7716
7717 static void __exit osd_exit(void)
7718 {
7719         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7720         lu_kmem_fini(ldiskfs_caches);
7721 }
7722
7723 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7724 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7725 MODULE_VERSION(LUSTRE_VERSION_STRING);
7726 MODULE_LICENSE("GPL");
7727
7728 module_init(osd_init);
7729 module_exit(osd_exit);