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