lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 499                               "maximum entry limit\n",
 500                               osd_name(osd), parent->i_ino, PFID(fid),
 501                               rc == -ENOSPC ? "has reached" : "is approaching");
 502                 else
 503                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 504                               "maximum entry limit\n",
 505                               osd_name(osd), parent->i_ino,
 506                               rc == -ENOSPC ? "has reached" : "is approaching");
 507
 508                 /* ignore such error now */
 509                 if (rc == -ENOBUFS)
 510                         rc = 0;
 511         }
 512
 513         return rc;
 514 }
 515
 516
 517 struct inode *
 518 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 519              struct osd_inode_id *id, struct lu_fid *fid)
 520 {
 521         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 522         struct inode *inode;
 523         int rc;
 524
 525         inode = osd_iget(info, dev, id);
 526         if (IS_ERR(inode))
 527                 return inode;
 528
 529         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 530         if (!rc) {
 531                 *fid = loa->loa_lma.lma_self_fid;
 532         } else if (rc == -ENODATA) {
 533                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 534                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 535                 else
 536                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 537         } else {
 538                 iput(inode);
 539                 inode = ERR_PTR(rc);
 540         }
 541         return inode;
 542 }
 543
 544 static struct inode *osd_iget_check(struct osd_thread_info *info,
 545                                     struct osd_device *dev,
 546                                     const struct lu_fid *fid,
 547                                     struct osd_inode_id *id,
 548                                     bool trusted)
 549 {
 550         struct inode *inode;
 551         int rc = 0;
 552
 553         ENTRY;
 554
 555         /*
 556          * The cached OI mapping is trustable. If we cannot locate the inode
 557          * via the cached OI mapping, then return the failure to the caller
 558          * directly without further OI checking.
 559          */
 560
 561 again:
 562         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_18_23, contains self-fid
 711  * \retval -v: other failure cases
 712  */
 713 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 714                  struct dentry *dentry, struct lu_fid *fid)
 715 {
 716         struct filter_fid *ff = &info->oti_ff;
 717         struct ost_id *ostid = &info->oti_ostid;
 718         int rc;
 719
 720         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 721         if (rc == sizeof(struct filter_fid_18_23)) {
 722                 struct filter_fid_18_23 *ff_old = (void *)ff;
 723
 724                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 725                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 726                 /*
 727                  * XXX: use 0 as the index for compatibility, the caller will
 728                  * handle index related issues when necessary.
 729                  */
 730                 if (!rc)
 731                         ostid_to_fid(fid, ostid, 0);
 732         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 733                 rc = 1;
 734         } else if (rc >= 0) {
 735                 rc = -EINVAL;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int osd_lma_self_repair(struct osd_thread_info *info,
 742                                struct osd_device *osd, struct inode *inode,
 743                                const struct lu_fid *fid, __u32 compat)
 744 {
 745         handle_t *jh;
 746         int rc;
 747
 748         LASSERT(current->journal_info == NULL);
 749
 750         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 751                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 752         if (IS_ERR(jh)) {
 753                 rc = PTR_ERR(jh);
 754                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 755                       osd_name(osd), rc);
 756                 return rc;
 757         }
 758
 759         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 760         if (rc != 0)
 761                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 762                       osd_name(osd), rc);
 763         ldiskfs_journal_stop(jh);
 764         return rc;
 765 }
 766
 767 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 768 {
 769         struct osd_thread_info *info = osd_oti_get(env);
 770         struct osd_device *osd = osd_obj2dev(obj);
 771         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 772         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 773         struct inode *inode = obj->oo_inode;
 774         struct dentry *dentry = &info->oti_obj_dentry;
 775         struct lu_fid *fid = NULL;
 776         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 777         int rc;
 778
 779         ENTRY;
 780
 781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 782                              (void *)loa, sizeof(*loa));
 783         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 784                 fid = &lma->lma_self_fid;
 785                 rc = osd_get_idif(info, inode, dentry, fid);
 786                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 787                         /*
 788                          * For the given OST-object, if it has neither LMA nor
 789                          * FID in XATTR_NAME_FID, then the given FID (which is
 790                          * contained in the @obj, from client RPC for locating
 791                          * the OST-object) is trusted. We use it to generate
 792                          * the LMA.
 793                          */
 794                         osd_lma_self_repair(info, osd, inode, rfid,
 795                                             LMAC_FID_ON_OST);
 796                         RETURN(0);
 797                 }
 798         }
 799
 800         if (rc < 0)
 801                 RETURN(rc);
 802
 803         if (rc > 0) {
 804                 rc = 0;
 805                 lustre_lma_swab(lma);
 806                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 807                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 808                               S_ISREG(inode->i_mode)))) {
 809                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 810                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 811                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 812                               PFID(rfid), inode->i_ino);
 813                         rc = -EOPNOTSUPP;
 814                 } else {
 815                         fid = &lma->lma_self_fid;
 816                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 817                             osd->od_is_ost)
 818                                 obj->oo_pfid_in_lma = 1;
 819                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 820                             !osd->od_is_ost)
 821                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 822                 }
 823         }
 824
 825         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 826                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 827                         struct ost_id   *oi   = &info->oti_ostid;
 828                         struct lu_fid   *fid1 = &info->oti_fid3;
 829                         __u32            idx  = fid_idif_ost_idx(rfid);
 830
 831                         /*
 832                          * For old IDIF, the OST index is not part of the IDIF,
 833                          * Means that different OSTs may have the same IDIFs.
 834                          * Under such case, we need to make some compatible
 835                          * check to make sure to trigger OI scrub properly.
 836                          */
 837                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 838                                 /* Given @rfid is new, LMA is old. */
 839                                 fid_to_ostid(fid, oi);
 840                                 ostid_to_fid(fid1, oi, idx);
 841                                 if (lu_fid_eq(fid1, rfid)) {
 842                                         if (osd->od_index_in_idif)
 843                                                 osd_lma_self_repair(info, osd,
 844                                                         inode, rfid,
 845                                                         LMAC_FID_ON_OST);
 846                                         RETURN(0);
 847                                 }
 848                         }
 849                 }
 850
 851                 rc = -EREMCHG;
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 struct osd_check_lmv_buf {
 858 #ifdef HAVE_DIR_CONTEXT
 859         /* please keep it as first member */
 860         struct dir_context ctx;
 861 #endif
 862         struct osd_thread_info *oclb_info;
 863         struct osd_device *oclb_dev;
 864         struct osd_idmap_cache *oclb_oic;
 865         int oclb_items;
 866         bool oclb_found;
 867 };
 868
 869 /**
 870  * It is called internally by ->readdir() to filter out the
 871  * local slave object's FID of the striped directory.
 872  *
 873  * \retval      1 found the local slave's FID
 874  * \retval      0 continue to check next item
 875  * \retval      -ve for failure
 876  */
 877 #ifdef HAVE_FILLDIR_USE_CTX
 878 static int osd_stripe_dir_filldir(struct dir_context *buf,
 879 #else
 880 static int osd_stripe_dir_filldir(void *buf,
 881 #endif
 882                                   const char *name, int namelen,
 883                                   loff_t offset, __u64 ino, unsigned int d_type)
 884 {
 885         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 886         struct osd_thread_info *oti = oclb->oclb_info;
 887         struct lu_fid *fid = &oti->oti_fid3;
 888         struct osd_inode_id *id = &oti->oti_id3;
 889         struct osd_device *dev = oclb->oclb_dev;
 890         struct osd_idmap_cache *oic = oclb->oclb_oic;
 891         struct inode *inode;
 892
 893         oclb->oclb_items++;
 894
 895         if (name[0] == '.')
 896                 return 0;
 897
 898         fid_zero(fid);
 899         sscanf(name + 1, SFID, RFID(fid));
 900         if (!fid_is_sane(fid))
 901                 return 0;
 902
 903         if (osd_remote_fid(oti->oti_env, dev, fid))
 904                 return 0;
 905
 906         osd_id_gen(id, ino, OSD_OII_NOGEN);
 907         inode = osd_iget(oti, dev, id);
 908         if (IS_ERR(inode))
 909                 return PTR_ERR(inode);
 910
 911         iput(inode);
 912         osd_add_oi_cache(oti, dev, id, fid);
 913         oic->oic_fid = *fid;
 914         oic->oic_lid = *id;
 915         oic->oic_dev = dev;
 916         osd_oii_insert(dev, oic, true);
 917         oclb->oclb_found = true;
 918
 919         return 1;
 920 }
 921
 922 /*
 923  * When lookup item under striped directory, we need to locate the master
 924  * MDT-object of the striped directory firstly, then the client will send
 925  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 926  * and the item's name. If the system is restored from MDT file level backup,
 927  * then before the OI scrub completely built the OI files, the OI mappings of
 928  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 929  * not a problem for the master MDT-object. Because when locate the master
 930  * MDT-object, we will do name based lookup (for the striped directory itself)
 931  * firstly, during such process we can setup the correct OI mapping for the
 932  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 933  * the client will not trigger name based lookup on the MDT to locate the slave
 934  * MDT-object before locating item under the striped directory, then when
 935  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 936  * is invalid and does not know what the right OI mapping is, then the MDT has
 937  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 938  * the OI file, related OI mapping is unknown yet, please try again later. And
 939  * then client will re-try the RPC again and again until related OI mapping has
 940  * been updated. That is quite inefficient.
 941  *
 942  * To resolve above trouble, we will handle it as the following two cases:
 943  *
 944  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 945  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 946  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 947  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 948  *    directly. Please check osd_fid_lookup().
 949  *
 950  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 951  *    Under such case, during lookup the master MDT-object, we will lookup the
 952  *    slave MDT-object via readdir against the master MDT-object, because the
 953  *    slave MDT-objects information are stored as sub-directories with the name
 954  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 955  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 956  *    the correct OI mapping for the slave MDT-object.
 957  */
 958 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 959                          struct inode *inode, struct osd_idmap_cache *oic)
 960 {
 961         struct lu_buf *buf = &oti->oti_big_buf;
 962         struct dentry *dentry = &oti->oti_obj_dentry;
 963         struct file *filp = &oti->oti_file;
 964         const struct file_operations *fops;
 965         struct lmv_mds_md_v1 *lmv1;
 966         struct osd_check_lmv_buf oclb = {
 967 #ifdef HAVE_DIR_CONTEXT
 968                 .ctx.actor = osd_stripe_dir_filldir,
 969 #endif
 970                 .oclb_info = oti,
 971                 .oclb_dev = dev,
 972                 .oclb_oic = oic,
 973                 .oclb_found = false,
 974         };
 975         int rc = 0;
 976
 977         ENTRY;
 978
 979 again:
 980         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 981                              buf->lb_len);
 982         if (rc == -ERANGE) {
 983                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 984                 if (rc > 0) {
 985                         lu_buf_realloc(buf, rc);
 986                         if (buf->lb_buf == NULL)
 987                                 GOTO(out, rc = -ENOMEM);
 988
 989                         goto again;
 990                 }
 991         }
 992
 993         if (unlikely(rc == 0 || rc == -ENODATA))
 994                 GOTO(out, rc = 0);
 995
 996         if (rc < 0)
 997                 GOTO(out, rc);
 998
 999         if (unlikely(buf->lb_buf == NULL)) {
1000                 lu_buf_realloc(buf, rc);
1001                 if (buf->lb_buf == NULL)
1002                         GOTO(out, rc = -ENOMEM);
1003
1004                 goto again;
1005         }
1006
1007         lmv1 = buf->lb_buf;
1008         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1009                 GOTO(out, rc = 0);
1010
1011         fops = inode->i_fop;
1012         dentry->d_inode = inode;
1013         dentry->d_sb = inode->i_sb;
1014         filp->f_pos = 0;
1015         filp->f_path.dentry = dentry;
1016         filp->f_mode = FMODE_64BITHASH;
1017         filp->f_mapping = inode->i_mapping;
1018         filp->f_op = fops;
1019         filp->private_data = NULL;
1020         set_file_inode(filp, inode);
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024 #ifdef HAVE_DIR_CONTEXT
1025                 oclb.ctx.pos = filp->f_pos;
1026 #ifdef HAVE_ITERATE_SHARED
1027                 rc = fops->iterate_shared(filp, &oclb.ctx);
1028 #else
1029                 rc = fops->iterate(filp, &oclb.ctx);
1030 #endif
1031                 filp->f_pos = oclb.ctx.pos;
1032 #else
1033                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1034 #endif
1035         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1036                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1037         fops->release(inode, filp);
1038
1039 out:
1040         if (rc < 0)
1041                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1042                        DFID": rc = %d\n", osd_ino2name(inode),
1043                        inode->i_ino, inode->i_generation,
1044                        PFID(&oic->oic_fid), rc);
1045         else
1046                 rc = 0;
1047
1048         RETURN(rc);
1049 }
1050
1051 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1052                           const struct lu_fid *fid,
1053                           const struct lu_object_conf *conf)
1054 {
1055         struct osd_thread_info *info;
1056         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1057         struct osd_device *dev;
1058         struct osd_idmap_cache *oic;
1059         struct osd_inode_id *id;
1060         struct inode *inode = NULL;
1061         struct lustre_scrub *scrub;
1062         struct scrub_file *sf;
1063         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1064         __u32 saved_ino;
1065         __u32 saved_gen;
1066         int result = 0;
1067         int rc1 = 0;
1068         bool remote = false;
1069         bool trusted = true;
1070         bool updated = false;
1071         bool checked = false;
1072
1073         ENTRY;
1074
1075         LINVRNT(osd_invariant(obj));
1076         LASSERT(obj->oo_inode == NULL);
1077         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1078
1079         dev = osd_dev(ldev);
1080         scrub = &dev->od_scrub.os_scrub;
1081         sf = &scrub->os_file;
1082         info = osd_oti_get(env);
1083         LASSERT(info);
1084         oic = &info->oti_cache;
1085
1086         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1087                 RETURN(-ENOENT);
1088
1089         /*
1090          * For the object is created as locking anchor, or for the object to
1091          * be created on disk. No need to osd_oi_lookup() at here because FID
1092          * shouldn't never be re-used, if it's really a duplicate FID from
1093          * unexpected reason, we should be able to detect it later by calling
1094          * do_create->osd_oi_insert().
1095          */
1096         if (conf && conf->loc_flags & LOC_F_NEW)
1097                 GOTO(out, result = 0);
1098
1099         /* Search order: 1. per-thread cache. */
1100         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1101                 id = &oic->oic_lid;
1102                 goto iget;
1103         }
1104
1105         id = &info->oti_id;
1106         if (!list_empty(&scrub->os_inconsistent_items)) {
1107                 /* Search order: 2. OI scrub pending list. */
1108                 result = osd_oii_lookup(dev, fid, id);
1109                 if (!result)
1110                         goto iget;
1111         }
1112
1113         /*
1114          * The OI mapping in the OI file can be updated by the OI scrub
1115          * when we locate the inode via FID. So it may be not trustable.
1116          */
1117         trusted = false;
1118
1119         /* Search order: 3. OI files. */
1120         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1121         if (result == -ENOENT) {
1122                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1123                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1124                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1125                                       sf->sf_oi_bitmap))
1126                         GOTO(out, result = 0);
1127
1128                 goto trigger;
1129         }
1130
1131         if (result)
1132                 GOTO(out, result);
1133
1134 iget:
1135         obj->oo_inode = NULL;
1136         /* for later passes through checks, not true on first pass */
1137         if (!IS_ERR_OR_NULL(inode))
1138                 iput(inode);
1139
1140         inode = osd_iget_check(info, dev, fid, id, trusted);
1141         if (!IS_ERR(inode)) {
1142                 obj->oo_inode = inode;
1143                 result = 0;
1144                 if (remote)
1145                         goto trigger;
1146
1147                 goto check_lma;
1148         }
1149
1150         result = PTR_ERR(inode);
1151         if (result == -ENOENT || result == -ESTALE)
1152                 GOTO(out, result = 0);
1153
1154         if (result != -EREMCHG)
1155                 GOTO(out, result);
1156
1157 trigger:
1158         /*
1159          * We still have chance to get the valid inode: for the
1160          * object which is referenced by remote name entry, the
1161          * object on the local MDT will be linked under the dir
1162          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1163          *
1164          * We do not know whether the object for the given FID
1165          * is referenced by some remote name entry or not, and
1166          * especially for DNE II, a multiple-linked object may
1167          * have many name entries reside on many MDTs.
1168          *
1169          * To simplify the operation, OSD will not distinguish
1170          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1171          * only happened for the RPC from other MDT during the
1172          * OI scrub, or for the client side RPC with FID only,
1173          * such as FID to path, or from old connected client.
1174          */
1175         if (!remote) {
1176                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1177                 if (!rc1) {
1178                         remote = true;
1179                         trusted = true;
1180                         flags |= SS_AUTO_PARTIAL;
1181                         flags &= ~SS_AUTO_FULL;
1182                         goto iget;
1183                 }
1184         }
1185
1186         if (thread_is_running(&scrub->os_thread)) {
1187                 if (scrub->os_partial_scan && !scrub->os_in_join)
1188                         goto join;
1189
1190                 if (IS_ERR_OR_NULL(inode) || result)
1191                         GOTO(out, result = -EINPROGRESS);
1192
1193                 LASSERT(remote);
1194                 LASSERT(obj->oo_inode == inode);
1195
1196                 osd_add_oi_cache(info, dev, id, fid);
1197                 osd_oii_insert(dev, oic, true);
1198                 goto found;
1199         }
1200
1201         if (dev->od_auto_scrub_interval == AS_NEVER) {
1202                 if (!remote)
1203                         GOTO(out, result = -EREMCHG);
1204
1205                 LASSERT(!result);
1206                 LASSERT(obj->oo_inode == inode);
1207
1208                 osd_add_oi_cache(info, dev, id, fid);
1209                 goto found;
1210         }
1211
1212 join:
1213         rc1 = osd_scrub_start(env, dev, flags);
1214         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1215                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1216         if (rc1 && rc1 != -EALREADY)
1217                 GOTO(out, result = -EREMCHG);
1218
1219         if (IS_ERR_OR_NULL(inode) || result)
1220                 GOTO(out, result = -EINPROGRESS);
1221
1222         LASSERT(remote);
1223         LASSERT(obj->oo_inode == inode);
1224
1225         osd_add_oi_cache(info, dev, id, fid);
1226         osd_oii_insert(dev, oic, true);
1227         goto found;
1228
1229 check_lma:
1230         checked = true;
1231         if (unlikely(obj->oo_header))
1232                 goto found;
1233
1234         result = osd_check_lma(env, obj);
1235         if (!result)
1236                 goto found;
1237
1238         LASSERTF(id->oii_ino == inode->i_ino &&
1239                  id->oii_gen == inode->i_generation,
1240                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1241                  PFID(fid), id->oii_ino, id->oii_gen,
1242                  inode->i_ino, inode->i_generation);
1243
1244         saved_ino = inode->i_ino;
1245         saved_gen = inode->i_generation;
1246
1247         if (unlikely(result == -ENODATA)) {
1248                 /*
1249                  * If the OI scrub updated the OI mapping by race, it
1250                  * must be valid. Trust the inode that has no LMA EA.
1251                  */
1252                 if (updated)
1253                         goto found;
1254
1255                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1256                 if (!result) {
1257                         /*
1258                          * The OI mapping is still there, the inode is still
1259                          * valid. It is just becaues the inode has no LMA EA.
1260                          */
1261                         if (saved_ino == id->oii_ino &&
1262                             saved_gen == id->oii_gen)
1263                                 goto found;
1264
1265                         /*
1266                          * It is the OI scrub updated the OI mapping by race.
1267                          * The new OI mapping must be valid.
1268                          */
1269                         trusted = true;
1270                         updated = true;
1271                         goto iget;
1272                 }
1273
1274                 /*
1275                  * "result == -ENOENT" means that the OI mappinghas been
1276                  * removed by race, so the inode belongs to other object.
1277                  *
1278                  * Others error can be returned  directly.
1279                  */
1280                 if (result == -ENOENT) {
1281                         LASSERT(trusted);
1282
1283                         obj->oo_inode = NULL;
1284                         result = 0;
1285                 }
1286         }
1287
1288         if (result != -EREMCHG)
1289                 GOTO(out, result);
1290
1291         LASSERT(!updated);
1292
1293         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1294         /*
1295          * "result == -ENOENT" means the cached OI mapping has been removed
1296          * from the OI file by race, above inode belongs to other object.
1297          */
1298         if (result == -ENOENT) {
1299                 LASSERT(trusted);
1300
1301                 obj->oo_inode = NULL;
1302                 GOTO(out, result = 0);
1303         }
1304
1305         if (result)
1306                 GOTO(out, result);
1307
1308         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1309                 result = -EREMCHG;
1310                 goto trigger;
1311         }
1312
1313         /*
1314          * It is the OI scrub updated the OI mapping by race.
1315          * The new OI mapping must be valid.
1316          */
1317         trusted = true;
1318         updated = true;
1319         goto iget;
1320
1321 found:
1322         if (!checked) {
1323                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1324                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1325
1326                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1327                 if (!result) {
1328                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1329                             dev->od_is_ost)
1330                                 obj->oo_pfid_in_lma = 1;
1331                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1332                             !dev->od_is_ost)
1333                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1334                 } else if (result != -ENODATA) {
1335                         GOTO(out, result);
1336                 }
1337         }
1338
1339         obj->oo_compat_dot_created = 1;
1340         obj->oo_compat_dotdot_created = 1;
1341
1342         if (S_ISDIR(inode->i_mode) &&
1343             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1344                 osd_check_lmv(info, dev, inode, oic);
1345
1346         result = osd_attach_jinode(inode);
1347         if (result)
1348                 GOTO(out, result);
1349
1350         if (!ldiskfs_pdo)
1351                 GOTO(out, result = 0);
1352
1353         LASSERT(!obj->oo_hl_head);
1354         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1355
1356         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1357
1358 out:
1359         if (result || !obj->oo_inode) {
1360                 if (!IS_ERR_OR_NULL(inode))
1361                         iput(inode);
1362
1363                 obj->oo_inode = NULL;
1364                 if (trusted)
1365                         fid_zero(&oic->oic_fid);
1366         }
1367
1368         LINVRNT(osd_invariant(obj));
1369         return result;
1370 }
1371
1372 /*
1373  * Concurrency: shouldn't matter.
1374  */
1375 static void osd_object_init0(struct osd_object *obj)
1376 {
1377         LASSERT(obj->oo_inode != NULL);
1378         obj->oo_dt.do_body_ops = &osd_body_ops;
1379         obj->oo_dt.do_lu.lo_header->loh_attr |=
1380                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1381 }
1382
1383 /*
1384  * Concurrency: no concurrent access is possible that early in object
1385  * life-cycle.
1386  */
1387 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1388                            const struct lu_object_conf *conf)
1389 {
1390         struct osd_object *obj = osd_obj(l);
1391         int result;
1392
1393         LINVRNT(osd_invariant(obj));
1394
1395         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1396                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1397                 l->lo_header->loh_attr |= LOHA_EXISTS;
1398                 return 0;
1399         }
1400
1401         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1402         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1403         if (result == 0 && obj->oo_inode != NULL) {
1404                 struct osd_thread_info *oti = osd_oti_get(env);
1405                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1406
1407                 osd_object_init0(obj);
1408                 if (unlikely(obj->oo_header))
1409                         return 0;
1410
1411                 result = osd_get_lma(oti, obj->oo_inode,
1412                                      &oti->oti_obj_dentry, loa);
1413                 if (!result) {
1414                         /*
1415                          * Convert LMAI flags to lustre LMA flags
1416                          * and cache it to oo_lma_flags
1417                          */
1418                         obj->oo_lma_flags =
1419                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1420                 } else if (result == -ENODATA) {
1421                         result = 0;
1422                 }
1423         }
1424
1425         LINVRNT(osd_invariant(obj));
1426         return result;
1427 }
1428
1429 /*
1430  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1431  * The left part is for value, binary mode.
1432  */
1433 struct osd_xattr_entry {
1434         struct list_head        oxe_list;
1435         size_t                  oxe_len;
1436         size_t                  oxe_namelen;
1437         bool                    oxe_exist;
1438         struct rcu_head         oxe_rcu;
1439         char                    oxe_buf[0];
1440 };
1441
1442 static int osd_oxc_get(struct osd_object *obj, const char *name,
1443                        struct lu_buf *buf)
1444 {
1445         struct osd_xattr_entry *tmp;
1446         struct osd_xattr_entry *oxe = NULL;
1447         size_t namelen = strlen(name);
1448         int rc;
1449
1450         ENTRY;
1451
1452         rcu_read_lock();
1453         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1454                 if (namelen == tmp->oxe_namelen &&
1455                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1456                         oxe = tmp;
1457                         break;
1458                 }
1459         }
1460
1461         if (oxe == NULL)
1462                 GOTO(out, rc = -ENOENT);
1463
1464         if (!oxe->oxe_exist)
1465                 GOTO(out, rc = -ENODATA);
1466
1467         /* vallen */
1468         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1469         LASSERT(rc > 0);
1470
1471         if (buf->lb_buf == NULL)
1472                 GOTO(out, rc);
1473
1474         if (buf->lb_len < rc)
1475                 GOTO(out, rc = -ERANGE);
1476
1477         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1478         EXIT;
1479 out:
1480         rcu_read_unlock();
1481
1482         return rc;
1483 }
1484
1485 static void osd_oxc_free(struct rcu_head *head)
1486 {
1487         struct osd_xattr_entry *oxe;
1488
1489         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1490         OBD_FREE(oxe, oxe->oxe_len);
1491 }
1492
1493 static void osd_oxc_add(struct osd_object *obj, const char *name,
1494                         const char *buf, int buflen)
1495 {
1496         struct osd_xattr_entry *oxe;
1497         struct osd_xattr_entry *old = NULL;
1498         struct osd_xattr_entry *tmp;
1499         size_t namelen = strlen(name);
1500         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1501
1502         OBD_ALLOC(oxe, len);
1503         if (oxe == NULL)
1504                 return;
1505
1506         INIT_LIST_HEAD(&oxe->oxe_list);
1507         oxe->oxe_len = len;
1508         oxe->oxe_namelen = namelen;
1509         memcpy(oxe->oxe_buf, name, namelen);
1510         if (buflen > 0) {
1511                 LASSERT(buf != NULL);
1512                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1513                 oxe->oxe_exist = true;
1514         } else {
1515                 oxe->oxe_exist = false;
1516         }
1517
1518         /* this should be rarely called, just remove old and add new */
1519         spin_lock(&obj->oo_guard);
1520         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1521                 if (namelen == tmp->oxe_namelen &&
1522                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1523                         old = tmp;
1524                         break;
1525                 }
1526         }
1527         if (old != NULL) {
1528                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1529                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1530         } else {
1531                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1532         }
1533         spin_unlock(&obj->oo_guard);
1534 }
1535
1536 static void osd_oxc_del(struct osd_object *obj, const char *name)
1537 {
1538         struct osd_xattr_entry *oxe;
1539         size_t namelen = strlen(name);
1540
1541         spin_lock(&obj->oo_guard);
1542         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1543                 if (namelen == oxe->oxe_namelen &&
1544                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1545                         list_del_rcu(&oxe->oxe_list);
1546                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1547                         break;
1548                 }
1549         }
1550         spin_unlock(&obj->oo_guard);
1551 }
1552
1553 static void osd_oxc_fini(struct osd_object *obj)
1554 {
1555         struct osd_xattr_entry *oxe, *next;
1556
1557         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1558                 list_del(&oxe->oxe_list);
1559                 OBD_FREE(oxe, oxe->oxe_len);
1560         }
1561 }
1562
1563 /*
1564  * Concurrency: no concurrent access is possible that late in object
1565  * life-cycle.
1566  */
1567 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1568 {
1569         struct osd_object *obj = osd_obj(l);
1570         struct lu_object_header *h = obj->oo_header;
1571
1572         LINVRNT(osd_invariant(obj));
1573
1574         osd_oxc_fini(obj);
1575         dt_object_fini(&obj->oo_dt);
1576         if (obj->oo_hl_head != NULL)
1577                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1578         OBD_FREE_PTR(obj);
1579         if (unlikely(h)) {
1580                 lu_object_header_fini(h);
1581                 OBD_FREE_PTR(h);
1582         }
1583 }
1584
1585 /*
1586  * Concurrency: no concurrent access is possible that late in object
1587  * life-cycle.
1588  */
1589 static void osd_index_fini(struct osd_object *o)
1590 {
1591         struct iam_container *bag;
1592
1593         if (o->oo_dir != NULL) {
1594                 bag = &o->oo_dir->od_container;
1595                 if (o->oo_inode != NULL) {
1596                         if (bag->ic_object == o->oo_inode)
1597                                 iam_container_fini(bag);
1598                 }
1599                 OBD_FREE_PTR(o->oo_dir);
1600                 o->oo_dir = NULL;
1601         }
1602 }
1603
1604 /*
1605  * Concurrency: no concurrent access is possible that late in object
1606  * life-cycle (for all existing callers, that is. New callers have to provide
1607  * their own locking.)
1608  */
1609 static int osd_inode_unlinked(const struct inode *inode)
1610 {
1611         return inode->i_nlink == 0;
1612 }
1613
1614 enum {
1615         OSD_TXN_OI_DELETE_CREDITS    = 20,
1616         OSD_TXN_INODE_DELETE_CREDITS = 20
1617 };
1618
1619 /*
1620  * Journal
1621  */
1622
1623 #if OSD_THANDLE_STATS
1624 /**
1625  * Set time when the handle is allocated
1626  */
1627 static void osd_th_alloced(struct osd_thandle *oth)
1628 {
1629         oth->oth_alloced = ktime_get();
1630 }
1631
1632 /**
1633  * Set time when the handle started
1634  */
1635 static void osd_th_started(struct osd_thandle *oth)
1636 {
1637         oth->oth_started = ktime_get();
1638 }
1639
1640 /**
1641  * Check whether the we deal with this handle for too long.
1642  */
1643 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1644                                 ktime_t alloced, ktime_t started,
1645                                 ktime_t closed)
1646 {
1647         ktime_t now = ktime_get();
1648
1649         LASSERT(dev != NULL);
1650
1651         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1652                             ktime_us_delta(started, alloced));
1653         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1654                             ktime_us_delta(closed, started));
1655         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1656                             ktime_us_delta(now, closed));
1657
1658         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1659                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1660                                 oth, now, alloced, started, closed);
1661                 libcfs_debug_dumpstack(NULL);
1662         }
1663 }
1664
1665 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1666 {                                                                       \
1667         ktime_t __closed = ktime_get();                                 \
1668         ktime_t __alloced = oth->oth_alloced;                           \
1669         ktime_t __started = oth->oth_started;                           \
1670                                                                         \
1671         expr;                                                           \
1672         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1673 }
1674
1675 #else /* OSD_THANDLE_STATS */
1676
1677 #define osd_th_alloced(h)                  do {} while(0)
1678 #define osd_th_started(h)                  do {} while(0)
1679 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1680
1681 #endif /* OSD_THANDLE_STATS */
1682
1683 /*
1684  * Concurrency: doesn't access mutable data.
1685  */
1686 static int osd_param_is_not_sane(const struct osd_device *dev,
1687                                  const struct thandle *th)
1688 {
1689         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1690
1691         return oh->ot_credits > osd_transaction_size(dev);
1692 }
1693
1694 /*
1695  * Concurrency: shouldn't matter.
1696  */
1697 static void osd_trans_commit_cb(struct super_block *sb,
1698                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1699 {
1700         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1701         struct thandle *th = &oh->ot_super;
1702         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1703         struct dt_txn_commit_cb *dcb, *tmp;
1704
1705         LASSERT(oh->ot_handle == NULL);
1706
1707         if (error)
1708                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1709
1710         dt_txn_hook_commit(th);
1711
1712         /* call per-transaction callbacks if any */
1713         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1714                                  dcb_linkage) {
1715                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1716                          "commit callback entry: magic=%x name='%s'\n",
1717                          dcb->dcb_magic, dcb->dcb_name);
1718                 list_del_init(&dcb->dcb_linkage);
1719                 dcb->dcb_func(NULL, th, dcb, error);
1720         }
1721
1722         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1723         lu_device_put(lud);
1724         th->th_dev = NULL;
1725
1726         OBD_FREE_PTR(oh);
1727 }
1728
1729 #ifndef HAVE_SB_START_WRITE
1730 # define sb_start_write(sb) do {} while (0)
1731 # define sb_end_write(sb) do {} while (0)
1732 #endif
1733
1734 static struct thandle *osd_trans_create(const struct lu_env *env,
1735                                         struct dt_device *d)
1736 {
1737         struct osd_thread_info *oti = osd_oti_get(env);
1738         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1739         struct osd_thandle *oh;
1740         struct thandle *th;
1741
1742         ENTRY;
1743
1744         if (d->dd_rdonly) {
1745                 CERROR("%s: someone try to start transaction under "
1746                        "readonly mode, should be disabled.\n",
1747                        osd_name(osd_dt_dev(d)));
1748                 dump_stack();
1749                 RETURN(ERR_PTR(-EROFS));
1750         }
1751
1752         /* on pending IO in this thread should left from prev. request */
1753         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1754
1755         sb_start_write(osd_sb(osd_dt_dev(d)));
1756
1757         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1758         if (!oh) {
1759                 sb_end_write(osd_sb(osd_dt_dev(d)));
1760                 RETURN(ERR_PTR(-ENOMEM));
1761         }
1762
1763         oh->ot_quota_trans = &oti->oti_quota_trans;
1764         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1765         th = &oh->ot_super;
1766         th->th_dev = d;
1767         th->th_result = 0;
1768         oh->ot_credits = 0;
1769         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1770         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1771         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1772         osd_th_alloced(oh);
1773
1774         memset(oti->oti_declare_ops, 0,
1775                sizeof(oti->oti_declare_ops));
1776         memset(oti->oti_declare_ops_cred, 0,
1777                sizeof(oti->oti_declare_ops_cred));
1778         memset(oti->oti_declare_ops_used, 0,
1779                sizeof(oti->oti_declare_ops_used));
1780
1781         oti->oti_ins_cache_depth++;
1782
1783         RETURN(th);
1784 }
1785
1786 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1787 {
1788         struct osd_thread_info *oti = osd_oti_get(env);
1789         struct osd_thandle *oh;
1790
1791         oh = container_of0(th, struct osd_thandle, ot_super);
1792         LASSERT(oh != NULL);
1793
1794         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1795               oti->oti_declare_ops[OSD_OT_CREATE],
1796               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1797               oti->oti_declare_ops_used[OSD_OT_CREATE],
1798               oti->oti_declare_ops[OSD_OT_DESTROY],
1799               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1800               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1801         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1802               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1803               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1805               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1806               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1807               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1808         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1809               oti->oti_declare_ops[OSD_OT_WRITE],
1810               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1811               oti->oti_declare_ops_used[OSD_OT_WRITE],
1812               oti->oti_declare_ops[OSD_OT_PUNCH],
1813               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1814               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1815               oti->oti_declare_ops[OSD_OT_QUOTA],
1816               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1817               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1818         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1819               oti->oti_declare_ops[OSD_OT_INSERT],
1820               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1821               oti->oti_declare_ops_used[OSD_OT_INSERT],
1822               oti->oti_declare_ops[OSD_OT_DELETE],
1823               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1824               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1825         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1826               oti->oti_declare_ops[OSD_OT_REF_ADD],
1827               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1829               oti->oti_declare_ops[OSD_OT_REF_DEL],
1830               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1831               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1832 }
1833
1834 /*
1835  * Concurrency: shouldn't matter.
1836  */
1837 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1838                            struct thandle *th)
1839 {
1840         struct osd_thread_info *oti = osd_oti_get(env);
1841         struct osd_device *dev = osd_dt_dev(d);
1842         handle_t *jh;
1843         struct osd_thandle *oh;
1844         int rc;
1845
1846         ENTRY;
1847
1848         LASSERT(current->journal_info == NULL);
1849
1850         oh = container_of0(th, struct osd_thandle, ot_super);
1851         LASSERT(oh != NULL);
1852         LASSERT(oh->ot_handle == NULL);
1853
1854         rc = dt_txn_hook_start(env, d, th);
1855         if (rc != 0)
1856                 GOTO(out, rc);
1857
1858         if (unlikely(osd_param_is_not_sane(dev, th))) {
1859                 static unsigned long last_printed;
1860                 static int last_credits;
1861
1862                 /*
1863                  * don't make noise on a tiny testing systems
1864                  * actual credits misuse will be caught anyway
1865                  */
1866                 if (last_credits != oh->ot_credits &&
1867                     time_after(jiffies, last_printed +
1868                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1869                     osd_transaction_size(dev) > 512) {
1870                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1871                               oh->ot_credits, osd_transaction_size(dev));
1872                         osd_trans_dump_creds(env, th);
1873                         libcfs_debug_dumpstack(NULL);
1874                         last_credits = oh->ot_credits;
1875                         last_printed = jiffies;
1876                 }
1877                 /*
1878                  * XXX Limit the credits to 'max_transaction_buffers', and
1879                  *     let the underlying filesystem to catch the error if
1880                  *     we really need so many credits.
1881                  *
1882                  *     This should be removed when we can calculate the
1883                  *     credits precisely.
1884                  */
1885                 oh->ot_credits = osd_transaction_size(dev);
1886         } else if (ldiskfs_track_declares_assert != 0) {
1887                 /*
1888                  * reserve few credits to prevent an assertion in JBD
1889                  * our debugging mechanism will be able to detected
1890                  * overuse. this can help to debug single-update
1891                  * transactions
1892                  */
1893                 oh->ot_credits += 10;
1894                 if (unlikely(osd_param_is_not_sane(dev, th)))
1895                         oh->ot_credits = osd_transaction_size(dev);
1896         }
1897
1898         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1899                 GOTO(out, rc = -EIO);
1900
1901         /*
1902          * XXX temporary stuff. Some abstraction layer should
1903          * be used.
1904          */
1905         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1906         osd_th_started(oh);
1907         if (!IS_ERR(jh)) {
1908                 oh->ot_handle = jh;
1909                 LASSERT(oti->oti_txns == 0);
1910
1911                 lu_device_get(&d->dd_lu_dev);
1912                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1913                               "osd-tx", th);
1914                 oti->oti_txns++;
1915                 rc = 0;
1916         } else {
1917                 rc = PTR_ERR(jh);
1918         }
1919 out:
1920         RETURN(rc);
1921 }
1922
1923 static int osd_seq_exists(const struct lu_env *env,
1924                           struct osd_device *osd, u64 seq)
1925 {
1926         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1927         struct seq_server_site *ss = osd_seq_site(osd);
1928         int rc;
1929
1930         ENTRY;
1931
1932         LASSERT(ss != NULL);
1933         LASSERT(ss->ss_server_fld != NULL);
1934
1935         rc = osd_fld_lookup(env, osd, seq, range);
1936         if (rc != 0) {
1937                 if (rc != -ENOENT)
1938                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1939                                osd_name(osd), seq, rc);
1940                 RETURN(0);
1941         }
1942
1943         RETURN(ss->ss_node_id == range->lsr_index);
1944 }
1945
1946 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1947 {
1948         struct dt_txn_commit_cb *dcb;
1949         struct dt_txn_commit_cb *tmp;
1950
1951         /* call per-transaction stop callbacks if any */
1952         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1953                                  dcb_linkage) {
1954                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1955                          "commit callback entry: magic=%x name='%s'\n",
1956                          dcb->dcb_magic, dcb->dcb_name);
1957                 list_del_init(&dcb->dcb_linkage);
1958                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1959         }
1960 }
1961
1962 /*
1963  * Concurrency: shouldn't matter.
1964  */
1965 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1966                           struct thandle *th)
1967 {
1968         struct osd_thread_info *oti = osd_oti_get(env);
1969         struct osd_thandle *oh;
1970         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1971         struct osd_device *osd = osd_dt_dev(th->th_dev);
1972         struct qsd_instance *qsd = osd_def_qsd(osd);
1973         struct lquota_trans *qtrans;
1974         struct list_head truncates = LIST_HEAD_INIT(truncates);
1975         int rc = 0, remove_agents = 0;
1976
1977         ENTRY;
1978
1979         oh = container_of0(th, struct osd_thandle, ot_super);
1980
1981         remove_agents = oh->ot_remove_agents;
1982
1983         qtrans = oh->ot_quota_trans;
1984         oh->ot_quota_trans = NULL;
1985
1986         /* move locks to local list, stop tx, execute truncates */
1987         list_splice(&oh->ot_trunc_locks, &truncates);
1988
1989         if (oh->ot_handle != NULL) {
1990                 int rc2;
1991
1992                 handle_t *hdl = oh->ot_handle;
1993
1994                 /*
1995                  * add commit callback
1996                  * notice we don't do this in osd_trans_start()
1997                  * as underlying transaction can change during truncate
1998                  */
1999                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2000                                              &oh->ot_jcb);
2001
2002                 LASSERT(oti->oti_txns == 1);
2003                 oti->oti_txns--;
2004
2005                 rc = dt_txn_hook_stop(env, th);
2006                 if (rc != 0)
2007                         CERROR("%s: failed in transaction hook: rc = %d\n",
2008                                osd_name(osd), rc);
2009
2010                 osd_trans_stop_cb(oh, rc);
2011                 /* hook functions might modify th_sync */
2012                 hdl->h_sync = th->th_sync;
2013
2014                 oh->ot_handle = NULL;
2015                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2016                 if (rc2 != 0)
2017                         CERROR("%s: failed to stop transaction: rc = %d\n",
2018                                osd_name(osd), rc2);
2019                 if (!rc)
2020                         rc = rc2;
2021
2022                 osd_process_truncates(&truncates);
2023         } else {
2024                 osd_trans_stop_cb(oh, th->th_result);
2025                 OBD_FREE_PTR(oh);
2026         }
2027
2028         osd_trunc_unlock_all(&truncates);
2029
2030         /* inform the quota slave device that the transaction is stopping */
2031         qsd_op_end(env, qsd, qtrans);
2032
2033         /*
2034          * as we want IO to journal and data IO be concurrent, we don't block
2035          * awaiting data IO completion in osd_do_bio(), instead we wait here
2036          * once transaction is submitted to the journal. all reqular requests
2037          * don't do direct IO (except read/write), thus this wait_event becomes
2038          * no-op for them.
2039          *
2040          * IMPORTANT: we have to wait till any IO submited by the thread is
2041          * completed otherwise iobuf may be corrupted by different request
2042          */
2043         wait_event(iobuf->dr_wait,
2044                        atomic_read(&iobuf->dr_numreqs) == 0);
2045         osd_fini_iobuf(osd, iobuf);
2046         if (!rc)
2047                 rc = iobuf->dr_error;
2048
2049         if (unlikely(remove_agents != 0))
2050                 osd_process_scheduled_agent_removals(env, osd);
2051
2052         oti->oti_ins_cache_depth--;
2053         /* reset OI cache for safety */
2054         if (oti->oti_ins_cache_depth == 0)
2055                 oti->oti_ins_cache_used = 0;
2056
2057         sb_end_write(osd_sb(osd));
2058
2059         RETURN(rc);
2060 }
2061
2062 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2063 {
2064         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2065                                                ot_super);
2066
2067         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2068         LASSERT(&dcb->dcb_func != NULL);
2069         if (dcb->dcb_flags & DCB_TRANS_STOP)
2070                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2071         else
2072                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2073
2074         return 0;
2075 }
2076
2077 /*
2078  * Called just before object is freed. Releases all resources except for
2079  * object itself (that is released by osd_object_free()).
2080  *
2081  * Concurrency: no concurrent access is possible that late in object
2082  * life-cycle.
2083  */
2084 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2085 {
2086         struct osd_object *obj = osd_obj(l);
2087         struct inode *inode = obj->oo_inode;
2088
2089         LINVRNT(osd_invariant(obj));
2090
2091         /*
2092          * If object is unlinked remove fid->ino mapping from object index.
2093          */
2094
2095         osd_index_fini(obj);
2096         if (inode != NULL) {
2097                 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2098                 qid_t                uid = i_uid_read(inode);
2099                 qid_t                gid = i_gid_read(inode);
2100
2101                 obj->oo_inode = NULL;
2102                 iput(inode);
2103                 if (!obj->oo_header && qsd) {
2104                         struct osd_thread_info *info = osd_oti_get(env);
2105                         struct lquota_id_info *qi = &info->oti_qi;
2106
2107                         /* Release granted quota to master if necessary */
2108                         qi->lqi_id.qid_uid = uid;
2109                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2110
2111                         qi->lqi_id.qid_uid = gid;
2112                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2113
2114                         qi->lqi_id.qid_uid = i_projid_read(inode);
2115                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2116                 }
2117         }
2118 }
2119
2120 /*
2121  * Concurrency: ->loo_object_release() is called under site spin-lock.
2122  */
2123 static void osd_object_release(const struct lu_env *env,
2124                                struct lu_object *l)
2125 {
2126         struct osd_object *o = osd_obj(l);
2127
2128         /*
2129          * nobody should be releasing a non-destroyed object with nlink=0
2130          * the API allows this, but ldiskfs doesn't like and then report
2131          * this inode as deleted
2132          */
2133         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2134                   o->oo_inode->i_nlink == 0));
2135 }
2136
2137 /*
2138  * Concurrency: shouldn't matter.
2139  */
2140 static int osd_object_print(const struct lu_env *env, void *cookie,
2141                             lu_printer_t p, const struct lu_object *l)
2142 {
2143         struct osd_object *o = osd_obj(l);
2144         struct iam_descr *d;
2145
2146         if (o->oo_dir != NULL)
2147                 d = o->oo_dir->od_container.ic_descr;
2148         else
2149                 d = NULL;
2150         return (*p)(env, cookie,
2151                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2152                     o, o->oo_inode,
2153                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2154                     o->oo_inode ? o->oo_inode->i_generation : 0,
2155                     d ? d->id_ops->id_name : "plain");
2156 }
2157
2158 /*
2159  * Concurrency: shouldn't matter.
2160  */
2161 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2162                 struct obd_statfs *sfs)
2163 {
2164         struct osd_device *osd = osd_dt_dev(d);
2165         struct super_block *sb = osd_sb(osd);
2166         struct kstatfs *ksfs;
2167         __u64 reserved;
2168         int result = 0;
2169
2170         if (unlikely(osd->od_mnt == NULL))
2171                 return -EINPROGRESS;
2172
2173         /* osd_lproc.c call this without env, allocate ksfs for that case */
2174         if (unlikely(env == NULL)) {
2175                 OBD_ALLOC_PTR(ksfs);
2176                 if (ksfs == NULL)
2177                         return -ENOMEM;
2178         } else {
2179                 ksfs = &osd_oti_get(env)->oti_ksfs;
2180         }
2181
2182         result = sb->s_op->statfs(sb->s_root, ksfs);
2183         if (result)
2184                 goto out;
2185
2186         statfs_pack(sfs, ksfs);
2187         if (unlikely(sb->s_flags & MS_RDONLY))
2188                 sfs->os_state |= OS_STATE_READONLY;
2189         if (ldiskfs_has_feature_extents(sb))
2190                 sfs->os_maxbytes = sb->s_maxbytes;
2191         else
2192                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2193
2194         /*
2195          * Reserve some space so to avoid fragmenting the filesystem too much.
2196          * Fragmentation not only impacts performance, but can also increase
2197          * metadata overhead significantly, causing grant calculation to be
2198          * wrong.
2199          *
2200          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2201          */
2202         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2203         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2204         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2205                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2206
2207         sfs->os_blocks -= reserved;
2208         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2209         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2210
2211 out:
2212         if (unlikely(env == NULL))
2213                 OBD_FREE_PTR(ksfs);
2214         return result;
2215 }
2216
2217 /**
2218  * Estimate space needed for file creations. We assume the largest filename
2219  * which is 2^64 - 1, hence a filename of 20 chars.
2220  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2221  */
2222 #ifdef __LDISKFS_DIR_REC_LEN
2223 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2224 #else
2225 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2226 #endif
2227
2228 /*
2229  * Concurrency: doesn't access mutable data.
2230  */
2231 static void osd_conf_get(const struct lu_env *env,
2232                          const struct dt_device *dev,
2233                          struct dt_device_param *param)
2234 {
2235         struct osd_device *d = osd_dt_dev(dev);
2236         struct super_block *sb = osd_sb(d);
2237         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2238         const char *name;
2239         int ea_overhead;
2240
2241         /*
2242          * XXX should be taken from not-yet-existing fs abstraction layer.
2243          */
2244         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2245         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2246         param->ddp_symlink_max  = sb->s_blocksize;
2247         param->ddp_mount_type   = LDD_MT_LDISKFS;
2248         if (ldiskfs_has_feature_extents(sb))
2249                 param->ddp_maxbytes = sb->s_maxbytes;
2250         else
2251                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2252         /*
2253          * inode are statically allocated, so per-inode space consumption
2254          * is the space consumed by the directory entry
2255          */
2256         param->ddp_inodespace     = PER_OBJ_USAGE;
2257         /*
2258          * EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2259          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2260          * maximum length to not under count the actual number of extents
2261          * needed for writing a file.
2262          */
2263         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2264         /* worst-case extent insertion metadata overhead */
2265         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2266         param->ddp_mntopts = 0;
2267         if (test_opt(sb, XATTR_USER))
2268                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2269         if (test_opt(sb, POSIX_ACL))
2270                 param->ddp_mntopts |= MNTOPT_ACL;
2271
2272         /*
2273          * LOD might calculate the max stripe count based on max_ea_size,
2274          * so we need take account in the overhead as well,
2275          * xattr_header + magic + xattr_entry_head
2276          */
2277         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2278                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2279
2280 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2281         if (ldiskfs_has_feature_ea_inode(sb))
2282                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2283                                                                 ea_overhead;
2284         else
2285 #endif
2286                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2287
2288         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2289                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2290
2291         /*
2292          * Preferred RPC size for efficient disk IO.  4MB shows good
2293          * all-around performance for ldiskfs, but use bigalloc chunk size
2294          * by default if larger.
2295          */
2296 #if defined(LDISKFS_CLUSTER_SIZE)
2297         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2298                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2299         else
2300 #endif
2301                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2302
2303         param->ddp_t10_cksum_type = 0;
2304         if (bi) {
2305                 unsigned short interval = blk_integrity_interval(bi);
2306                 name = blk_integrity_name(bi);
2307                 /*
2308                  * Expected values:
2309                  * T10-DIF-TYPE1-CRC
2310                  * T10-DIF-TYPE3-CRC
2311                  * T10-DIF-TYPE1-IP
2312                  * T10-DIF-TYPE3-IP
2313                  */
2314                 if (strncmp(name, "T10-DIF-TYPE",
2315                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2316                         /* also skip "1/3-" at end */
2317                         const int type_off = sizeof("T10-DIF-TYPE.");
2318                         char type_number = name[type_off - 2];
2319
2320                         if (interval != 512 && interval != 4096) {
2321                                 CERROR("%s: unsupported T10PI sector size %u\n",
2322                                        d->od_svname, interval);
2323                         } else if (type_number != '1' && type_number != '3') {
2324                                 CERROR("%s: unsupported T10PI type %s\n",
2325                                        d->od_svname, name);
2326                         } else if (strcmp(name + type_off, "CRC") == 0) {
2327                                 d->od_t10_type = type_number == '1' ?
2328                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2329                                 param->ddp_t10_cksum_type = interval == 512 ?
2330                                         OBD_CKSUM_T10CRC512 :
2331                                         OBD_CKSUM_T10CRC4K;
2332                         } else if (strcmp(name + type_off, "IP") == 0) {
2333                                 d->od_t10_type = type_number == '1' ?
2334                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2335                                 param->ddp_t10_cksum_type = interval == 512 ?
2336                                         OBD_CKSUM_T10IP512 :
2337                                         OBD_CKSUM_T10IP4K;
2338                         } else {
2339                                 CERROR("%s: unsupported checksum type of "
2340                                        "T10PI type '%s'",
2341                                        d->od_svname, name);
2342                         }
2343
2344                 } else {
2345                         CERROR("%s: unsupported T10PI type '%s'",
2346                                d->od_svname, name);
2347                 }
2348         }
2349 }
2350
2351 /*
2352  * Concurrency: shouldn't matter.
2353  */
2354 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2355 {
2356         int rc;
2357         struct super_block *s = osd_sb(osd_dt_dev(d));
2358         ENTRY;
2359
2360         down_read(&s->s_umount);
2361         rc = s->s_op->sync_fs(s, 1);
2362         up_read(&s->s_umount);
2363
2364         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2365                rc);
2366
2367         return rc;
2368 }
2369
2370 /**
2371  * Start commit for OSD device.
2372  *
2373  * An implementation of dt_commit_async method for OSD device.
2374  * Asychronously starts underlayng fs sync and thereby a transaction
2375  * commit.
2376  *
2377  * \param env environment
2378  * \param d dt device
2379  *
2380  * \see dt_device_operations
2381  */
2382 static int osd_commit_async(const struct lu_env *env,
2383                             struct dt_device *d)
2384 {
2385         struct super_block *s = osd_sb(osd_dt_dev(d));
2386         int rc;
2387
2388         ENTRY;
2389
2390         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2391         down_read(&s->s_umount);
2392         rc = s->s_op->sync_fs(s, 0);
2393         up_read(&s->s_umount);
2394
2395         RETURN(rc);
2396 }
2397
2398 /* Our own copy of the set readonly functions if present, or NU if not. */
2399 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2400 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2401 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2402
2403 /*
2404  * Concurrency: shouldn't matter.
2405  */
2406 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2407 {
2408         struct super_block *sb = osd_sb(osd_dt_dev(d));
2409         struct block_device *dev = sb->s_bdev;
2410         int rc = -EOPNOTSUPP;
2411
2412         ENTRY;
2413
2414         if (priv_dev_set_rdonly) {
2415                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2416
2417                 rc = 0;
2418                 CERROR("*** setting %s read-only ***\n",
2419                        osd_dt_dev(d)->od_svname);
2420
2421                 if (sb->s_op->freeze_fs) {
2422                         rc = sb->s_op->freeze_fs(sb);
2423                         if (rc)
2424                                 goto out;
2425                 }
2426
2427                 if (jdev && (jdev != dev)) {
2428                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2429                                (long)jdev);
2430                         priv_dev_set_rdonly(jdev);
2431                 }
2432                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2433                 priv_dev_set_rdonly(dev);
2434
2435                 if (sb->s_op->unfreeze_fs)
2436                         sb->s_op->unfreeze_fs(sb);
2437         }
2438
2439 out:
2440         if (rc)
2441                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2442                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2443
2444         RETURN(rc);
2445 }
2446
2447 /**
2448  * Note: we do not count into QUOTA here.
2449  * If we mount with --data_journal we may need more.
2450  */
2451 const int osd_dto_credits_noquota[DTO_NR] = {
2452         /**
2453          * Insert.
2454          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2455          * SINGLEDATA_TRANS_BLOCKS(8)
2456          * XXX Note: maybe iam need more, since iam have more level than
2457          *           EXT3 htree.
2458          */
2459         [DTO_INDEX_INSERT]  = 16,
2460         /**
2461          * Delete
2462          * just modify a single entry, probably merge few within a block
2463          */
2464         [DTO_INDEX_DELETE]  = 1,
2465         /**
2466          * Used for OI scrub
2467          */
2468         [DTO_INDEX_UPDATE]  = 16,
2469         /**
2470          * 4(inode, inode bits, groups, GDT)
2471          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2472          */
2473         [DTO_OBJECT_CREATE] = 4,
2474         /**
2475          * 4(inode, inode bits, groups, GDT)
2476          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2477          */
2478         [DTO_OBJECT_DELETE] = 4,
2479         /**
2480          * Attr set credits (inode)
2481          */
2482         [DTO_ATTR_SET_BASE] = 1,
2483         /**
2484          * Xattr set. The same as xattr of EXT3.
2485          * DATA_TRANS_BLOCKS(14)
2486          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2487          * are also counted in. Do not know why?
2488          */
2489         [DTO_XATTR_SET]     = 14,
2490         /**
2491          * credits for inode change during write.
2492          */
2493         [DTO_WRITE_BASE]    = 3,
2494         /**
2495          * credits for single block write.
2496          */
2497         [DTO_WRITE_BLOCK]   = 14,
2498         /**
2499          * Attr set credits for chown.
2500          * This is extra credits for setattr, and it is null without quota
2501          */
2502         [DTO_ATTR_SET_CHOWN] = 0
2503 };
2504
2505 static const struct dt_device_operations osd_dt_ops = {
2506         .dt_root_get       = osd_root_get,
2507         .dt_statfs         = osd_statfs,
2508         .dt_trans_create   = osd_trans_create,
2509         .dt_trans_start    = osd_trans_start,
2510         .dt_trans_stop     = osd_trans_stop,
2511         .dt_trans_cb_add   = osd_trans_cb_add,
2512         .dt_conf_get       = osd_conf_get,
2513         .dt_sync           = osd_sync,
2514         .dt_ro             = osd_ro,
2515         .dt_commit_async   = osd_commit_async,
2516 };
2517
2518 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2519                           unsigned int role)
2520 {
2521         struct osd_object *obj = osd_dt_obj(dt);
2522         struct osd_thread_info *oti = osd_oti_get(env);
2523
2524         LINVRNT(osd_invariant(obj));
2525
2526         LASSERT(obj->oo_owner != env);
2527         down_read_nested(&obj->oo_sem, role);
2528
2529         LASSERT(obj->oo_owner == NULL);
2530         oti->oti_r_locks++;
2531 }
2532
2533 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2534                            unsigned int role)
2535 {
2536         struct osd_object *obj = osd_dt_obj(dt);
2537         struct osd_thread_info *oti = osd_oti_get(env);
2538
2539         LINVRNT(osd_invariant(obj));
2540
2541         LASSERT(obj->oo_owner != env);
2542         down_write_nested(&obj->oo_sem, role);
2543
2544         LASSERT(obj->oo_owner == NULL);
2545         obj->oo_owner = env;
2546         oti->oti_w_locks++;
2547 }
2548
2549 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2550 {
2551         struct osd_object *obj = osd_dt_obj(dt);
2552         struct osd_thread_info *oti = osd_oti_get(env);
2553
2554         LINVRNT(osd_invariant(obj));
2555
2556         LASSERT(oti->oti_r_locks > 0);
2557         oti->oti_r_locks--;
2558         up_read(&obj->oo_sem);
2559 }
2560
2561 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2562 {
2563         struct osd_object *obj = osd_dt_obj(dt);
2564         struct osd_thread_info *oti = osd_oti_get(env);
2565
2566         LINVRNT(osd_invariant(obj));
2567
2568         LASSERT(obj->oo_owner == env);
2569         LASSERT(oti->oti_w_locks > 0);
2570         oti->oti_w_locks--;
2571         obj->oo_owner = NULL;
2572         up_write(&obj->oo_sem);
2573 }
2574
2575 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2576 {
2577         struct osd_object *obj = osd_dt_obj(dt);
2578
2579         LINVRNT(osd_invariant(obj));
2580
2581         return obj->oo_owner == env;
2582 }
2583
2584 static struct timespec *osd_inode_time(const struct lu_env *env,
2585                                        struct inode *inode, __u64 seconds)
2586 {
2587         struct osd_thread_info *oti = osd_oti_get(env);
2588         struct timespec *t = &oti->oti_time;
2589
2590         t->tv_sec = seconds;
2591         t->tv_nsec = 0;
2592         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2593         return t;
2594 }
2595
2596 static void osd_inode_getattr(const struct lu_env *env,
2597                               struct inode *inode, struct lu_attr *attr)
2598 {
2599         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2600                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2601                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2602                            LA_BLKSIZE | LA_TYPE;
2603
2604         attr->la_atime = inode->i_atime.tv_sec;
2605         attr->la_mtime = inode->i_mtime.tv_sec;
2606         attr->la_ctime = inode->i_ctime.tv_sec;
2607         attr->la_mode    = inode->i_mode;
2608         attr->la_size    = i_size_read(inode);
2609         attr->la_blocks  = inode->i_blocks;
2610         attr->la_uid     = i_uid_read(inode);
2611         attr->la_gid     = i_gid_read(inode);
2612         attr->la_projid  = i_projid_read(inode);
2613         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2614         attr->la_nlink   = inode->i_nlink;
2615         attr->la_rdev    = inode->i_rdev;
2616         attr->la_blksize = 1 << inode->i_blkbits;
2617         attr->la_blkbits = inode->i_blkbits;
2618         /*
2619          * Ext4 did not transfer inherit flags from raw inode
2620          * to inode flags, and ext4 internally test raw inode
2621          * @i_flags directly. Instead of patching ext4, we do it here.
2622          */
2623         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2624                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2625 }
2626
2627 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2628                         struct lu_attr *attr)
2629 {
2630         struct osd_object *obj = osd_dt_obj(dt);
2631
2632         if (unlikely(!dt_object_exists(dt)))
2633                 return -ENOENT;
2634         if (unlikely(obj->oo_destroyed))
2635                 return -ENOENT;
2636
2637         LASSERT(!dt_object_remote(dt));
2638         LINVRNT(osd_invariant(obj));
2639
2640         spin_lock(&obj->oo_guard);
2641         osd_inode_getattr(env, obj->oo_inode, attr);
2642         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2643                 attr->la_valid |= LA_FLAGS;
2644                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2645         }
2646         spin_unlock(&obj->oo_guard);
2647
2648         return 0;
2649 }
2650
2651 static int osd_declare_attr_qid(const struct lu_env *env,
2652                                 struct osd_object *obj,
2653                                 struct osd_thandle *oh, long long bspace,
2654                                 qid_t old_id, qid_t new_id, bool enforce,
2655                                 unsigned int type, bool ignore_edquot)
2656 {
2657         int rc;
2658         struct osd_thread_info *info = osd_oti_get(env);
2659         struct lquota_id_info  *qi = &info->oti_qi;
2660
2661         qi->lqi_type = type;
2662         /* inode accounting */
2663         qi->lqi_is_blk = false;
2664
2665         /* one more inode for the new id ... */
2666         qi->lqi_id.qid_uid = new_id;
2667         qi->lqi_space      = 1;
2668         /* Reserve credits for the new id */
2669         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2670         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2671                 rc = 0;
2672         if (rc)
2673                 RETURN(rc);
2674
2675         /* and one less inode for the current id */
2676         qi->lqi_id.qid_uid = old_id;
2677         qi->lqi_space = -1;
2678         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2679         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2680                 rc = 0;
2681         if (rc)
2682                 RETURN(rc);
2683
2684         /* block accounting */
2685         qi->lqi_is_blk = true;
2686
2687         /* more blocks for the new id ... */
2688         qi->lqi_id.qid_uid = new_id;
2689         qi->lqi_space      = bspace;
2690         /*
2691          * Credits for the new uid has been reserved, re-use "obj"
2692          * to save credit reservation.
2693          */
2694         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2695         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2696                 rc = 0;
2697         if (rc)
2698                 RETURN(rc);
2699
2700         /* and finally less blocks for the current uid */
2701         qi->lqi_id.qid_uid = old_id;
2702         qi->lqi_space      = -bspace;
2703         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2704         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2705                 rc = 0;
2706
2707         RETURN(rc);
2708 }
2709
2710 static int osd_declare_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_thandle *oh;
2716         struct osd_object *obj;
2717         qid_t uid;
2718         qid_t gid;
2719         long long bspace;
2720         int rc = 0;
2721         bool enforce;
2722
2723         ENTRY;
2724
2725         LASSERT(dt != NULL);
2726         LASSERT(handle != NULL);
2727
2728         obj = osd_dt_obj(dt);
2729         LASSERT(osd_invariant(obj));
2730
2731         oh = container_of0(handle, struct osd_thandle, ot_super);
2732         LASSERT(oh->ot_handle == NULL);
2733
2734         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2735                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2736
2737         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2738                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2739
2740         if (attr == NULL || obj->oo_inode == NULL)
2741                 RETURN(rc);
2742
2743         bspace   = obj->oo_inode->i_blocks << 9;
2744         bspace   = toqb(bspace);
2745
2746         /*
2747          * Changing ownership is always preformed by super user, it should not
2748          * fail with EDQUOT unless required explicitly.
2749          *
2750          * We still need to call the osd_declare_qid() to calculate the journal
2751          * credits for updating quota accounting files and to trigger quota
2752          * space adjustment once the operation is completed.
2753          */
2754         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2755                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2756
2757                 if (!ignore_edquot)
2758                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2759                                "(the quota space is %lld)\n",
2760                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2761                                attr->la_gid, bspace);
2762
2763                 /* USERQUOTA */
2764                 uid = i_uid_read(obj->oo_inode);
2765                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2766                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2767                                           attr->la_uid, enforce, USRQUOTA,
2768                                           true);
2769                 if (rc)
2770                         RETURN(rc);
2771
2772                 gid = i_gid_read(obj->oo_inode);
2773                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2774                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2775                                           i_gid_read(obj->oo_inode),
2776                                           attr->la_gid, enforce, GRPQUOTA,
2777                                           ignore_edquot);
2778                 if (rc)
2779                         RETURN(rc);
2780
2781         }
2782 #ifdef HAVE_PROJECT_QUOTA
2783         if (attr->la_valid & LA_PROJID) {
2784                 __u32 projid = i_projid_read(obj->oo_inode);
2785
2786                 enforce = (attr->la_valid & LA_PROJID) &&
2787                                         (attr->la_projid != projid);
2788                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2789                                           (qid_t)projid, (qid_t)attr->la_projid,
2790                                           enforce, PRJQUOTA, true);
2791                 if (rc)
2792                         RETURN(rc);
2793         }
2794 #endif
2795         RETURN(rc);
2796 }
2797
2798 static int osd_inode_setattr(const struct lu_env *env,
2799                              struct inode *inode, const struct lu_attr *attr)
2800 {
2801         __u64 bits = attr->la_valid;
2802
2803         /* Only allow set size for regular file */
2804         if (!S_ISREG(inode->i_mode))
2805                 bits &= ~(LA_SIZE | LA_BLOCKS);
2806
2807         if (bits == 0)
2808                 return 0;
2809
2810         if (bits & LA_ATIME)
2811                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2812         if (bits & LA_CTIME)
2813                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2814         if (bits & LA_MTIME)
2815                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2816         if (bits & LA_SIZE) {
2817                 spin_lock(&inode->i_lock);
2818                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2819                 i_size_write(inode, attr->la_size);
2820                 spin_unlock(&inode->i_lock);
2821         }
2822
2823         /*
2824          * OSD should not change "i_blocks" which is used by quota.
2825          * "i_blocks" should be changed by ldiskfs only.
2826          */
2827         if (bits & LA_MODE)
2828                 inode->i_mode = (inode->i_mode & S_IFMT) |
2829                                 (attr->la_mode & ~S_IFMT);
2830         if (bits & LA_UID)
2831                 i_uid_write(inode, attr->la_uid);
2832         if (bits & LA_GID)
2833                 i_gid_write(inode, attr->la_gid);
2834         if (bits & LA_PROJID)
2835                 i_projid_write(inode, attr->la_projid);
2836         if (bits & LA_NLINK)
2837                 set_nlink(inode, attr->la_nlink);
2838         if (bits & LA_RDEV)
2839                 inode->i_rdev = attr->la_rdev;
2840
2841         if (bits & LA_FLAGS) {
2842                 /* always keep S_NOCMTIME */
2843                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2844                                  S_NOCMTIME;
2845                 /*
2846                  * Ext4 did not transfer inherit flags from
2847                  * @inode->i_flags to raw inode i_flags when writing
2848                  * flags, we do it explictly here.
2849                  */
2850                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2851                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2852                 else
2853                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2854         }
2855         return 0;
2856 }
2857
2858 #ifdef HAVE_PROJECT_QUOTA
2859 static int osd_transfer_project(struct inode *inode, __u32 projid)
2860 {
2861         struct super_block *sb = inode->i_sb;
2862         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2863         int err;
2864         kprojid_t kprojid;
2865         struct ldiskfs_iloc iloc;
2866         struct ldiskfs_inode *raw_inode;
2867         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2868
2869         if (!ldiskfs_has_feature_project(sb)) {
2870                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2871                         == LDISKFS_DEF_PROJID);
2872                 if (projid != LDISKFS_DEF_PROJID)
2873                         return -EOPNOTSUPP;
2874                 else
2875                         return 0;
2876         }
2877
2878         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2879                 return -EOPNOTSUPP;
2880
2881         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2882         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2883                 return 0;
2884
2885         err = ldiskfs_get_inode_loc(inode, &iloc);
2886         if (err)
2887                 return err;
2888
2889         raw_inode = ldiskfs_raw_inode(&iloc);
2890         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2891                 err = -EOVERFLOW;
2892                 brelse(iloc.bh);
2893                 return err;
2894         }
2895         brelse(iloc.bh);
2896
2897         dquot_initialize(inode);
2898         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2899         if (transfer_to[PRJQUOTA]) {
2900                 err = __dquot_transfer(inode, transfer_to);
2901                 dqput(transfer_to[PRJQUOTA]);
2902                 if (err)
2903                         return err;
2904         }
2905
2906         return err;
2907 }
2908 #endif
2909
2910 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2911 {
2912         int rc;
2913
2914         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2915             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2916                 struct iattr iattr;
2917
2918                 ll_vfs_dq_init(inode);
2919                 iattr.ia_valid = 0;
2920                 if (attr->la_valid & LA_UID)
2921                         iattr.ia_valid |= ATTR_UID;
2922                 if (attr->la_valid & LA_GID)
2923                         iattr.ia_valid |= ATTR_GID;
2924                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2925                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2926
2927                 rc = ll_vfs_dq_transfer(inode, &iattr);
2928                 if (rc) {
2929                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2930                                "enforcement enabled on the ldiskfs "
2931                                "filesystem?\n", inode->i_sb->s_id, rc);
2932                         return rc;
2933                 }
2934         }
2935
2936         /* Handle project id transfer here properly */
2937         if (attr->la_valid & LA_PROJID &&
2938             attr->la_projid != i_projid_read(inode)) {
2939 #ifdef HAVE_PROJECT_QUOTA
2940                 rc = osd_transfer_project(inode, attr->la_projid);
2941 #else
2942                 rc = -ENOTSUPP;
2943 #endif
2944                 if (rc) {
2945                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2946                                "enforcement enabled on the ldiskfs "
2947                                "filesystem?\n", inode->i_sb->s_id, rc);
2948                         return rc;
2949                 }
2950         }
2951         return 0;
2952 }
2953
2954 static int osd_attr_set(const struct lu_env *env,
2955                         struct dt_object *dt,
2956                         const struct lu_attr *attr,
2957                         struct thandle *handle)
2958 {
2959         struct osd_object *obj = osd_dt_obj(dt);
2960         struct inode *inode;
2961         int rc;
2962
2963         if (!dt_object_exists(dt))
2964                 return -ENOENT;
2965
2966         LASSERT(handle != NULL);
2967         LASSERT(!dt_object_remote(dt));
2968         LASSERT(osd_invariant(obj));
2969
2970         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2971
2972         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2973             !osd_obj2dev(obj)->od_is_ost) {
2974                 struct osd_thread_info *oti = osd_oti_get(env);
2975                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2976                 struct lu_fid *fid1 = &oti->oti_fid;
2977                 struct osd_inode_id *id = &oti->oti_id;
2978                 struct iam_path_descr *ipd;
2979                 struct iam_container *bag;
2980                 struct osd_thandle *oh;
2981                 int rc;
2982
2983                 fid_cpu_to_be(fid1, fid0);
2984                 memset(id, 1, sizeof(*id));
2985                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2986                                   fid0)->oi_dir.od_container;
2987                 ipd = osd_idx_ipd_get(env, bag);
2988                 if (unlikely(ipd == NULL))
2989                         RETURN(-ENOMEM);
2990
2991                 oh = container_of0(handle, struct osd_thandle, ot_super);
2992                 rc = iam_update(oh->ot_handle, bag,
2993                                 (const struct iam_key *)fid1,
2994                                 (const struct iam_rec *)id, ipd);
2995                 osd_ipd_put(env, bag, ipd);
2996                 return(rc > 0 ? 0 : rc);
2997         }
2998
2999         inode = obj->oo_inode;
3000
3001         rc = osd_quota_transfer(inode, attr);
3002         if (rc)
3003                 return rc;
3004
3005         spin_lock(&obj->oo_guard);
3006         rc = osd_inode_setattr(env, inode, attr);
3007         spin_unlock(&obj->oo_guard);
3008         if (rc != 0)
3009                 GOTO(out, rc);
3010
3011         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3012
3013         if (!(attr->la_valid & LA_FLAGS))
3014                 GOTO(out, rc);
3015
3016         /* Let's check if there are extra flags need to be set into LMA */
3017         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3018                 struct osd_thread_info *info = osd_oti_get(env);
3019                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3020
3021                 LASSERT(!obj->oo_pfid_in_lma);
3022
3023                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3024                                  &info->oti_ost_attrs);
3025                 if (rc)
3026                         GOTO(out, rc);
3027
3028                 lma->lma_incompat |=
3029                         lustre_to_lma_flags(attr->la_flags);
3030                 lustre_lma_swab(lma);
3031                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3032                                      lma, sizeof(*lma), XATTR_REPLACE);
3033                 if (rc != 0) {
3034                         struct osd_device *osd = osd_obj2dev(obj);
3035
3036                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3037                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3038                               lma->lma_incompat, rc);
3039                 } else {
3040                         obj->oo_lma_flags =
3041                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3042                 }
3043                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3044         }
3045 out:
3046         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3047
3048         return rc;
3049 }
3050
3051 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3052                                            struct osd_object *obj,
3053                                            const char *name, const int namelen)
3054 {
3055         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3056 }
3057
3058 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3059                       umode_t mode, struct dt_allocation_hint *hint,
3060                       struct thandle *th)
3061 {
3062         int result;
3063         struct osd_device *osd = osd_obj2dev(obj);
3064         struct osd_thandle *oth;
3065         struct dt_object *parent = NULL;
3066         struct inode *inode;
3067
3068         LINVRNT(osd_invariant(obj));
3069         LASSERT(obj->oo_inode == NULL);
3070         LASSERT(obj->oo_hl_head == NULL);
3071
3072         if (S_ISDIR(mode) && ldiskfs_pdo) {
3073                 obj->oo_hl_head =
3074                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3075                 if (obj->oo_hl_head == NULL)
3076                         return -ENOMEM;
3077         }
3078
3079         oth = container_of(th, struct osd_thandle, ot_super);
3080         LASSERT(oth->ot_handle->h_transaction != NULL);
3081
3082         if (hint != NULL && hint->dah_parent != NULL &&
3083             !dt_object_remote(hint->dah_parent))
3084                 parent = hint->dah_parent;
3085
3086         inode = ldiskfs_create_inode(oth->ot_handle,
3087                                      parent ? osd_dt_obj(parent)->oo_inode :
3088                                               osd_sb(osd)->s_root->d_inode,
3089                                      mode);
3090         if (!IS_ERR(inode)) {
3091                 /* Do not update file c/mtime in ldiskfs. */
3092                 inode->i_flags |= S_NOCMTIME;
3093
3094                 /*
3095                  * For new created object, it must be consistent,
3096                  * and it is unnecessary to scrub against it.
3097                  */
3098                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3099
3100                 obj->oo_inode = inode;
3101                 result = 0;
3102         } else {
3103                 if (obj->oo_hl_head != NULL) {
3104                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3105                         obj->oo_hl_head = NULL;
3106                 }
3107                 result = PTR_ERR(inode);
3108         }
3109         LINVRNT(osd_invariant(obj));
3110         return result;
3111 }
3112
3113 enum {
3114         OSD_NAME_LEN = 255
3115 };
3116
3117 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3118                      struct lu_attr *attr,
3119                      struct dt_allocation_hint *hint,
3120                      struct dt_object_format *dof,
3121                      struct thandle *th)
3122 {
3123         int result;
3124         struct osd_thandle *oth;
3125         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3126
3127         LASSERT(S_ISDIR(attr->la_mode));
3128
3129         oth = container_of(th, struct osd_thandle, ot_super);
3130         LASSERT(oth->ot_handle->h_transaction != NULL);
3131         result = osd_mkfile(info, obj, mode, hint, th);
3132
3133         return result;
3134 }
3135
3136 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3137                         struct lu_attr *attr,
3138                         struct dt_allocation_hint *hint,
3139                         struct dt_object_format *dof,
3140                         struct thandle *th)
3141 {
3142         int result;
3143         struct osd_thandle *oth;
3144         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3145
3146         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3147
3148         LASSERT(S_ISREG(attr->la_mode));
3149
3150         oth = container_of(th, struct osd_thandle, ot_super);
3151         LASSERT(oth->ot_handle->h_transaction != NULL);
3152
3153         result = osd_mkfile(info, obj, mode, hint, th);
3154         if (result == 0) {
3155                 LASSERT(obj->oo_inode != NULL);
3156                 if (feat->dif_flags & DT_IND_VARKEY)
3157                         result = iam_lvar_create(obj->oo_inode,
3158                                                  feat->dif_keysize_max,
3159                                                  feat->dif_ptrsize,
3160                                                  feat->dif_recsize_max,
3161                                                  oth->ot_handle);
3162                 else
3163                         result = iam_lfix_create(obj->oo_inode,
3164                                                  feat->dif_keysize_max,
3165                                                  feat->dif_ptrsize,
3166                                                  feat->dif_recsize_max,
3167                                                  oth->ot_handle);
3168         }
3169         return result;
3170 }
3171
3172 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3173                      struct lu_attr *attr,
3174                      struct dt_allocation_hint *hint,
3175                      struct dt_object_format *dof,
3176                      struct thandle *th)
3177 {
3178         LASSERT(S_ISREG(attr->la_mode));
3179         return osd_mkfile(info, obj, (attr->la_mode &
3180                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3181 }
3182
3183 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3184                      struct lu_attr *attr,
3185                      struct dt_allocation_hint *hint,
3186                      struct dt_object_format *dof,
3187                      struct thandle *th)
3188 {
3189         LASSERT(S_ISLNK(attr->la_mode));
3190         return osd_mkfile(info, obj, (attr->la_mode &
3191                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3192 }
3193
3194 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3195                      struct lu_attr *attr,
3196                      struct dt_allocation_hint *hint,
3197                      struct dt_object_format *dof,
3198                      struct thandle *th)
3199 {
3200         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3201         int result;
3202
3203         LINVRNT(osd_invariant(obj));
3204         LASSERT(obj->oo_inode == NULL);
3205         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3206                 S_ISFIFO(mode) || S_ISSOCK(mode));
3207
3208         result = osd_mkfile(info, obj, mode, hint, th);
3209         if (result == 0) {
3210                 LASSERT(obj->oo_inode != NULL);
3211                 /*
3212                  * This inode should be marked dirty for i_rdev.  Currently
3213                  * that is done in the osd_attr_init().
3214                  */
3215                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3216                                    attr->la_rdev);
3217         }
3218         LINVRNT(osd_invariant(obj));
3219         return result;
3220 }
3221
3222 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3223                               struct lu_attr *,
3224                               struct dt_allocation_hint *hint,
3225                               struct dt_object_format *dof,
3226                               struct thandle *);
3227
3228 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3229 {
3230         osd_obj_type_f result;
3231
3232         switch (type) {
3233         case DFT_DIR:
3234                 result = osd_mkdir;
3235                 break;
3236         case DFT_REGULAR:
3237                 result = osd_mkreg;
3238                 break;
3239         case DFT_SYM:
3240                 result = osd_mksym;
3241                 break;
3242         case DFT_NODE:
3243                 result = osd_mknod;
3244                 break;
3245         case DFT_INDEX:
3246                 result = osd_mk_index;
3247                 break;
3248
3249         default:
3250                 LBUG();
3251                 break;
3252         }
3253         return result;
3254 }
3255
3256
3257 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3258                         struct dt_object *parent, struct dt_object *child,
3259                         umode_t child_mode)
3260 {
3261         LASSERT(ah);
3262
3263         ah->dah_parent = parent;
3264         ah->dah_mode = child_mode;
3265
3266         if (parent != NULL && !dt_object_remote(parent)) {
3267                 /* will help to find FID->ino at dt_insert("..") */
3268                 struct osd_object *pobj = osd_dt_obj(parent);
3269
3270                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3271         }
3272 }
3273
3274 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3275                           struct lu_attr *attr, struct dt_object_format *dof)
3276 {
3277         struct inode *inode = obj->oo_inode;
3278         __u64 valid = attr->la_valid;
3279         int result;
3280
3281         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3282
3283         if (dof->dof_type != DFT_NODE)
3284                 attr->la_valid &= ~LA_RDEV;
3285         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3286                 attr->la_valid &= ~LA_ATIME;
3287         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3288                 attr->la_valid &= ~LA_CTIME;
3289         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3290                 attr->la_valid &= ~LA_MTIME;
3291
3292         result = osd_quota_transfer(inode, attr);
3293         if (result)
3294                 return;
3295
3296         if (attr->la_valid != 0) {
3297                 result = osd_inode_setattr(info->oti_env, inode, attr);
3298                 /*
3299                  * The osd_inode_setattr() should always succeed here.  The
3300                  * only error that could be returned is EDQUOT when we are
3301                  * trying to change the UID or GID of the inode. However, this
3302                  * should not happen since quota enforcement is no longer
3303                  * enabled on ldiskfs (lquota takes care of it).
3304                  */
3305                 LASSERTF(result == 0, "%d\n", result);
3306                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3307         }
3308
3309         attr->la_valid = valid;
3310 }
3311
3312 /**
3313  * Helper function for osd_create()
3314  *
3315  * \retval 0, on success
3316  */
3317 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3318                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3319                         struct dt_object_format *dof, struct thandle *th)
3320 {
3321         int result;
3322         __u32 umask;
3323
3324         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3325
3326         /* we drop umask so that permissions we pass are not affected */
3327         umask = current->fs->umask;
3328         current->fs->umask = 0;
3329
3330         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3331                                                   th);
3332         if (likely(obj->oo_inode != NULL)) {
3333                 LASSERT(obj->oo_inode->i_state & I_NEW);
3334
3335                 /*
3336                  * Unlock the inode before attr initialization to avoid
3337                  * unnecessary dqget operations. LU-6378
3338                  */
3339                 unlock_new_inode(obj->oo_inode);
3340         }
3341
3342         if (likely(result == 0)) {
3343                 osd_attr_init(info, obj, attr, dof);
3344                 osd_object_init0(obj);
3345         }
3346
3347         /* restore previous umask value */
3348         current->fs->umask = umask;
3349
3350         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3351
3352         return result;
3353 }
3354
3355 /**
3356  * Helper function for osd_create()
3357  *
3358  * \retval 0, on success
3359  */
3360 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3361                            const struct lu_fid *fid, struct thandle *th)
3362 {
3363         struct osd_thread_info *info = osd_oti_get(env);
3364         struct osd_inode_id    *id   = &info->oti_id;
3365         struct osd_device      *osd  = osd_obj2dev(obj);
3366         struct osd_thandle     *oh;
3367         int rc;
3368
3369         LASSERT(obj->oo_inode != NULL);
3370
3371         oh = container_of0(th, struct osd_thandle, ot_super);
3372         LASSERT(oh->ot_handle);
3373         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3374
3375         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3376         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3377                            OI_CHECK_FLD, NULL);
3378         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3379
3380         return rc;
3381 }
3382
3383 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3384                    u64 seq, struct lu_seq_range *range)
3385 {
3386         struct seq_server_site *ss = osd_seq_site(osd);
3387
3388         if (fid_seq_is_idif(seq)) {
3389                 fld_range_set_ost(range);
3390                 range->lsr_index = idif_ost_idx(seq);
3391                 return 0;
3392         }
3393
3394         if (!fid_seq_in_fldb(seq)) {
3395                 fld_range_set_mdt(range);
3396                 if (ss != NULL)
3397                         /*
3398                          * FIXME: If ss is NULL, it suppose not get lsr_index
3399                          * at all
3400                          */
3401                         range->lsr_index = ss->ss_node_id;
3402                 return 0;
3403         }
3404
3405         LASSERT(ss != NULL);
3406         fld_range_set_any(range);
3407         /* OSD will only do local fld lookup */
3408         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3409 }
3410
3411 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3412                               struct lu_attr *attr,
3413                               struct dt_allocation_hint *hint,
3414                               struct dt_object_format *dof,
3415                               struct thandle *handle)
3416 {
3417         struct osd_thandle *oh;
3418         int rc;
3419
3420         ENTRY;
3421
3422         LASSERT(handle != NULL);
3423
3424         oh = container_of0(handle, struct osd_thandle, ot_super);
3425         LASSERT(oh->ot_handle == NULL);
3426
3427         /*
3428          * EA object consumes more credits than regular object: osd_mk_index
3429          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3430          * leaf_node, could involves the block, block bitmap, groups, GDT
3431          * change for each block, so add 4 * 2 credits in that case.
3432          */
3433         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3434                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3435                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3436         /*
3437          * Reuse idle OI block may cause additional one OI block
3438          * to be changed.
3439          */
3440         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3441                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3442
3443         /* will help to find FID->ino mapping at dt_insert() */
3444         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3445                                    osd_dt_obj(dt));
3446         if (rc != 0)
3447                 RETURN(rc);
3448
3449         if (!attr)
3450                 RETURN(0);
3451
3452         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3453                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3454                                    NULL, OSD_QID_INODE);
3455         if (rc != 0)
3456                 RETURN(rc);
3457
3458         RETURN(rc);
3459 }
3460
3461 /**
3462  * Called to destroy on-disk representation of the object
3463  *
3464  * Concurrency: must be locked
3465  */
3466 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3467                                struct thandle *th)
3468 {
3469         struct osd_object *obj = osd_dt_obj(dt);
3470         struct inode *inode = obj->oo_inode;
3471         struct osd_thandle *oh;
3472         int rc;
3473
3474         ENTRY;
3475
3476         if (inode == NULL)
3477                 RETURN(-ENOENT);
3478
3479         oh = container_of0(th, struct osd_thandle, ot_super);
3480         LASSERT(oh->ot_handle == NULL);
3481
3482         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3483                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3484
3485         /* For removing agent entry */
3486         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3487                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3488
3489         /*
3490          * Recycle idle OI leaf may cause additional three OI blocks
3491          * to be changed.
3492          */
3493         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3494                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3495                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3496         /* one less inode */
3497         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3498                                    i_projid_read(inode), -1, oh, obj, NULL,
3499                                    OSD_QID_INODE);
3500         if (rc)
3501                 RETURN(rc);
3502         /* data to be truncated */
3503         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3504                                    i_projid_read(inode), 0, oh, obj, NULL,
3505                                    OSD_QID_BLK);
3506         if (rc)
3507                 RETURN(rc);
3508
3509         /*
3510          * will help to find FID->ino when this object is being
3511          * added to PENDING
3512          */
3513         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3514
3515         RETURN(rc);
3516 }
3517
3518 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3519                        struct thandle *th)
3520 {
3521         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3522         struct osd_object *obj = osd_dt_obj(dt);
3523         struct inode *inode = obj->oo_inode;
3524         struct osd_device *osd = osd_obj2dev(obj);
3525         struct osd_thandle *oh;
3526         int result;
3527
3528         ENTRY;
3529
3530         oh = container_of0(th, struct osd_thandle, ot_super);
3531         LASSERT(oh->ot_handle);
3532         LASSERT(inode);
3533         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3534
3535         if (unlikely(fid_is_acct(fid)))
3536                 RETURN(-EPERM);
3537
3538         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3539                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3540                 if (result != 0)
3541                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3542                                osd_name(osd), PFID(fid), result);
3543         }
3544
3545         if (S_ISDIR(inode->i_mode)) {
3546                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3547                         inode->i_nlink == 2);
3548
3549                 spin_lock(&obj->oo_guard);
3550                 clear_nlink(inode);
3551                 spin_unlock(&obj->oo_guard);
3552                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3553         }
3554
3555         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3556
3557         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3558
3559         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3560                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3561                                        oh->ot_handle, OI_CHECK_FLD);
3562
3563         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3564         /* XXX: add to ext3 orphan list */
3565         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3566
3567         /* not needed in the cache anymore */
3568         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3569         obj->oo_destroyed = 1;
3570
3571         RETURN(0);
3572 }
3573
3574 /**
3575  * Put the fid into lustre_mdt_attrs, and then place the structure
3576  * inode's ea. This fid should not be altered during the life time
3577  * of the inode.
3578  *
3579  * \retval +ve, on success
3580  * \retval -ve, on error
3581  *
3582  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3583  */
3584 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3585                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3586 {
3587         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3588         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3589         int rc;
3590
3591         ENTRY;
3592
3593         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3594                 RETURN(0);
3595
3596         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3597                 rc = -ENOMEM;
3598
3599         lustre_loa_init(loa, fid, compat, incompat);
3600         lustre_loa_swab(loa, false);
3601
3602         /*
3603          * For the OST device with 256 bytes inode size by default,
3604          * the PFID EA will be stored together with LMA EA to avoid
3605          * performance trouble. Otherwise the PFID EA can be stored
3606          * independently. LU-8998
3607          */
3608         if ((compat & LMAC_FID_ON_OST) &&
3609             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3610                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3611                                      sizeof(*loa), XATTR_CREATE);
3612         else
3613                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3614                                      sizeof(*lma), XATTR_CREATE);
3615         /*
3616          * LMA may already exist, but we need to check that all the
3617          * desired compat/incompat flags have been added.
3618          */
3619         if (unlikely(rc == -EEXIST)) {
3620                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3621                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3622                 if (rc < 0)
3623                         RETURN(rc);
3624
3625                 if (rc < sizeof(*lma))
3626                         RETURN(-EINVAL);
3627
3628                 lustre_loa_swab(loa, true);
3629                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3630                     ((compat == 0 && incompat == 0) ||
3631                      (!(~lma->lma_compat & compat) &&
3632                       !(~lma->lma_incompat & incompat))))
3633                         RETURN(0);
3634
3635                 lma->lma_self_fid = *fid;
3636                 lma->lma_compat |= compat;
3637                 lma->lma_incompat |= incompat;
3638                 if (rc == sizeof(*lma)) {
3639                         lustre_lma_swab(lma);
3640                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3641                                              sizeof(*lma), XATTR_REPLACE);
3642                 } else {
3643                         lustre_loa_swab(loa, false);
3644                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3645                                              sizeof(*loa), XATTR_REPLACE);
3646                 }
3647         }
3648
3649         RETURN(rc);
3650 }
3651
3652 /**
3653  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3654  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3655  * To have compatilibility with 1.8 ldiskfs driver we need to have
3656  * magic number at start of fid data.
3657  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3658  * its inmemory API.
3659  */
3660 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3661                                          const struct lu_fid *fid)
3662 {
3663         if (!fid_is_namespace_visible(fid) ||
3664             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3665                 param->edp_magic = 0;
3666                 return;
3667         }
3668
3669         param->edp_magic = LDISKFS_LUFID_MAGIC;
3670         param->edp_len =  sizeof(struct lu_fid) + 1;
3671         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3672 }
3673
3674 /**
3675  * Try to read the fid from inode ea into dt_rec.
3676  *
3677  * \param fid object fid.
3678  *
3679  * \retval 0 on success
3680  */
3681 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3682                           __u32 ino, struct lu_fid *fid,
3683                           struct osd_inode_id *id)
3684 {
3685         struct osd_thread_info *info  = osd_oti_get(env);
3686         struct inode *inode;
3687
3688         ENTRY;
3689
3690         osd_id_gen(id, ino, OSD_OII_NOGEN);
3691         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3692         if (IS_ERR(inode))
3693                 RETURN(PTR_ERR(inode));
3694
3695         iput(inode);
3696         RETURN(0);
3697 }
3698
3699 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3700                                         struct inode *dir,
3701                                         struct inode *parent_dir,
3702                                         const struct lu_fid *dot_fid,
3703                                         const struct lu_fid *dot_dot_fid,
3704                                         struct osd_thandle *oth)
3705 {
3706         struct ldiskfs_dentry_param *dot_ldp;
3707         struct ldiskfs_dentry_param *dot_dot_ldp;
3708         __u32 saved_nlink = dir->i_nlink;
3709         int rc;
3710
3711         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3712         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3713
3714         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3715         dot_ldp->edp_magic = 0;
3716
3717         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3718                                     dir, dot_ldp, dot_dot_ldp);
3719         /*
3720          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3721          * the subseqent ref_add() will increase the dir->i_nlink
3722          * as 3. That is incorrect for new created directory.
3723          *
3724          * It looks like hack, because we want to make the OSD API
3725          * to be order-independent for new created directory object
3726          * between dt_insert(..) and ref_add() operations.
3727          *
3728          * Here, we only restore the in-RAM dir-inode's nlink attr,
3729          * becuase if the nlink attr is not 2, then there will be
3730          * ref_add() called following the dt_insert(..), such call
3731          * will make both the in-RAM and on-disk dir-inode's nlink
3732          * attr to be set as 2. LU-7447
3733          */
3734         set_nlink(dir, saved_nlink);
3735         return rc;
3736 }
3737
3738 /**
3739  * Create an local agent inode for remote entry
3740  */
3741 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3742                                                   struct osd_device *osd,
3743                                                   struct osd_object *pobj,
3744                                                   const struct lu_fid *fid,
3745                                                   __u32 type,
3746                                                   struct thandle *th)
3747 {
3748         struct osd_thread_info *info = osd_oti_get(env);
3749         struct inode *local;
3750         struct osd_thandle *oh;
3751         int rc;
3752
3753         ENTRY;
3754
3755         LASSERT(th);
3756         oh = container_of(th, struct osd_thandle, ot_super);
3757         LASSERT(oh->ot_handle->h_transaction != NULL);
3758
3759         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3760         if (IS_ERR(local)) {
3761                 CERROR("%s: create local error %d\n", osd_name(osd),
3762                        (int)PTR_ERR(local));
3763                 RETURN(local);
3764         }
3765
3766         /*
3767          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3768          * correct gid on remote file, not agent here
3769          */
3770         local->i_gid = current_fsgid();
3771         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3772
3773         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3774          * That gives e2fsck something to look at and be happy, and allows
3775          * debugging if we need to determine where this symlink came from.
3776          */
3777         if (S_ISLNK(type)) {
3778                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3779                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3780                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3781
3782                 i_size_write(local, rc);
3783                 LDISKFS_I(local)->i_disksize = rc;
3784         }
3785         unlock_new_inode(local);
3786
3787         /* Agent inode should not have project ID */
3788 #ifdef  HAVE_PROJECT_QUOTA
3789         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3790             i_projid_read(pobj->oo_inode) != 0) {
3791                 rc = osd_transfer_project(local, 0);
3792                 if (rc) {
3793                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3794                                "quota enforcement enabled on the ldiskfs "
3795                                "filesystem?\n", local->i_sb->s_id, rc);
3796                         RETURN(ERR_PTR(rc));
3797                 }
3798         }
3799 #endif
3800         /* Set special LMA flag for local agent inode */
3801         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3802         if (rc != 0) {
3803                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3804                        osd_name(osd), PFID(fid), rc);
3805                 RETURN(ERR_PTR(rc));
3806         }
3807
3808         if (!S_ISDIR(type))
3809                 RETURN(local);
3810
3811         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3812                                          lu_object_fid(&pobj->oo_dt.do_lu),
3813                                          fid, oh);
3814         if (rc != 0) {
3815                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3816                         osd_name(osd), PFID(fid), rc);
3817                 RETURN(ERR_PTR(rc));
3818         }
3819
3820         RETURN(local);
3821 }
3822
3823 /**
3824  * when direntry is deleted, we have to take care of possible agent inode
3825  * referenced by that. unfortunately we can't do this at that point:
3826  * iget() within a running transaction leads to deadlock and we better do
3827  * not call that every delete declaration to save performance. so we put
3828  * a potention agent inode on a list and process that once the transaction
3829  * is over. Notice it's not any worse in terms of real orphans as regular
3830  * object destroy doesn't put inodes on the on-disk orphan list. this should
3831  * be addressed separately
3832  */
3833 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3834                                             struct osd_thandle *oh,
3835                                             __u32 ino)
3836 {
3837         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3838         struct osd_obj_orphan *oor;
3839
3840         OBD_ALLOC_PTR(oor);
3841         if (oor == NULL)
3842                 return -ENOMEM;
3843
3844         oor->oor_ino = ino;
3845         oor->oor_env = (struct lu_env *)env;
3846         spin_lock(&osd->od_osfs_lock);
3847         list_add(&oor->oor_list, &osd->od_orphan_list);
3848         spin_unlock(&osd->od_osfs_lock);
3849
3850         oh->ot_remove_agents = 1;
3851
3852         return 0;
3853
3854 }
3855
3856 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3857                                                 struct osd_device *osd)
3858 {
3859         struct osd_thread_info *info = osd_oti_get(env);
3860         struct osd_obj_orphan *oor, *tmp;
3861         struct osd_inode_id id;
3862         struct list_head list;
3863         struct inode *inode;
3864         struct lu_fid fid;
3865         handle_t *jh;
3866         __u32 ino;
3867
3868         INIT_LIST_HEAD(&list);
3869
3870         spin_lock(&osd->od_osfs_lock);
3871         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3872                 if (oor->oor_env == env) {
3873                         list_del(&oor->oor_list);
3874                         list_add(&oor->oor_list, &list);
3875                 }
3876         }
3877         spin_unlock(&osd->od_osfs_lock);
3878
3879         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3880
3881                 ino = oor->oor_ino;
3882
3883                 list_del(&oor->oor_list);
3884                 OBD_FREE_PTR(oor);
3885
3886                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3887                 inode = osd_iget_fid(info, osd, &id, &fid);
3888                 if (IS_ERR(inode))
3889                         continue;
3890
3891                 if (!osd_remote_fid(env, osd, &fid)) {
3892                         iput(inode);
3893                         continue;
3894                 }
3895
3896                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3897                 clear_nlink(inode);
3898                 mark_inode_dirty(inode);
3899                 ldiskfs_journal_stop(jh);
3900                 iput(inode);
3901         }
3902
3903         return 0;
3904 }
3905
3906 /**
3907  * OSD layer object create function for OST objects (b=11826).
3908  *
3909  * The FID is inserted into inode xattr here.
3910  *
3911  * \retval   0, on success
3912  * \retval -ve, on error
3913  */
3914 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3915                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3916                       struct dt_object_format *dof, struct thandle *th)
3917 {
3918         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3919         struct osd_object *obj = osd_dt_obj(dt);
3920         struct osd_thread_info *info = osd_oti_get(env);
3921         int result, on_ost = 0;
3922
3923         ENTRY;
3924
3925         if (dt_object_exists(dt))
3926                 RETURN(-EEXIST);
3927
3928         LINVRNT(osd_invariant(obj));
3929         LASSERT(!dt_object_remote(dt));
3930         LASSERT(osd_is_write_locked(env, obj));
3931         LASSERT(th != NULL);
3932
3933         if (unlikely(fid_is_acct(fid)))
3934                 /*
3935                  * Quota files can't be created from the kernel any more,
3936                  * 'tune2fs -O quota' will take care of creating them
3937                  */
3938                 RETURN(-EPERM);
3939
3940         result = __osd_create(info, obj, attr, hint, dof, th);
3941         if (result == 0) {
3942                 if (fid_is_idif(fid) &&
3943                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3944                         struct lu_fid *tfid = &info->oti_fid;
3945                         struct ost_id *oi   = &info->oti_ostid;
3946
3947                         fid_to_ostid(fid, oi);
3948                         ostid_to_fid(tfid, oi, 0);
3949                         on_ost = 1;
3950                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3951                                                 LMAC_FID_ON_OST, 0);
3952                 } else {
3953                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3954                                                fid, OI_CHECK_FLD);
3955                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3956                                                 on_ost ? LMAC_FID_ON_OST : 0,
3957                                                 0);
3958                 }
3959                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3960                         obj->oo_dt.do_body_ops = &osd_body_ops;
3961         }
3962
3963         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3964                 result = __osd_oi_insert(env, obj, fid, th);
3965
3966         /*
3967          * a small optimization - dt_insert() isn't usually applied
3968          * to OST objects, so we don't need to cache OI mapping for
3969          * OST objects
3970          */
3971         if (result == 0 && on_ost == 0) {
3972                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3973
3974                 result = osd_idc_find_and_init(env, osd, obj);
3975                 LASSERT(result == 0);
3976         }
3977
3978         LASSERT(ergo(result == 0,
3979                      dt_object_exists(dt) && !dt_object_remote(dt)));
3980         LINVRNT(osd_invariant(obj));
3981         RETURN(result);
3982 }
3983
3984 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3985                                struct thandle *handle)
3986 {
3987         struct osd_thandle *oh;
3988
3989         /* it's possible that object doesn't exist yet */
3990         LASSERT(handle != NULL);
3991
3992         oh = container_of0(handle, struct osd_thandle, ot_super);
3993         LASSERT(oh->ot_handle == NULL);
3994
3995         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3996                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3997
3998         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3999
4000         return 0;
4001 }
4002
4003 /*
4004  * Concurrency: @dt is write locked.
4005  */
4006 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4007                        struct thandle *th)
4008 {
4009         struct osd_object *obj = osd_dt_obj(dt);
4010         struct inode *inode = obj->oo_inode;
4011         struct osd_thandle *oh;
4012         int rc = 0;
4013
4014         if (!dt_object_exists(dt) || obj->oo_destroyed)
4015                 return -ENOENT;
4016
4017         LINVRNT(osd_invariant(obj));
4018         LASSERT(!dt_object_remote(dt));
4019         LASSERT(osd_is_write_locked(env, obj));
4020         LASSERT(th != NULL);
4021
4022         oh = container_of0(th, struct osd_thandle, ot_super);
4023         LASSERT(oh->ot_handle != NULL);
4024
4025         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4026
4027         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4028                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4029         /*
4030          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4031          * (65000) subdirectories by storing "1" in i_nlink if the link count
4032          * would otherwise overflow. Directory tranversal tools understand
4033          * that (st_nlink == 1) indicates that the filesystem dose not track
4034          * hard links count on the directory, and will not abort subdirectory
4035          * scanning early once (st_nlink - 2) subdirs have been found.
4036          *
4037          * This also has to properly handle the case of inodes with nlink == 0
4038          * in case they are being linked into the PENDING directory
4039          */
4040         spin_lock(&obj->oo_guard);
4041         if (unlikely(inode->i_nlink == 0))
4042                 /* inc_nlink from 0 may cause WARN_ON */
4043                 set_nlink(inode, 1);
4044         else {
4045                 ldiskfs_inc_count(oh->ot_handle, inode);
4046                 if (!S_ISDIR(inode->i_mode))
4047                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4048         }
4049         spin_unlock(&obj->oo_guard);
4050
4051         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4052         LINVRNT(osd_invariant(obj));
4053
4054         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4055
4056         return rc;
4057 }
4058
4059 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4060                                struct thandle *handle)
4061 {
4062         struct osd_thandle *oh;
4063
4064         if (!dt_object_exists(dt))
4065                 return -ENOENT;
4066
4067         LASSERT(!dt_object_remote(dt));
4068         LASSERT(handle != NULL);
4069
4070         oh = container_of0(handle, struct osd_thandle, ot_super);
4071         LASSERT(oh->ot_handle == NULL);
4072
4073         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4074                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4075
4076         return 0;
4077 }
4078
4079 /*
4080  * Concurrency: @dt is write locked.
4081  */
4082 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4083                        struct thandle *th)
4084 {
4085         struct osd_object *obj = osd_dt_obj(dt);
4086         struct inode *inode = obj->oo_inode;
4087         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4088         struct osd_thandle *oh;
4089
4090         if (!dt_object_exists(dt))
4091                 return -ENOENT;
4092
4093         LINVRNT(osd_invariant(obj));
4094         LASSERT(!dt_object_remote(dt));
4095         LASSERT(osd_is_write_locked(env, obj));
4096         LASSERT(th != NULL);
4097
4098         oh = container_of0(th, struct osd_thandle, ot_super);
4099         LASSERT(oh->ot_handle != NULL);
4100
4101         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4102
4103         spin_lock(&obj->oo_guard);
4104         /*
4105          * That can be result of upgrade from old Lustre version and
4106          * applied only to local files.  Just skip this ref_del call.
4107          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4108          */
4109         if (inode->i_nlink == 0) {
4110                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4111                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4112                              ", maybe an upgraded file? (LU-3915)\n",
4113                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4114                 spin_unlock(&obj->oo_guard);
4115                 return 0;
4116         }
4117
4118         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4119                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4120
4121         ldiskfs_dec_count(oh->ot_handle, inode);
4122         spin_unlock(&obj->oo_guard);
4123
4124         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4125         LINVRNT(osd_invariant(obj));
4126
4127         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4128
4129         return 0;
4130 }
4131
4132 /*
4133  * Concurrency: @dt is read locked.
4134  */
4135 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4136                          struct lu_buf *buf, const char *name)
4137 {
4138         struct osd_object      *obj    = osd_dt_obj(dt);
4139         struct inode           *inode  = obj->oo_inode;
4140         struct osd_thread_info *info   = osd_oti_get(env);
4141         struct dentry          *dentry = &info->oti_obj_dentry;
4142         bool cache_xattr = false;
4143         int rc;
4144
4145         LASSERT(buf);
4146
4147         /* version get is not real XATTR but uses xattr API */
4148         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4149                 dt_obj_version_t *ver = buf->lb_buf;
4150
4151                 /*
4152                  * for version we are just using xattr API but change inode
4153                  * field instead
4154                  */
4155                 if (buf->lb_len == 0)
4156                         return sizeof(dt_obj_version_t);
4157
4158                 if (buf->lb_len < sizeof(dt_obj_version_t))
4159                         return -ERANGE;
4160
4161                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4162                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4163
4164                 *ver = LDISKFS_I(inode)->i_fs_version;
4165
4166                 return sizeof(dt_obj_version_t);
4167         }
4168
4169         if (!dt_object_exists(dt))
4170                 return -ENOENT;
4171
4172         LASSERT(!dt_object_remote(dt));
4173         LASSERT(inode->i_op != NULL);
4174 #ifdef HAVE_IOP_XATTR
4175         LASSERT(inode->i_op->getxattr != NULL);
4176 #endif
4177
4178         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4179             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4180                 cache_xattr = true;
4181
4182         if (cache_xattr) {
4183                 rc = osd_oxc_get(obj, name, buf);
4184                 if (rc != -ENOENT)
4185                         return rc;
4186         }
4187
4188         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4189                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4190                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4191                 struct filter_fid *ff;
4192                 struct ost_layout *ol;
4193
4194                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4195                 if (rc)
4196                         return rc;
4197
4198                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4199
4200                 rc = sizeof(*ff);
4201                 if (buf->lb_len == 0 || !buf->lb_buf)
4202                         return rc;
4203
4204                 if (buf->lb_len < rc)
4205                         return -ERANGE;
4206
4207                 ff = buf->lb_buf;
4208                 ol = &ff->ff_layout;
4209                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4210                                                   PFID_STRIPE_IDX_BITS);
4211                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4212                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4213                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4214                 if (lma->lma_compat & LMAC_COMP_INFO) {
4215                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4216                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4217                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4218                 } else {
4219                         ol->ol_comp_start = 0;
4220                         ol->ol_comp_end = 0;
4221                         ol->ol_comp_id = 0;
4222                 }
4223         } else {
4224                 rc = __osd_xattr_get(inode, dentry, name,
4225                                      buf->lb_buf, buf->lb_len);
4226         }
4227
4228         if (cache_xattr) {
4229                 if (rc == -ENOENT || rc == -ENODATA)
4230                         osd_oxc_add(obj, name, NULL, 0);
4231                 else if (rc > 0 && buf->lb_buf != NULL)
4232                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4233         }
4234
4235         return rc;
4236 }
4237
4238 static int osd_declare_xattr_set(const struct lu_env *env,
4239                                  struct dt_object *dt,
4240                                  const struct lu_buf *buf, const char *name,
4241                                  int fl, struct thandle *handle)
4242 {
4243         struct osd_thandle *oh;
4244         int credits = 0;
4245         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4246
4247         LASSERT(handle != NULL);
4248
4249         oh = container_of0(handle, struct osd_thandle, ot_super);
4250         LASSERT(oh->ot_handle == NULL);
4251
4252         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4253                 /*
4254                  * For non-upgrading case, the LMA is set first and
4255                  * usually fit inode. But for upgrade case, the LMA
4256                  * may be in another separated EA block.
4257                  */
4258                 if (dt_object_exists(dt)) {
4259                         if (fl == LU_XATTR_REPLACE)
4260                                 credits = 1;
4261                         else
4262                                 goto upgrade;
4263                 }
4264         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4265                 credits = 1;
4266         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4267                 /* We may need to delete the old PFID EA. */
4268                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4269                 if (fl == LU_XATTR_REPLACE)
4270                         credits += 1;
4271                 else
4272                         goto upgrade;
4273         } else {
4274                 /*
4275                  * If some name entry resides on remote MDT, then will create
4276                  * agent entry under remote parent. On the other hand, if the
4277                  * remote entry will be removed, then related agent entry may
4278                  * need to be removed from the remote parent. So there may be
4279                  * kinds of cases, let's declare enough credits. The credits
4280                  * for create agent entry is enough for remove case.
4281                  */
4282                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4283                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4284                         if (dt_object_exists(dt))
4285                                 credits += 1; /* For updating LMA */
4286                 }
4287
4288 upgrade:
4289                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4290
4291                 if (buf != NULL) {
4292                         ssize_t buflen;
4293
4294                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4295                                 /*
4296                                  * learn xattr size from osd_xattr_get if
4297                                  * attribute has not been read yet
4298                                  */
4299                                 buflen = __osd_xattr_get(
4300                                     osd_dt_obj(dt)->oo_inode,
4301                                     &osd_oti_get(env)->oti_obj_dentry,
4302                                     name, NULL, 0);
4303                                 if (buflen < 0)
4304                                         buflen = 0;
4305                         } else {
4306                                 buflen = buf->lb_len;
4307                         }
4308
4309                         if (buflen > sb->s_blocksize) {
4310                                 credits += osd_calc_bkmap_credits(
4311                                     sb, NULL, 0, -1,
4312                                     (buflen + sb->s_blocksize - 1) >>
4313                                     sb->s_blocksize_bits);
4314                         }
4315                 }
4316                 /*
4317                  * xattr set may involve inode quota change, reserve credits for
4318                  * dquot_initialize()
4319                  */
4320                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4321         }
4322
4323         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4324
4325         return 0;
4326 }
4327
4328 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4329                               const struct lu_buf *buf, int fl,
4330                               struct thandle *handle)
4331 {
4332         struct osd_thread_info *info = osd_oti_get(env);
4333         struct dentry *dentry = &info->oti_obj_dentry;
4334         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4335         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4336         struct inode *inode = obj->oo_inode;
4337         struct filter_fid *ff = buf->lb_buf;
4338         struct ost_layout *ol = &ff->ff_layout;
4339         int flags = XATTR_REPLACE;
4340         int rc;
4341
4342         ENTRY;
4343
4344         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4345                 RETURN(-EINVAL);
4346
4347         rc = osd_get_lma(info, inode, dentry, loa);
4348         if (rc == -ENODATA) {
4349                 /* Usually for upgarding from old device */
4350                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4351                                 LMAC_FID_ON_OST, 0);
4352                 flags = XATTR_CREATE;
4353         } else if (rc) {
4354                 RETURN(rc);
4355         }
4356
4357         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4358                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4359                         RETURN(-EEXIST);
4360
4361                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4362                         /* Separate PFID EA from LMA */
4363                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4364                         lustre_lma_swab(lma);
4365                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4366                                              sizeof(*lma), XATTR_REPLACE);
4367                         if (!rc) {
4368                                 obj->oo_pfid_in_lma = 0;
4369                                 rc = LU_XATTR_CREATE;
4370                         }
4371
4372                         RETURN(rc);
4373                 }
4374         } else {
4375                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4376                         RETURN(fl);
4377
4378                 /*
4379                  * Old client does not send stripe information,
4380                  * then store the PFID EA on disk separatedly.
4381                  */
4382                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4383                              ol->ol_stripe_size == 0))
4384                         RETURN(fl);
4385
4386                 /* Remove old PFID EA entry firstly. */
4387                 ll_vfs_dq_init(inode);
4388                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4389                 if (rc == -ENODATA) {
4390                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4391                                 RETURN(rc);
4392                 } else if (rc) {
4393                         RETURN(rc);
4394                 }
4395         }
4396
4397         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4398         if (likely(ol->ol_stripe_size != 0)) {
4399                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4400                                              PFID_STRIPE_IDX_BITS;
4401                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4402                 lma->lma_compat |= LMAC_STRIPE_INFO;
4403                 if (ol->ol_comp_id != 0) {
4404                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4405                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4406                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4407                         lma->lma_compat |= LMAC_COMP_INFO;
4408                 }
4409         }
4410
4411         lustre_loa_swab(loa, false);
4412
4413         /* Store the PFID EA inside LMA. */
4414         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4415                              flags);
4416         if (!rc)
4417                 obj->oo_pfid_in_lma = 1;
4418
4419         RETURN(rc);
4420 }
4421
4422 /*
4423  * In DNE environment, the object (in spite of regular file or directory)
4424  * and its name entry may reside on different MDTs. Under such case, we will
4425  * create an agent entry on the MDT where the object resides. The agent entry
4426  * references the object locally, that makes the object to be visible to the
4427  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4428  * such as 'tar' can handle the object properly.
4429  *
4430  * We handle the agent entry during set linkEA that is the common interface
4431  * for both regular file and directroy, can handle kinds of cases, such as
4432  * create/link/unlink/rename, and so on.
4433  *
4434  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4435  *
4436  * XXX: There are two known issues:
4437  * 1. For one object, we will create at most one agent entry even if there
4438  *    may be more than one cross-MDTs hard links on the object. So the local
4439  *    e2fsck may claim that the object's nlink is larger than the name entries
4440  *    that reference such inode. And in further, the e2fsck will fix the nlink
4441  *    attribute to match the local references. Then it will cause the object's
4442  *    nlink attribute to be inconsistent with the global references. it is bad
4443  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4444  *    other hand, the global namespace LFSCK can repair the object's attribute
4445  *    according to the linkEA.
4446  * 2. There may be too many hard links on the object as to its linkEA overflow,
4447  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4448  *    case happened, then at this point, we do not know whether there are some
4449  *    cross-MDTs reference. But there are local references, it guarantees that
4450  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4451  */
4452 static int osd_xattr_handle_linkea(const struct lu_env *env,
4453                                    struct osd_device *osd,
4454                                    struct osd_object *obj,
4455                                    const struct lu_buf *buf,
4456                                    struct thandle *handle)
4457 {
4458         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4459         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4460         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4461         struct lu_name tmpname;
4462         struct osd_thandle *oh;
4463         int rc;
4464         bool remote = false;
4465
4466         ENTRY;
4467
4468         oh = container_of0(handle, struct osd_thandle, ot_super);
4469         LASSERT(oh->ot_handle != NULL);
4470
4471         rc = linkea_init_with_rec(&ldata);
4472         if (!rc) {
4473                 linkea_first_entry(&ldata);
4474                 while (ldata.ld_lee != NULL && !remote) {
4475                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4476                                             &tmpname, tfid);
4477                         if (osd_remote_fid(env, osd, tfid) > 0)
4478                                 remote = true;
4479                         else
4480                                 linkea_next_entry(&ldata);
4481                 }
4482         } else if (rc == -ENODATA) {
4483                 rc = 0;
4484         } else {
4485                 RETURN(rc);
4486         }
4487
4488         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4489                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4490                 if (rc)
4491                         CERROR("%s: failed to remove agent entry for "DFID
4492                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4493         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4494                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4495                 if (rc)
4496                         CERROR("%s: failed to create agent entry for "DFID
4497                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4498         }
4499
4500         RETURN(rc);
4501 }
4502
4503 /*
4504  * Concurrency: @dt is write locked.
4505  */
4506 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4507                          const struct lu_buf *buf, const char *name, int fl,
4508                          struct thandle *handle)
4509 {
4510         struct osd_object *obj = osd_dt_obj(dt);
4511         struct osd_device *osd = osd_obj2dev(obj);
4512         struct inode *inode = obj->oo_inode;
4513         struct osd_thread_info *info = osd_oti_get(env);
4514         int fs_flags = 0;
4515         int len;
4516         int rc;
4517
4518         ENTRY;
4519
4520         LASSERT(handle);
4521         LASSERT(buf);
4522
4523         /* version set is not real XATTR */
4524         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4525                 dt_obj_version_t *version = buf->lb_buf;
4526
4527                 /*
4528                  * for version we are just using xattr API but change inode
4529                  * field instead
4530                  */
4531                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4532
4533                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4534                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4535
4536                 LDISKFS_I(inode)->i_fs_version = *version;
4537                 /*
4538                  * Version is set after all inode operations are finished,
4539                  * so we should mark it dirty here
4540                  */
4541                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4542
4543                 RETURN(0);
4544         }
4545
4546         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4547                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4548
4549         len = buf->lb_len;
4550         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4551
4552         /*
4553          * For the OST device with 256 bytes inode size by default,
4554          * the PFID EA will be stored together with LMA EA to avoid
4555          * performance trouble. Otherwise the PFID EA can be stored
4556          * independently. LU-8998
4557          */
4558         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4559             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4560                 LASSERT(buf->lb_buf);
4561
4562                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4563                 if (fl <= 0)
4564                         RETURN(fl);
4565         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4566                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4567                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4568
4569                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4570                 if (rc)
4571                         RETURN(rc);
4572
4573                 lma->lma_incompat |= LMAI_STRIPED;
4574                 lustre_lma_swab(lma);
4575                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4576                                      sizeof(*lma), XATTR_REPLACE);
4577                 if (rc != 0)
4578                         RETURN(rc);
4579         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4580                 LASSERT(!osd->od_is_ost);
4581
4582                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4583                 if (rc)
4584                         RETURN(rc);
4585         }
4586
4587         if (fl & LU_XATTR_REPLACE)
4588                 fs_flags |= XATTR_REPLACE;
4589
4590         if (fl & LU_XATTR_CREATE)
4591                 fs_flags |= XATTR_CREATE;
4592
4593         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4594         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4595
4596         if (rc == 0 &&
4597             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4598              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4599                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4600
4601         return rc;
4602 }
4603
4604 /*
4605  * Concurrency: @dt is read locked.
4606  */
4607 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4608                           const struct lu_buf *buf)
4609 {
4610         struct osd_object *obj = osd_dt_obj(dt);
4611         struct inode *inode = obj->oo_inode;
4612         struct osd_thread_info *info = osd_oti_get(env);
4613         struct dentry *dentry = &info->oti_obj_dentry;
4614
4615         if (!dt_object_exists(dt))
4616                 return -ENOENT;
4617
4618         LASSERT(!dt_object_remote(dt));
4619         LASSERT(inode->i_op != NULL);
4620         LASSERT(inode->i_op->listxattr != NULL);
4621
4622         dentry->d_inode = inode;
4623         dentry->d_sb = inode->i_sb;
4624         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4625 }
4626
4627 static int osd_declare_xattr_del(const struct lu_env *env,
4628                                  struct dt_object *dt, const char *name,
4629                                  struct thandle *handle)
4630 {
4631         struct osd_thandle *oh;
4632         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4633
4634         LASSERT(!dt_object_remote(dt));
4635         LASSERT(handle != NULL);
4636
4637         oh = container_of0(handle, struct osd_thandle, ot_super);
4638         LASSERT(oh->ot_handle == NULL);
4639
4640         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4641                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4642         /*
4643          * xattr del may involve inode quota change, reserve credits for
4644          * dquot_initialize()
4645          */
4646         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4647
4648         return 0;
4649 }
4650
4651 /*
4652  * Concurrency: @dt is write locked.
4653  */
4654 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4655                          const char *name, struct thandle *handle)
4656 {
4657         struct osd_object *obj = osd_dt_obj(dt);
4658         struct inode *inode = obj->oo_inode;
4659         struct osd_thread_info *info = osd_oti_get(env);
4660         struct dentry *dentry = &info->oti_obj_dentry;
4661         int rc;
4662
4663         if (!dt_object_exists(dt))
4664                 return -ENOENT;
4665
4666         LASSERT(!dt_object_remote(dt));
4667         LASSERT(inode->i_op != NULL);
4668         LASSERT(handle != NULL);
4669 #ifdef HAVE_IOP_XATTR
4670         LASSERT(inode->i_op->removexattr != NULL);
4671 #endif
4672
4673         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4674
4675         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4676                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4677
4678                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4679                                  &info->oti_ost_attrs);
4680                 if (!rc) {
4681                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4682
4683                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4684                         lustre_lma_swab(lma);
4685                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4686                                              sizeof(*lma), XATTR_REPLACE);
4687                         if (!rc)
4688                                 obj->oo_pfid_in_lma = 0;
4689                 }
4690         } else {
4691                 ll_vfs_dq_init(inode);
4692                 dentry->d_inode = inode;
4693                 dentry->d_sb = inode->i_sb;
4694                 rc = osd_removexattr(dentry, inode, name);
4695         }
4696
4697         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4698
4699         if (rc == 0 &&
4700             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4701              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4702                 osd_oxc_del(obj, name);
4703
4704         return rc;
4705 }
4706
4707 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4708                            __u64 start, __u64 end)
4709 {
4710         struct osd_object *obj = osd_dt_obj(dt);
4711         struct inode *inode = obj->oo_inode;
4712         struct osd_thread_info *info = osd_oti_get(env);
4713         struct dentry *dentry = &info->oti_obj_dentry;
4714         struct file *file = &info->oti_file;
4715         int rc;
4716
4717         ENTRY;
4718
4719         dentry->d_inode = inode;
4720         dentry->d_sb = inode->i_sb;
4721         file->f_path.dentry = dentry;
4722         file->f_mapping = inode->i_mapping;
4723         file->f_op = inode->i_fop;
4724         set_file_inode(file, inode);
4725
4726         rc = ll_vfs_fsync_range(file, start, end, 0);
4727
4728         RETURN(rc);
4729 }
4730
4731 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4732 {
4733         return 0;
4734 }
4735
4736 /*
4737  * Index operations.
4738  */
4739
4740 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4741                                const struct dt_index_features *feat)
4742 {
4743         struct iam_descr *descr;
4744
4745         if (osd_object_is_root(o))
4746                 return feat == &dt_directory_features;
4747
4748         LASSERT(o->oo_dir != NULL);
4749
4750         descr = o->oo_dir->od_container.ic_descr;
4751         if (feat == &dt_directory_features) {
4752                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4753                         return 1;
4754                 else
4755                         return 0;
4756         } else {
4757                 return feat->dif_keysize_min <= descr->id_key_size &&
4758                        descr->id_key_size <= feat->dif_keysize_max &&
4759                        feat->dif_recsize_min <= descr->id_rec_size &&
4760                        descr->id_rec_size <= feat->dif_recsize_max &&
4761                        !(feat->dif_flags & (DT_IND_VARKEY |
4762                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4763                        ergo(feat->dif_flags & DT_IND_UPDATE,
4764                             1 /* XXX check that object (and fs) is writable */);
4765         }
4766 }
4767
4768 static int osd_iam_container_init(const struct lu_env *env,
4769                                   struct osd_object *obj,
4770                                   struct osd_directory *dir)
4771 {
4772         struct iam_container *bag = &dir->od_container;
4773         int result;
4774
4775         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4776         if (result != 0)
4777                 return result;
4778
4779         result = iam_container_setup(bag);
4780         if (result == 0)
4781                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4782         else
4783                 iam_container_fini(bag);
4784
4785         return result;
4786 }
4787
4788
4789 /*
4790  * Concurrency: no external locking is necessary.
4791  */
4792 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4793                          const struct dt_index_features *feat)
4794 {
4795         int result;
4796         int skip_iam = 0;
4797         struct osd_object *obj = osd_dt_obj(dt);
4798
4799         LINVRNT(osd_invariant(obj));
4800
4801         if (osd_object_is_root(obj)) {
4802                 dt->do_index_ops = &osd_index_ea_ops;
4803                 result = 0;
4804         } else if (feat == &dt_directory_features) {
4805                 dt->do_index_ops = &osd_index_ea_ops;
4806                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4807                         result = 0;
4808                 else
4809                         result = -ENOTDIR;
4810                 skip_iam = 1;
4811         } else if (unlikely(feat == &dt_otable_features)) {
4812                 dt->do_index_ops = &osd_otable_ops;
4813                 return 0;
4814         } else if (unlikely(feat == &dt_acct_features)) {
4815                 dt->do_index_ops = &osd_acct_index_ops;
4816                 result = 0;
4817                 skip_iam = 1;
4818         } else if (!osd_has_index(obj)) {
4819                 struct osd_directory *dir;
4820                 struct osd_device *osd = osd_obj2dev(obj);
4821                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4822
4823                 OBD_ALLOC_PTR(dir);
4824                 if (dir) {
4825
4826                         spin_lock(&obj->oo_guard);
4827                         if (obj->oo_dir == NULL)
4828                                 obj->oo_dir = dir;
4829                         else
4830                                 /*
4831                                  * Concurrent thread allocated container data.
4832                                  */
4833                                 OBD_FREE_PTR(dir);
4834                         spin_unlock(&obj->oo_guard);
4835                         /*
4836                          * Now, that we have container data, serialize its
4837                          * initialization.
4838                          */
4839                         down_write(&obj->oo_ext_idx_sem);
4840                         /*
4841                          * recheck under lock.
4842                          */
4843
4844                         if (osd_has_index(obj)) {
4845                                 result = 0;
4846                                 goto unlock;
4847                         }
4848
4849                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4850                         if (result || feat == &dt_lfsck_namespace_features ||
4851                             feat == &dt_lfsck_layout_orphan_features ||
4852                             feat == &dt_lfsck_layout_dangling_features)
4853                                 goto unlock;
4854
4855                         result = osd_index_register(osd, fid,
4856                                                     feat->dif_keysize_max,
4857                                                     feat->dif_recsize_max);
4858                         if (result < 0)
4859                                 CWARN("%s: failed to register index "
4860                                       DFID": rc = %d\n",
4861                                       osd_name(osd), PFID(fid), result);
4862                         else if (result > 0)
4863                                 result = 0;
4864                         else
4865                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4866                                        " (%d/%d) registered\n",
4867                                        osd_name(osd), PFID(fid),
4868                                        (int)feat->dif_keysize_max,
4869                                        (int)feat->dif_recsize_max);
4870
4871 unlock:
4872                         up_write(&obj->oo_ext_idx_sem);
4873                 } else {
4874                         result = -ENOMEM;
4875                 }
4876         } else {
4877                 result = 0;
4878         }
4879
4880         if (result == 0 && skip_iam == 0) {
4881                 if (!osd_iam_index_probe(env, obj, feat))
4882                         result = -ENOTDIR;
4883         }
4884         LINVRNT(osd_invariant(obj));
4885
4886         return result;
4887 }
4888
4889 static int osd_otable_it_attr_get(const struct lu_env *env,
4890                                  struct dt_object *dt,
4891                                  struct lu_attr *attr)
4892 {
4893         attr->la_valid = 0;
4894         return 0;
4895 }
4896
4897 static const struct dt_object_operations osd_obj_ops = {
4898         .do_read_lock           = osd_read_lock,
4899         .do_write_lock          = osd_write_lock,
4900         .do_read_unlock         = osd_read_unlock,
4901         .do_write_unlock        = osd_write_unlock,
4902         .do_write_locked        = osd_write_locked,
4903         .do_attr_get            = osd_attr_get,
4904         .do_declare_attr_set    = osd_declare_attr_set,
4905         .do_attr_set            = osd_attr_set,
4906         .do_ah_init             = osd_ah_init,
4907         .do_declare_create      = osd_declare_create,
4908         .do_create              = osd_create,
4909         .do_declare_destroy     = osd_declare_destroy,
4910         .do_destroy             = osd_destroy,
4911         .do_index_try           = osd_index_try,
4912         .do_declare_ref_add     = osd_declare_ref_add,
4913         .do_ref_add             = osd_ref_add,
4914         .do_declare_ref_del     = osd_declare_ref_del,
4915         .do_ref_del             = osd_ref_del,
4916         .do_xattr_get           = osd_xattr_get,
4917         .do_declare_xattr_set   = osd_declare_xattr_set,
4918         .do_xattr_set           = osd_xattr_set,
4919         .do_declare_xattr_del   = osd_declare_xattr_del,
4920         .do_xattr_del           = osd_xattr_del,
4921         .do_xattr_list          = osd_xattr_list,
4922         .do_object_sync         = osd_object_sync,
4923         .do_invalidate          = osd_invalidate,
4924 };
4925
4926 static const struct dt_object_operations osd_obj_otable_it_ops = {
4927         .do_attr_get    = osd_otable_it_attr_get,
4928         .do_index_try   = osd_index_try,
4929 };
4930
4931 static int osd_index_declare_iam_delete(const struct lu_env *env,
4932                                         struct dt_object *dt,
4933                                         const struct dt_key *key,
4934                                         struct thandle *handle)
4935 {
4936         struct osd_thandle *oh;
4937
4938         oh = container_of0(handle, struct osd_thandle, ot_super);
4939         LASSERT(oh->ot_handle == NULL);
4940
4941         /* Recycle  may cause additional three blocks to be changed. */
4942         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4943                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4944
4945         return 0;
4946 }
4947
4948 /**
4949  *      delete a (key, value) pair from index \a dt specified by \a key
4950  *
4951  *      \param  dt      osd index object
4952  *      \param  key     key for index
4953  *      \param  rec     record reference
4954  *      \param  handle  transaction handler
4955  *
4956  *      \retval  0  success
4957  *      \retval -ve   failure
4958  */
4959 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4960                                 const struct dt_key *key,
4961                                 struct thandle *handle)
4962 {
4963         struct osd_thread_info *oti = osd_oti_get(env);
4964         struct osd_object *obj = osd_dt_obj(dt);
4965         struct osd_thandle *oh;
4966         struct iam_path_descr *ipd;
4967         struct iam_container *bag = &obj->oo_dir->od_container;
4968         int rc;
4969
4970         ENTRY;
4971
4972         if (!dt_object_exists(dt))
4973                 RETURN(-ENOENT);
4974
4975         LINVRNT(osd_invariant(obj));
4976         LASSERT(!dt_object_remote(dt));
4977         LASSERT(bag->ic_object == obj->oo_inode);
4978         LASSERT(handle != NULL);
4979
4980         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4981
4982         ipd = osd_idx_ipd_get(env, bag);
4983         if (unlikely(ipd == NULL))
4984                 RETURN(-ENOMEM);
4985
4986         oh = container_of0(handle, struct osd_thandle, ot_super);
4987         LASSERT(oh->ot_handle != NULL);
4988         LASSERT(oh->ot_handle->h_transaction != NULL);
4989
4990         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4991                 /* swab quota uid/gid provided by caller */
4992                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4993                 key = (const struct dt_key *)&oti->oti_quota_id;
4994         }
4995
4996         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4997         osd_ipd_put(env, bag, ipd);
4998         LINVRNT(osd_invariant(obj));
4999         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5000         RETURN(rc);
5001 }
5002
5003 static int osd_index_declare_ea_delete(const struct lu_env *env,
5004                                        struct dt_object *dt,
5005                                        const struct dt_key *key,
5006                                        struct thandle *handle)
5007 {
5008         struct osd_thandle *oh;
5009         struct inode *inode;
5010         int rc, credits;
5011
5012         ENTRY;
5013
5014         LASSERT(!dt_object_remote(dt));
5015         LASSERT(handle != NULL);
5016
5017         oh = container_of0(handle, struct osd_thandle, ot_super);
5018         LASSERT(oh->ot_handle == NULL);
5019
5020         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5021         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5022
5023         inode = osd_dt_obj(dt)->oo_inode;
5024         if (inode == NULL)
5025                 RETURN(-ENOENT);
5026
5027         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5028                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5029                                    NULL, OSD_QID_BLK);
5030         RETURN(rc);
5031 }
5032
5033 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5034                                           struct dt_rec *fid)
5035 {
5036         struct osd_fid_pack *rec;
5037         int rc = -ENODATA;
5038
5039         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5040                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5041                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5042                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5043                         rc = -EINVAL;
5044         }
5045         return rc;
5046 }
5047
5048 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5049                           const struct lu_fid *fid)
5050 {
5051         struct seq_server_site *ss = osd_seq_site(osd);
5052
5053         ENTRY;
5054
5055         /* FID seqs not in FLDB, must be local seq */
5056         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5057                 RETURN(0);
5058
5059         /*
5060          * If FLD is not being initialized yet, it only happens during the
5061          * initialization, likely during mgs initialization, and we assume
5062          * this is local FID.
5063          */
5064         if (ss == NULL || ss->ss_server_fld == NULL)
5065                 RETURN(0);
5066
5067         /* Only check the local FLDB here */
5068         if (osd_seq_exists(env, osd, fid_seq(fid)))
5069                 RETURN(0);
5070
5071         RETURN(1);
5072 }
5073
5074 static void osd_take_care_of_agent(const struct lu_env *env,
5075                                    struct osd_device *osd,
5076                                    struct osd_thandle *oh,
5077                                    struct ldiskfs_dir_entry_2 *de)
5078 {
5079         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5080         struct osd_idmap_cache *idc;
5081         int rc, schedule = 0;
5082
5083         LASSERT(de != NULL);
5084
5085         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5086         if (likely(rc == 0)) {
5087                 idc = osd_idc_find_or_init(env, osd, fid);
5088                 if (IS_ERR(idc) || idc->oic_remote)
5089                         schedule = 1;
5090         } else if (rc == -ENODATA) {
5091                 /*
5092                  * can't get FID, postpone to the end of the
5093                  * transaction when iget() is safe
5094                  */
5095                 schedule = 1;
5096         } else {
5097                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5098         }
5099         if (schedule)
5100                 osd_schedule_agent_inode_removal(env, oh,
5101                                                  le32_to_cpu(de->inode));
5102 }
5103
5104 /**
5105  * Index delete function for interoperability mode (b11826).
5106  * It will remove the directory entry added by osd_index_ea_insert().
5107  * This entry is needed to maintain name->fid mapping.
5108  *
5109  * \param key,  key i.e. file entry to be deleted
5110  *
5111  * \retval   0, on success
5112  * \retval -ve, on error
5113  */
5114 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5115                                const struct dt_key *key, struct thandle *handle)
5116 {
5117         struct osd_object *obj = osd_dt_obj(dt);
5118         struct inode *dir = obj->oo_inode;
5119         struct dentry *dentry;
5120         struct osd_thandle *oh;
5121         struct ldiskfs_dir_entry_2 *de = NULL;
5122         struct buffer_head *bh;
5123         struct htree_lock *hlock = NULL;
5124         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5125         int rc;
5126
5127         ENTRY;
5128
5129         if (!dt_object_exists(dt))
5130                 RETURN(-ENOENT);
5131
5132         LINVRNT(osd_invariant(obj));
5133         LASSERT(!dt_object_remote(dt));
5134         LASSERT(handle != NULL);
5135
5136         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5137
5138         oh = container_of(handle, struct osd_thandle, ot_super);
5139         LASSERT(oh->ot_handle != NULL);
5140         LASSERT(oh->ot_handle->h_transaction != NULL);
5141
5142         ll_vfs_dq_init(dir);
5143         dentry = osd_child_dentry_get(env, obj,
5144                                       (char *)key, strlen((char *)key));
5145
5146         if (obj->oo_hl_head != NULL) {
5147                 hlock = osd_oti_get(env)->oti_hlock;
5148                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5149                                    dir, LDISKFS_HLOCK_DEL);
5150         } else {
5151                 down_write(&obj->oo_ext_idx_sem);
5152         }
5153
5154         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5155         if (!IS_ERR(bh)) {
5156                 /*
5157                  * If this is not the ".." entry, it might be a remote DNE
5158                  * entry and  we need to check if the FID is for a remote
5159                  * MDT.  If the FID is  not in the directory entry (e.g.
5160                  * upgraded 1.8 filesystem without dirdata enabled) then
5161                  * we need to get the FID from the LMA. For a remote directory
5162                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5163                  * case.
5164                  *
5165                  * Delete the entry before the agent inode in order to
5166                  * simplify error handling.  At worst an error after deleting
5167                  * the entry first might leak the agent inode afterward. The
5168                  * reverse would need filesystem abort in case of error deleting
5169                  * the entry after the agent had been removed, or leave a
5170                  * dangling entry pointing at a random inode.
5171                  */
5172                 if (strcmp((char *)key, dotdot) != 0)
5173                         osd_take_care_of_agent(env, osd, oh, de);
5174                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5175                 brelse(bh);
5176         } else {
5177                 rc = PTR_ERR(bh);
5178         }
5179         if (hlock != NULL)
5180                 ldiskfs_htree_unlock(hlock);
5181         else
5182                 up_write(&obj->oo_ext_idx_sem);
5183
5184         GOTO(out, rc);
5185 out:
5186         LASSERT(osd_invariant(obj));
5187         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5188         RETURN(rc);
5189 }
5190
5191 /**
5192  *      Lookup index for \a key and copy record to \a rec.
5193  *
5194  *      \param  dt      osd index object
5195  *      \param  key     key for index
5196  *      \param  rec     record reference
5197  *
5198  *      \retval  +ve  success : exact mach
5199  *      \retval  0    return record with key not greater than \a key
5200  *      \retval -ve   failure
5201  */
5202 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5203                                 struct dt_rec *rec, const struct dt_key *key)
5204 {
5205         struct osd_object *obj = osd_dt_obj(dt);
5206         struct iam_path_descr *ipd;
5207         struct iam_container *bag = &obj->oo_dir->od_container;
5208         struct osd_thread_info *oti = osd_oti_get(env);
5209         struct iam_iterator *it = &oti->oti_idx_it;
5210         struct iam_rec *iam_rec;
5211         int rc;
5212
5213         ENTRY;
5214
5215         if (!dt_object_exists(dt))
5216                 RETURN(-ENOENT);
5217
5218         LASSERT(osd_invariant(obj));
5219         LASSERT(!dt_object_remote(dt));
5220         LASSERT(bag->ic_object == obj->oo_inode);
5221
5222         ipd = osd_idx_ipd_get(env, bag);
5223         if (IS_ERR(ipd))
5224                 RETURN(-ENOMEM);
5225
5226         /* got ipd now we can start iterator. */
5227         iam_it_init(it, bag, 0, ipd);
5228
5229         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5230                 /* swab quota uid/gid provided by caller */
5231                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5232                 key = (const struct dt_key *)&oti->oti_quota_id;
5233         }
5234
5235         rc = iam_it_get(it, (struct iam_key *)key);
5236         if (rc >= 0) {
5237                 if (S_ISDIR(obj->oo_inode->i_mode))
5238                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5239                 else
5240                         iam_rec = (struct iam_rec *)rec;
5241
5242                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5243
5244                 if (S_ISDIR(obj->oo_inode->i_mode))
5245                         osd_fid_unpack((struct lu_fid *)rec,
5246                                        (struct osd_fid_pack *)iam_rec);
5247                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5248                         osd_quota_unpack(obj, rec);
5249         }
5250
5251         iam_it_put(it);
5252         iam_it_fini(it);
5253         osd_ipd_put(env, bag, ipd);
5254
5255         LINVRNT(osd_invariant(obj));
5256
5257         RETURN(rc);
5258 }
5259
5260 static int osd_index_declare_iam_insert(const struct lu_env *env,
5261                                         struct dt_object *dt,
5262                                         const struct dt_rec *rec,
5263                                         const struct dt_key *key,
5264                                         struct thandle *handle)
5265 {
5266         struct osd_thandle *oh;
5267
5268         LASSERT(handle != NULL);
5269
5270         oh = container_of0(handle, struct osd_thandle, ot_super);
5271         LASSERT(oh->ot_handle == NULL);
5272
5273         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5274                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5275
5276         return 0;
5277 }
5278
5279 /**
5280  *      Inserts (key, value) pair in \a dt index object.
5281  *
5282  *      \param  dt      osd index object
5283  *      \param  key     key for index
5284  *      \param  rec     record reference
5285  *      \param  th      transaction handler
5286  *
5287  *      \retval  0  success
5288  *      \retval -ve failure
5289  */
5290 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5291                                 const struct dt_rec *rec,
5292                                 const struct dt_key *key, struct thandle *th)
5293 {
5294         struct osd_object *obj = osd_dt_obj(dt);
5295         struct iam_path_descr *ipd;
5296         struct osd_thandle *oh;
5297         struct iam_container *bag;
5298         struct osd_thread_info *oti = osd_oti_get(env);
5299         struct iam_rec *iam_rec;
5300         int rc;
5301
5302         ENTRY;
5303
5304         if (!dt_object_exists(dt))
5305                 RETURN(-ENOENT);
5306
5307         LINVRNT(osd_invariant(obj));
5308         LASSERT(!dt_object_remote(dt));
5309
5310         bag = &obj->oo_dir->od_container;
5311         LASSERT(bag->ic_object == obj->oo_inode);
5312         LASSERT(th != NULL);
5313
5314         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5315
5316         ipd = osd_idx_ipd_get(env, bag);
5317         if (unlikely(ipd == NULL))
5318                 RETURN(-ENOMEM);
5319
5320         oh = container_of0(th, struct osd_thandle, ot_super);
5321         LASSERT(oh->ot_handle != NULL);
5322         LASSERT(oh->ot_handle->h_transaction != NULL);
5323         if (S_ISDIR(obj->oo_inode->i_mode)) {
5324                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5325                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5326                              &oti->oti_fid);
5327         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5328                 /* pack quota uid/gid */
5329                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5330                 key = (const struct dt_key *)&oti->oti_quota_id;
5331                 /* pack quota record */
5332                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5333                 iam_rec = (struct iam_rec *)rec;
5334         } else {
5335                 iam_rec = (struct iam_rec *)rec;
5336         }
5337
5338         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5339                         iam_rec, ipd);
5340         osd_ipd_put(env, bag, ipd);
5341         LINVRNT(osd_invariant(obj));
5342         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5343         RETURN(rc);
5344 }
5345
5346 /**
5347  * Calls ldiskfs_add_entry() to add directory entry
5348  * into the directory. This is required for
5349  * interoperability mode (b11826)
5350  *
5351  * \retval   0, on success
5352  * \retval -ve, on error
5353  */
5354 static int __osd_ea_add_rec(struct osd_thread_info *info,
5355                             struct osd_object *pobj, struct inode  *cinode,
5356                             const char *name, const struct lu_fid *fid,
5357                             struct htree_lock *hlock, struct thandle *th)
5358 {
5359         struct ldiskfs_dentry_param *ldp;
5360         struct dentry *child;
5361         struct osd_thandle *oth;
5362         int rc;
5363
5364         oth = container_of(th, struct osd_thandle, ot_super);
5365         LASSERT(oth->ot_handle != NULL);
5366         LASSERT(oth->ot_handle->h_transaction != NULL);
5367         LASSERT(pobj->oo_inode);
5368
5369         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5370         if (unlikely(pobj->oo_inode ==
5371                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5372                 ldp->edp_magic = 0;
5373         else
5374                 osd_get_ldiskfs_dirent_param(ldp, fid);
5375         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5376         child->d_fsdata = (void *)ldp;
5377         ll_vfs_dq_init(pobj->oo_inode);
5378         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5379                                    child, cinode, hlock);
5380         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5381                 struct ldiskfs_dir_entry_2      *de;
5382                 struct buffer_head              *bh;
5383                 int                              rc1;
5384
5385                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5386                                             NULL, hlock);
5387                 if (!IS_ERR(bh)) {
5388                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5389                                                                bh);
5390                         if (rc1 == 0) {
5391                                 if (S_ISDIR(cinode->i_mode))
5392                                         de->file_type = LDISKFS_DIRENT_LUFID |
5393                                                         LDISKFS_FT_REG_FILE;
5394                                 else
5395                                         de->file_type = LDISKFS_DIRENT_LUFID |
5396                                                         LDISKFS_FT_DIR;
5397                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5398                                                               NULL, bh);
5399                         }
5400                         brelse(bh);
5401                 }
5402         }
5403
5404         RETURN(rc);
5405 }
5406
5407 /**
5408  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5409  * into the directory.Also sets flags into osd object to
5410  * indicate dot and dotdot are created. This is required for
5411  * interoperability mode (b11826)
5412  *
5413  * \param dir   directory for dot and dotdot fixup.
5414  * \param obj   child object for linking
5415  *
5416  * \retval   0, on success
5417  * \retval -ve, on error
5418  */
5419 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5420                               struct osd_object *dir,
5421                               struct inode *parent_dir, const char *name,
5422                               const struct lu_fid *dot_fid,
5423                               const struct lu_fid *dot_dot_fid,
5424                               struct thandle *th)
5425 {
5426         struct inode *inode = dir->oo_inode;
5427         struct osd_thandle *oth;
5428         int result = 0;
5429
5430         oth = container_of(th, struct osd_thandle, ot_super);
5431         LASSERT(oth->ot_handle->h_transaction != NULL);
5432         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5433
5434         if (strcmp(name, dot) == 0) {
5435                 if (dir->oo_compat_dot_created) {
5436                         result = -EEXIST;
5437                 } else {
5438                         LASSERT(inode->i_ino == parent_dir->i_ino);
5439                         dir->oo_compat_dot_created = 1;
5440                         result = 0;
5441                 }
5442         } else if (strcmp(name, dotdot) == 0) {
5443                 if (!dir->oo_compat_dot_created)
5444                         return -EINVAL;
5445                 /* in case of rename, dotdot is already created */
5446                 if (dir->oo_compat_dotdot_created) {
5447                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5448                                                 dot_dot_fid, NULL, th);
5449                 }
5450
5451                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5452                         struct lu_fid tfid = *dot_dot_fid;
5453
5454                         tfid.f_oid--;
5455                         result = osd_add_dot_dotdot_internal(info,
5456                                         dir->oo_inode, parent_dir, dot_fid,
5457                                         &tfid, oth);
5458                 } else {
5459                         result = osd_add_dot_dotdot_internal(info,
5460                                         dir->oo_inode, parent_dir, dot_fid,
5461                                         dot_dot_fid, oth);
5462                 }
5463
5464                 if (result == 0)
5465                         dir->oo_compat_dotdot_created = 1;
5466         }
5467
5468         return result;
5469 }
5470
5471
5472 /**
5473  * It will call the appropriate osd_add* function and return the
5474  * value, return by respective functions.
5475  */
5476 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5477                           struct inode *cinode, const char *name,
5478                           const struct lu_fid *fid, struct thandle *th)
5479 {
5480         struct osd_thread_info *info = osd_oti_get(env);
5481         struct htree_lock *hlock;
5482         int rc;
5483
5484         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5485
5486         if (name[0] == '.' && (name[1] == '\0' ||
5487                                (name[1] == '.' && name[2] == '\0'))) {
5488                 if (hlock != NULL) {
5489                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5490                                            pobj->oo_inode, 0);
5491                 } else {
5492                         down_write(&pobj->oo_ext_idx_sem);
5493                 }
5494
5495                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5496                                         lu_object_fid(&pobj->oo_dt.do_lu),
5497                                         fid, th);
5498         } else {
5499                 if (hlock != NULL) {
5500                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5501                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5502                 } else {
5503                         down_write(&pobj->oo_ext_idx_sem);
5504                 }
5505
5506                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5507                         struct lu_fid *tfid = &info->oti_fid;
5508
5509                         *tfid = *fid;
5510                         tfid->f_ver = ~0;
5511                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5512                                               tfid, hlock, th);
5513                 } else {
5514                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5515                                               hlock, th);
5516                 }
5517         }
5518         if (hlock != NULL)
5519                 ldiskfs_htree_unlock(hlock);
5520         else
5521                 up_write(&pobj->oo_ext_idx_sem);
5522
5523         return rc;
5524 }
5525
5526 static int
5527 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5528                       struct osd_idmap_cache *oic)
5529 {
5530         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5531         struct lu_fid *fid = &oic->oic_fid;
5532         struct osd_inode_id *id = &oic->oic_lid;
5533         struct inode *inode = NULL;
5534         int once = 0;
5535         bool insert;
5536         int rc;
5537
5538         ENTRY;
5539
5540         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5541                 RETURN(0);
5542
5543         if (thread_is_running(&scrub->os_thread) &&
5544             scrub->os_pos_current > id->oii_ino)
5545                 RETURN(0);
5546
5547         if (dev->od_auto_scrub_interval == AS_NEVER ||
5548             ktime_get_real_seconds() <
5549             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5550                 RETURN(0);
5551
5552 again:
5553         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5554         if (rc == -ENOENT) {
5555                 __u32 gen = id->oii_gen;
5556
5557                 insert = true;
5558                 if (inode != NULL)
5559                         goto trigger;
5560
5561                 inode = osd_iget(oti, dev, id);
5562                 /* The inode has been removed (by race maybe). */
5563                 if (IS_ERR(inode)) {
5564                         rc = PTR_ERR(inode);
5565
5566                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5567                 }
5568
5569                 /* The OI mapping is lost. */
5570                 if (gen != OSD_OII_NOGEN)
5571                         goto trigger;
5572
5573                 /*
5574                  * The inode may has been reused by others, we do not know,
5575                  * leave it to be handled by subsequent osd_fid_lookup().
5576                  */
5577                 GOTO(out, rc = 0);
5578         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5579                 GOTO(out, rc);
5580         }
5581
5582         insert = false;
5583
5584 trigger:
5585         if (thread_is_running(&scrub->os_thread)) {
5586                 if (inode == NULL) {
5587                         inode = osd_iget(oti, dev, id);
5588                         /* The inode has been removed (by race maybe). */
5589                         if (IS_ERR(inode)) {
5590                                 rc = PTR_ERR(inode);
5591
5592                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5593                         }
5594                 }
5595
5596                 rc = osd_oii_insert(dev, oic, insert);
5597                 /*
5598                  * There is race condition between osd_oi_lookup and OI scrub.
5599                  * The OI scrub finished just after osd_oi_lookup() failure.
5600                  * Under such case, it is unnecessary to trigger OI scrub again,
5601                  * but try to call osd_oi_lookup() again.
5602                  */
5603                 if (unlikely(rc == -EAGAIN))
5604                         goto again;
5605
5606                 if (!S_ISDIR(inode->i_mode))
5607                         rc = 0;
5608                 else
5609                         rc = osd_check_lmv(oti, dev, inode, oic);
5610
5611                 GOTO(out, rc);
5612         }
5613
5614         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5615                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5616                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5617                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5618                        "%s: trigger partial OI scrub for RPC inconsistency "
5619                        "checking FID "DFID": rc = %d\n",
5620                        osd_dev2name(dev), PFID(fid), rc);
5621                 if (rc == 0 || rc == -EALREADY)
5622                         goto again;
5623         }
5624
5625         GOTO(out, rc);
5626
5627 out:
5628         if (inode)
5629                 iput(inode);
5630
5631         RETURN(rc);
5632 }
5633
5634 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5635                                struct osd_device *dev,
5636                                struct osd_idmap_cache *oic,
5637                                struct lu_fid *fid, __u32 ino)
5638 {
5639         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5640         struct inode *inode;
5641         int rc;
5642
5643         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5644         inode = osd_iget(oti, dev, &oic->oic_lid);
5645         if (IS_ERR(inode)) {
5646                 fid_zero(&oic->oic_fid);
5647                 return PTR_ERR(inode);
5648         }
5649
5650         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5651         iput(inode);
5652         if (rc != 0)
5653                 fid_zero(&oic->oic_fid);
5654         else
5655                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5656         return rc;
5657 }
5658
5659 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5660                       struct osd_inode_id *id, const struct lu_fid *fid)
5661 {
5662         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5663                id->oii_ino, id->oii_gen, info);
5664         info->oti_cache.oic_lid = *id;
5665         info->oti_cache.oic_fid = *fid;
5666         info->oti_cache.oic_dev = osd;
5667 }
5668
5669 /**
5670  * Get parent FID from the linkEA.
5671  *
5672  * For a directory which parent resides on remote MDT, to satisfy the
5673  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5674  * the other hand, to make the lookup(..) on the directory can return
5675  * the real parent FID, we append the real parent FID after its ".."
5676  * name entry in the /REMOTE_PARENT_DIR.
5677  *
5678  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5679  * backup. So after the restore, we cannot get the right parent FID from
5680  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5681  * we have stored the real parent FID in the directory object's linkEA,
5682  * we can parse the linkEA for the real parent FID.
5683  *
5684  * \param[in] env       pointer to the thread context
5685  * \param[in] obj       pointer to the object to be handled
5686  * \param[out]fid       pointer to the buffer to hold the parent FID
5687  *
5688  * \retval              0 for getting the real parent FID successfully
5689  * \retval              negative error number on failure
5690  */
5691 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5692                                     struct osd_object *obj,
5693                                     struct lu_fid *fid)
5694 {
5695         struct osd_thread_info *oti = osd_oti_get(env);
5696         struct lu_buf *buf = &oti->oti_big_buf;
5697         struct dentry *dentry = &oti->oti_obj_dentry;
5698         struct inode *inode = obj->oo_inode;
5699         struct linkea_data ldata = { NULL };
5700         int rc;
5701
5702         ENTRY;
5703
5704         fid_zero(fid);
5705         if (!S_ISDIR(inode->i_mode))
5706                 RETURN(-EIO);
5707
5708 again:
5709         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5710                              buf->lb_buf, buf->lb_len);
5711         if (rc == -ERANGE) {
5712                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5713                                      NULL, 0);
5714                 if (rc > 0) {
5715                         lu_buf_realloc(buf, rc);
5716                         if (buf->lb_buf == NULL)
5717                                 RETURN(-ENOMEM);
5718
5719                         goto again;
5720                 }
5721         }
5722
5723         if (unlikely(rc == 0))
5724                 RETURN(-ENODATA);
5725
5726         if (rc < 0)
5727                 RETURN(rc);
5728
5729         if (unlikely(buf->lb_buf == NULL)) {
5730                 lu_buf_realloc(buf, rc);
5731                 if (buf->lb_buf == NULL)
5732                         RETURN(-ENOMEM);
5733
5734                 goto again;
5735         }
5736
5737         ldata.ld_buf = buf;
5738         rc = linkea_init_with_rec(&ldata);
5739         if (!rc) {
5740                 linkea_first_entry(&ldata);
5741                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5742         }
5743
5744         RETURN(rc);
5745 }
5746
5747 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5748                                     struct inode *inode,
5749                                     const struct lu_fid *pfid,
5750                                     const char *name, const int namelen)
5751 {
5752         struct osd_thread_info *oti = osd_oti_get(env);
5753         struct lu_buf *buf = &oti->oti_big_buf;
5754         struct dentry *dentry = &oti->oti_obj_dentry;
5755         struct linkea_data ldata = { NULL };
5756         struct lu_name cname = { .ln_name = name,
5757                                  .ln_namelen = namelen };
5758         int rc;
5759
5760         ENTRY;
5761
5762 again:
5763         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5764                              buf->lb_buf, buf->lb_len);
5765         if (rc == -ERANGE)
5766                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5767
5768         if (rc < 0)
5769                 RETURN(rc);
5770
5771         if (unlikely(rc == 0))
5772                 RETURN(-ENODATA);
5773
5774         if (buf->lb_len < rc) {
5775                 lu_buf_realloc(buf, rc);
5776                 if (buf->lb_buf == NULL)
5777                         RETURN(-ENOMEM);
5778
5779                 goto again;
5780         }
5781
5782         ldata.ld_buf = buf;
5783         rc = linkea_init_with_rec(&ldata);
5784         if (!rc)
5785                 rc = linkea_links_find(&ldata, &cname, pfid);
5786
5787         RETURN(rc);
5788 }
5789
5790 /**
5791  * Calls ->lookup() to find dentry. From dentry get inode and
5792  * read inode's ea to get fid. This is required for  interoperability
5793  * mode (b11826)
5794  *
5795  * \retval   0, on success
5796  * \retval -ve, on error
5797  */
5798 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5799                              struct dt_rec *rec, const struct dt_key *key)
5800 {
5801         struct inode *dir = obj->oo_inode;
5802         struct dentry *dentry;
5803         struct ldiskfs_dir_entry_2 *de;
5804         struct buffer_head *bh;
5805         struct lu_fid *fid = (struct lu_fid *)rec;
5806         struct htree_lock *hlock = NULL;
5807         int ino;
5808         int rc;
5809
5810         ENTRY;
5811
5812         LASSERT(dir->i_op != NULL);
5813         LASSERT(dir->i_op->lookup != NULL);
5814
5815         dentry = osd_child_dentry_get(env, obj,
5816                                       (char *)key, strlen((char *)key));
5817
5818         if (obj->oo_hl_head != NULL) {
5819                 hlock = osd_oti_get(env)->oti_hlock;
5820                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5821                                    dir, LDISKFS_HLOCK_LOOKUP);
5822         } else {
5823                 down_read(&obj->oo_ext_idx_sem);
5824         }
5825
5826         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5827         if (!IS_ERR(bh)) {
5828                 struct osd_thread_info *oti = osd_oti_get(env);
5829                 struct osd_inode_id *id = &oti->oti_id;
5830                 struct osd_idmap_cache *oic = &oti->oti_cache;
5831                 struct osd_device *dev = osd_obj2dev(obj);
5832
5833                 ino = le32_to_cpu(de->inode);
5834                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5835                         brelse(bh);
5836                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5837                         GOTO(out, rc);
5838                 }
5839
5840                 rc = osd_get_fid_from_dentry(de, rec);
5841
5842                 /* done with de, release bh */
5843                 brelse(bh);
5844                 if (rc != 0) {
5845                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5846                                 const char *name = (const char *)key;
5847
5848                                 /*
5849                                  * If the parent is on remote MDT, and there
5850                                  * is no FID-in-dirent, then we have to get
5851                                  * the parent FID from the linkEA.
5852                                  */
5853                                 if (likely(strlen(name) == 2 &&
5854                                            name[0] == '.' && name[1] == '.'))
5855                                         rc = osd_get_pfid_from_linkea(env, obj,
5856                                                                       fid);
5857                         } else {
5858                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5859                         }
5860                 } else {
5861                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5862                 }
5863
5864                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5865                         fid_zero(&oic->oic_fid);
5866
5867                         GOTO(out, rc);
5868                 }
5869
5870                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5871                 rc = osd_consistency_check(oti, dev, oic);
5872                 if (rc == -ENOENT)
5873                         fid_zero(&oic->oic_fid);
5874                 else
5875                         /* Other error should not affect lookup result. */
5876                         rc = 0;
5877         } else {
5878                 rc = PTR_ERR(bh);
5879         }
5880
5881         GOTO(out, rc);
5882
5883 out:
5884         if (hlock != NULL)
5885                 ldiskfs_htree_unlock(hlock);
5886         else
5887                 up_read(&obj->oo_ext_idx_sem);
5888         return rc;
5889 }
5890
5891 static int osd_index_declare_ea_insert(const struct lu_env *env,
5892                                        struct dt_object *dt,
5893                                        const struct dt_rec *rec,
5894                                        const struct dt_key *key,
5895                                        struct thandle *handle)
5896 {
5897         struct osd_thandle *oh;
5898         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5899         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5900         const struct lu_fid *fid = rec1->rec_fid;
5901         int credits, rc = 0;
5902         struct osd_idmap_cache *idc;
5903
5904         ENTRY;
5905
5906         LASSERT(!dt_object_remote(dt));
5907         LASSERT(handle != NULL);
5908         LASSERT(fid != NULL);
5909         LASSERT(rec1->rec_type != 0);
5910
5911         oh = container_of0(handle, struct osd_thandle, ot_super);
5912         LASSERT(oh->ot_handle == NULL);
5913
5914         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5915
5916         /*
5917          * we can't call iget() while a transactions is running
5918          * (this can lead to a deadlock), but we need to know
5919          * inum and object type. so we find this information at
5920          * declaration and cache in per-thread info
5921          */
5922         idc = osd_idc_find_or_init(env, osd, fid);
5923         if (IS_ERR(idc))
5924                 RETURN(PTR_ERR(idc));
5925         if (idc->oic_remote) {
5926                 /*
5927                  * a reference to remote inode is represented by an
5928                  * agent inode which we have to create
5929                  */
5930                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5931                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5932         }
5933
5934         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5935
5936         if (osd_dt_obj(dt)->oo_inode != NULL) {
5937                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5938
5939                 /*
5940                  * We ignore block quota on meta pool (MDTs), so needn't
5941                  * calculate how many blocks will be consumed by this index
5942                  * insert
5943                  */
5944                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5945                                            i_gid_read(inode),
5946                                            i_projid_read(inode), 0,
5947                                            oh, osd_dt_obj(dt), NULL,
5948                                            OSD_QID_BLK);
5949                 if (rc)
5950                         RETURN(rc);
5951
5952 #ifdef HAVE_PROJECT_QUOTA
5953                 /*
5954                  * Reserve credits for local agent inode to transfer
5955                  * to 0, quota enforcement is ignored in this case.
5956                  */
5957                 if (idc->oic_remote &&
5958                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5959                     i_projid_read(inode) != 0)
5960                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5961                                                   0, i_projid_read(inode),
5962                                                   0, false, PRJQUOTA, true);
5963 #endif
5964         }
5965
5966         RETURN(rc);
5967 }
5968
5969 /**
5970  * Index add function for interoperability mode (b11826).
5971  * It will add the directory entry.This entry is needed to
5972  * maintain name->fid mapping.
5973  *
5974  * \param key it is key i.e. file entry to be inserted
5975  * \param rec it is value of given key i.e. fid
5976  *
5977  * \retval   0, on success
5978  * \retval -ve, on error
5979  */
5980 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5981                                const struct dt_rec *rec,
5982                                const struct dt_key *key, struct thandle *th)
5983 {
5984         struct osd_object *obj = osd_dt_obj(dt);
5985         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5986         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5987         const struct lu_fid *fid = rec1->rec_fid;
5988         const char *name = (const char *)key;
5989         struct osd_thread_info *oti = osd_oti_get(env);
5990         struct inode *child_inode = NULL;
5991         struct osd_idmap_cache *idc;
5992         int rc;
5993
5994         ENTRY;
5995
5996         if (!dt_object_exists(dt))
5997                 RETURN(-ENOENT);
5998
5999         LASSERT(osd_invariant(obj));
6000         LASSERT(!dt_object_remote(dt));
6001         LASSERT(th != NULL);
6002
6003         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6004
6005         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6006
6007         idc = osd_idc_find(env, osd, fid);
6008         if (unlikely(idc == NULL)) {
6009                 idc = osd_idc_find_or_init(env, osd, fid);
6010                 if (IS_ERR(idc)) {
6011                         /*
6012                          * this dt_insert() wasn't declared properly, so
6013                          * FID is missing in OI cache. we better do not
6014                          * lookup FID in FLDB/OI and don't risk to deadlock,
6015                          * but in some special cases (lfsck testing, etc)
6016                          * it's much simpler than fixing a caller.
6017                          *
6018                          * normally this error should be placed after the first
6019                          * find, but migrate may attach source stripes to
6020                          * target, which doesn't create stripes.
6021                          */
6022                         CERROR("%s: "DFID" wasn't declared for insert\n",
6023                                osd_name(osd), PFID(fid));
6024                         dump_stack();
6025                         RETURN(PTR_ERR(idc));
6026                 }
6027         }
6028
6029         if (idc->oic_remote) {
6030                 /* Insert remote entry */
6031                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6032                         child_inode =
6033                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6034                 } else {
6035                         child_inode = osd_create_local_agent_inode(env, osd,
6036                                         obj, fid, rec1->rec_type & S_IFMT, th);
6037                         if (IS_ERR(child_inode))
6038                                 RETURN(PTR_ERR(child_inode));
6039                 }
6040         } else {
6041                 /* Insert local entry */
6042                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6043                         /* for a reason OI cache wasn't filled properly */
6044                         CERROR("%s: OIC for "DFID" isn't filled\n",
6045                                osd_name(osd), PFID(fid));
6046                         RETURN(-EINVAL);
6047                 }
6048                 child_inode = oti->oti_inode;
6049                 if (unlikely(child_inode == NULL)) {
6050                         struct ldiskfs_inode_info *lii;
6051
6052                         OBD_ALLOC_PTR(lii);
6053                         if (lii == NULL)
6054                                 RETURN(-ENOMEM);
6055                         child_inode = oti->oti_inode = &lii->vfs_inode;
6056                 }
6057                 child_inode->i_sb = osd_sb(osd);
6058                 child_inode->i_ino = idc->oic_lid.oii_ino;
6059                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6060         }
6061
6062         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6063
6064         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6065                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6066
6067         if (child_inode && child_inode != oti->oti_inode)
6068                 iput(child_inode);
6069         LASSERT(osd_invariant(obj));
6070         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6071         RETURN(rc);
6072 }
6073
6074 /**
6075  *  Initialize osd Iterator for given osd index object.
6076  *
6077  *  \param  dt      osd index object
6078  */
6079
6080 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6081                                      struct dt_object *dt,
6082                                      __u32 unused)
6083 {
6084         struct osd_it_iam *it;
6085         struct osd_object *obj = osd_dt_obj(dt);
6086         struct lu_object *lo = &dt->do_lu;
6087         struct iam_path_descr *ipd;
6088         struct iam_container *bag = &obj->oo_dir->od_container;
6089
6090         if (!dt_object_exists(dt))
6091                 return ERR_PTR(-ENOENT);
6092
6093         OBD_ALLOC_PTR(it);
6094         if (it == NULL)
6095                 return ERR_PTR(-ENOMEM);
6096
6097         ipd = osd_it_ipd_get(env, bag);
6098         if (likely(ipd != NULL)) {
6099                 it->oi_obj = obj;
6100                 it->oi_ipd = ipd;
6101                 lu_object_get(lo);
6102                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6103                 return (struct dt_it *)it;
6104         } else {
6105                 OBD_FREE_PTR(it);
6106                 return ERR_PTR(-ENOMEM);
6107         }
6108 }
6109
6110 /**
6111  * free given Iterator.
6112  */
6113 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6114 {
6115         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6116         struct osd_object *obj = it->oi_obj;
6117
6118         iam_it_fini(&it->oi_it);
6119         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6120         osd_object_put(env, obj);
6121         OBD_FREE_PTR(it);
6122 }
6123
6124 /**
6125  *  Move Iterator to record specified by \a key
6126  *
6127  *  \param  di      osd iterator
6128  *  \param  key     key for index
6129  *
6130  *  \retval +ve  di points to record with least key not larger than key
6131  *  \retval  0   di points to exact matched key
6132  *  \retval -ve  failure
6133  */
6134
6135 static int osd_it_iam_get(const struct lu_env *env,
6136                           struct dt_it *di, const struct dt_key *key)
6137 {
6138         struct osd_thread_info *oti = osd_oti_get(env);
6139         struct osd_it_iam *it = (struct osd_it_iam *)di;
6140
6141         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6142                 /* swab quota uid/gid */
6143                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6144                 key = (struct dt_key *)&oti->oti_quota_id;
6145         }
6146
6147         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6148 }
6149
6150 /**
6151  *  Release Iterator
6152  *
6153  *  \param  di      osd iterator
6154  */
6155 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6156 {
6157         struct osd_it_iam *it = (struct osd_it_iam *)di;
6158
6159         iam_it_put(&it->oi_it);
6160 }
6161
6162 /**
6163  *  Move iterator by one record
6164  *
6165  *  \param  di      osd iterator
6166  *
6167  *  \retval +1   end of container reached
6168  *  \retval  0   success
6169  *  \retval -ve  failure
6170  */
6171
6172 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6173 {
6174         struct osd_it_iam *it = (struct osd_it_iam *)di;
6175
6176         return iam_it_next(&it->oi_it);
6177 }
6178
6179 /**
6180  * Return pointer to the key under iterator.
6181  */
6182
6183 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6184                                      const struct dt_it *di)
6185 {
6186         struct osd_thread_info *oti = osd_oti_get(env);
6187         struct osd_it_iam *it = (struct osd_it_iam *)di;
6188         struct osd_object *obj = it->oi_obj;
6189         struct dt_key *key;
6190
6191         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6192
6193         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6194                 /* swab quota uid/gid */
6195                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6196                 key = (struct dt_key *)&oti->oti_quota_id;
6197         }
6198
6199         return key;
6200 }
6201
6202 /**
6203  * Return size of key under iterator (in bytes)
6204  */
6205
6206 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6207 {
6208         struct osd_it_iam *it = (struct osd_it_iam *)di;
6209
6210         return iam_it_key_size(&it->oi_it);
6211 }
6212
6213 static inline void
6214 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6215 {
6216         /* check if file type is required */
6217         if (ent->lde_attrs & LUDA_TYPE) {
6218                 struct luda_type *lt;
6219                 int align = sizeof(*lt) - 1;
6220
6221                 len = (len + align) & ~align;
6222                 lt = (struct luda_type *)(ent->lde_name + len);
6223                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6224         }
6225
6226         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6227 }
6228
6229 /**
6230  * build lu direct from backend fs dirent.
6231  */
6232
6233 static inline void
6234 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6235                    char *name, __u16 namelen, __u16 type, __u32 attr)
6236 {
6237         ent->lde_attrs = attr | LUDA_FID;
6238         fid_cpu_to_le(&ent->lde_fid, fid);
6239
6240         ent->lde_hash = cpu_to_le64(offset);
6241         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6242
6243         strncpy(ent->lde_name, name, namelen);
6244         ent->lde_name[namelen] = '\0';
6245         ent->lde_namelen = cpu_to_le16(namelen);
6246
6247         /* append lustre attributes */
6248         osd_it_append_attrs(ent, namelen, type);
6249 }
6250
6251 /**
6252  * Return pointer to the record under iterator.
6253  */
6254 static int osd_it_iam_rec(const struct lu_env *env,
6255                           const struct dt_it *di,
6256                           struct dt_rec *dtrec, __u32 attr)
6257 {
6258         struct osd_it_iam *it = (struct osd_it_iam *)di;
6259         struct osd_thread_info *info = osd_oti_get(env);
6260
6261         ENTRY;
6262
6263         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6264                 const struct osd_fid_pack *rec;
6265                 struct lu_fid *fid = &info->oti_fid;
6266                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6267                 char *name;
6268                 int namelen;
6269                 __u64 hash;
6270                 int rc;
6271
6272                 name = (char *)iam_it_key_get(&it->oi_it);
6273                 if (IS_ERR(name))
6274                         RETURN(PTR_ERR(name));
6275
6276                 namelen = iam_it_key_size(&it->oi_it);
6277
6278                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6279                 if (IS_ERR(rec))
6280                         RETURN(PTR_ERR(rec));
6281
6282                 rc = osd_fid_unpack(fid, rec);
6283                 if (rc)
6284                         RETURN(rc);
6285
6286                 hash = iam_it_store(&it->oi_it);
6287
6288                 /* IAM does not store object type in IAM index (dir) */
6289                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6290                                    0, LUDA_FID);
6291         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6292                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6293                            (struct iam_rec *)dtrec);
6294                 osd_quota_unpack(it->oi_obj, dtrec);
6295         } else {
6296                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6297                            (struct iam_rec *)dtrec);
6298         }
6299
6300         RETURN(0);
6301 }
6302
6303 /**
6304  * Returns cookie for current Iterator position.
6305  */
6306 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6307 {
6308         struct osd_it_iam *it = (struct osd_it_iam *)di;
6309
6310         return iam_it_store(&it->oi_it);
6311 }
6312
6313 /**
6314  * Restore iterator from cookie.
6315  *
6316  * \param  di      osd iterator
6317  * \param  hash    Iterator location cookie
6318  *
6319  * \retval +ve  di points to record with least key not larger than key.
6320  * \retval  0   di points to exact matched key
6321  * \retval -ve  failure
6322  */
6323
6324 static int osd_it_iam_load(const struct lu_env *env,
6325                            const struct dt_it *di, __u64 hash)
6326 {
6327         struct osd_it_iam *it = (struct osd_it_iam *)di;
6328
6329         return iam_it_load(&it->oi_it, hash);
6330 }
6331
6332 static const struct dt_index_operations osd_index_iam_ops = {
6333         .dio_lookup         = osd_index_iam_lookup,
6334         .dio_declare_insert = osd_index_declare_iam_insert,
6335         .dio_insert         = osd_index_iam_insert,
6336         .dio_declare_delete = osd_index_declare_iam_delete,
6337         .dio_delete         = osd_index_iam_delete,
6338         .dio_it     = {
6339                 .init     = osd_it_iam_init,
6340                 .fini     = osd_it_iam_fini,
6341                 .get      = osd_it_iam_get,
6342                 .put      = osd_it_iam_put,
6343                 .next     = osd_it_iam_next,
6344                 .key      = osd_it_iam_key,
6345                 .key_size = osd_it_iam_key_size,
6346                 .rec      = osd_it_iam_rec,
6347                 .store    = osd_it_iam_store,
6348                 .load     = osd_it_iam_load
6349         }
6350 };
6351
6352
6353 /**
6354  * Creates or initializes iterator context.
6355  *
6356  * \retval struct osd_it_ea, iterator structure on success
6357  *
6358  */
6359 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6360                                     struct dt_object *dt,
6361                                     __u32 attr)
6362 {
6363         struct osd_object *obj = osd_dt_obj(dt);
6364         struct osd_thread_info *info = osd_oti_get(env);
6365         struct osd_it_ea *oie;
6366         struct file *file;
6367         struct lu_object *lo = &dt->do_lu;
6368         struct dentry *obj_dentry;
6369
6370         ENTRY;
6371
6372         if (!dt_object_exists(dt) || obj->oo_destroyed)
6373                 RETURN(ERR_PTR(-ENOENT));
6374
6375         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6376         if (oie == NULL)
6377                 RETURN(ERR_PTR(-ENOMEM));
6378         obj_dentry = &oie->oie_dentry;
6379
6380         obj_dentry->d_inode = obj->oo_inode;
6381         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6382         obj_dentry->d_name.hash = 0;
6383
6384         oie->oie_rd_dirent       = 0;
6385         oie->oie_it_dirent       = 0;
6386         oie->oie_dirent          = NULL;
6387         if (unlikely(!info->oti_it_ea_buf_used)) {
6388                 oie->oie_buf = info->oti_it_ea_buf;
6389                 info->oti_it_ea_buf_used = 1;
6390         } else {
6391                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6392                 if (oie->oie_buf == NULL)
6393                         RETURN(ERR_PTR(-ENOMEM));
6394         }
6395         oie->oie_obj = obj;
6396
6397         file = &oie->oie_file;
6398
6399         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6400         if (attr & LUDA_64BITHASH)
6401                 file->f_mode    = FMODE_64BITHASH;
6402         else
6403                 file->f_mode    = FMODE_32BITHASH;
6404         file->f_path.dentry     = obj_dentry;
6405         file->f_mapping         = obj->oo_inode->i_mapping;
6406         file->f_op              = obj->oo_inode->i_fop;
6407         set_file_inode(file, obj->oo_inode);
6408
6409         lu_object_get(lo);
6410         RETURN((struct dt_it *)oie);
6411 }
6412
6413 /**
6414  * Destroy or finishes iterator context.
6415  *
6416  * \param di iterator structure to be destroyed
6417  */
6418 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6419 {
6420         struct osd_thread_info *info = osd_oti_get(env);
6421         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6422         struct osd_object *obj = oie->oie_obj;
6423         struct inode *inode = obj->oo_inode;
6424
6425         ENTRY;
6426         oie->oie_file.f_op->release(inode, &oie->oie_file);
6427         osd_object_put(env, obj);
6428         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6429                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6430         else
6431                 info->oti_it_ea_buf_used = 0;
6432         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6433         EXIT;
6434 }
6435
6436 /**
6437  * It position the iterator at given key, so that next lookup continues from
6438  * that key Or it is similar to dio_it->load() but based on a key,
6439  * rather than file position.
6440  *
6441  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6442  * to the beginning.
6443  *
6444  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6445  */
6446 static int osd_it_ea_get(const struct lu_env *env,
6447                          struct dt_it *di, const struct dt_key *key)
6448 {
6449         struct osd_it_ea *it = (struct osd_it_ea *)di;
6450
6451         ENTRY;
6452         LASSERT(((const char *)key)[0] == '\0');
6453         it->oie_file.f_pos = 0;
6454         it->oie_rd_dirent = 0;
6455         it->oie_it_dirent = 0;
6456         it->oie_dirent = NULL;
6457
6458         RETURN(+1);
6459 }
6460
6461 /**
6462  * Does nothing
6463  */
6464 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6465 {
6466 }
6467
6468 struct osd_filldir_cbs {
6469 #ifdef HAVE_DIR_CONTEXT
6470         struct dir_context ctx;
6471 #endif
6472         struct osd_it_ea  *it;
6473 };
6474 /**
6475  * It is called internally by ->readdir(). It fills the
6476  * iterator's in-memory data structure with required
6477  * information i.e. name, namelen, rec_size etc.
6478  *
6479  * \param buf in which information to be filled in.
6480  * \param name name of the file in given dir
6481  *
6482  * \retval 0 on success
6483  * \retval 1 on buffer full
6484  */
6485 #ifdef HAVE_FILLDIR_USE_CTX
6486 static int osd_ldiskfs_filldir(struct dir_context *buf,
6487 #else
6488 static int osd_ldiskfs_filldir(void *buf,
6489 #endif
6490                                const char *name, int namelen,
6491                                loff_t offset, __u64 ino, unsigned int d_type)
6492 {
6493         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6494         struct osd_object *obj = it->oie_obj;
6495         struct osd_it_ea_dirent *ent = it->oie_dirent;
6496         struct lu_fid *fid = &ent->oied_fid;
6497         struct osd_fid_pack *rec;
6498
6499         ENTRY;
6500
6501 /* this should never happen */
6502         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6503                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6504                 RETURN(-EIO);
6505         }
6506
6507         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6508             OSD_IT_EA_BUFSIZE)
6509                 RETURN(1);
6510
6511         /* "." is just the object itself. */
6512         if (namelen == 1 && name[0] == '.') {
6513                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6514         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6515                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6516                 if (osd_fid_unpack(fid, rec) != 0)
6517                         fid_zero(fid);
6518         } else {
6519                 fid_zero(fid);
6520         }
6521         d_type &= ~LDISKFS_DIRENT_LUFID;
6522
6523         /* NOT export local root. */
6524         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6525                 ino = obj->oo_inode->i_ino;
6526                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6527         }
6528
6529         ent->oied_ino     = ino;
6530         ent->oied_off     = offset;
6531         ent->oied_namelen = namelen;
6532         ent->oied_type    = d_type;
6533
6534         memcpy(ent->oied_name, name, namelen);
6535
6536         it->oie_rd_dirent++;
6537         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6538         RETURN(0);
6539 }
6540
6541 /**
6542  * Calls ->readdir() to load a directory entry at a time
6543  * and stored it in iterator's in-memory data structure.
6544  *
6545  * \param di iterator's in memory structure
6546  *
6547  * \retval   0 on success
6548  * \retval -ve on error
6549  * \retval +1 reach the end of entry
6550  */
6551 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6552                                const struct dt_it *di)
6553 {
6554         struct osd_it_ea *it = (struct osd_it_ea *)di;
6555         struct osd_object *obj = it->oie_obj;
6556         struct inode *inode = obj->oo_inode;
6557         struct htree_lock *hlock = NULL;
6558         struct file *filp = &it->oie_file;
6559         int rc = 0;
6560         struct osd_filldir_cbs buf = {
6561 #ifdef HAVE_DIR_CONTEXT
6562                 .ctx.actor = osd_ldiskfs_filldir,
6563 #endif
6564                 .it = it
6565         };
6566
6567         ENTRY;
6568         it->oie_dirent = it->oie_buf;
6569         it->oie_rd_dirent = 0;
6570
6571         if (obj->oo_hl_head != NULL) {
6572                 hlock = osd_oti_get(env)->oti_hlock;
6573                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6574                                    inode, LDISKFS_HLOCK_READDIR);
6575         } else {
6576                 down_read(&obj->oo_ext_idx_sem);
6577         }
6578
6579 #ifdef HAVE_DIR_CONTEXT
6580         buf.ctx.pos = filp->f_pos;
6581 #ifdef HAVE_ITERATE_SHARED
6582         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6583 #else
6584         rc = inode->i_fop->iterate(filp, &buf.ctx);
6585 #endif
6586         filp->f_pos = buf.ctx.pos;
6587 #else
6588         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6589 #endif
6590
6591         if (hlock != NULL)
6592                 ldiskfs_htree_unlock(hlock);
6593         else
6594                 up_read(&obj->oo_ext_idx_sem);
6595
6596         if (it->oie_rd_dirent == 0) {
6597                 /*
6598                  * If it does not get any dirent, it means it has been reached
6599                  * to the end of the dir
6600                  */
6601                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6602                 if (rc == 0)
6603                         rc = 1;
6604         } else {
6605                 it->oie_dirent = it->oie_buf;
6606                 it->oie_it_dirent = 1;
6607         }
6608
6609         RETURN(rc);
6610 }
6611
6612 /**
6613  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6614  * to load a directory entry at a time and stored it in
6615  * iterator's in-memory data structure.
6616  *
6617  * \param di iterator's in memory structure
6618  *
6619  * \retval +ve iterator reached to end
6620  * \retval   0 iterator not reached to end
6621  * \retval -ve on error
6622  */
6623 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6624 {
6625         struct osd_it_ea *it = (struct osd_it_ea *)di;
6626         int rc;
6627
6628         ENTRY;
6629
6630         if (it->oie_it_dirent < it->oie_rd_dirent) {
6631                 it->oie_dirent =
6632                         (void *)it->oie_dirent +
6633                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6634                                        it->oie_dirent->oied_namelen);
6635                 it->oie_it_dirent++;
6636                 rc = 0;
6637         } else {
6638                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6639                         rc = 1;
6640                 else
6641                         rc = osd_ldiskfs_it_fill(env, di);
6642         }
6643
6644         RETURN(rc);
6645 }
6646
6647 /**
6648  * Returns the key at current position from iterator's in memory structure.
6649  *
6650  * \param di iterator's in memory structure
6651  *
6652  * \retval key i.e. struct dt_key on success
6653  */
6654 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6655                                     const struct dt_it *di)
6656 {
6657         struct osd_it_ea *it = (struct osd_it_ea *)di;
6658
6659         return (struct dt_key *)it->oie_dirent->oied_name;
6660 }
6661
6662 /**
6663  * Returns key's size at current position from iterator's in memory structure.
6664  *
6665  * \param di iterator's in memory structure
6666  *
6667  * \retval key_size i.e. struct dt_key on success
6668  */
6669 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6670 {
6671         struct osd_it_ea *it = (struct osd_it_ea *)di;
6672
6673         return it->oie_dirent->oied_namelen;
6674 }
6675
6676 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6677 {
6678         if (LDISKFS_DIR_REC_LEN(de) >=
6679             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6680                 return true;
6681
6682         return false;
6683 }
6684
6685 static inline bool
6686 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6687                      unsigned int blocksize, bool dotdot)
6688 {
6689         if (dotdot)
6690                 return osd_dotdot_has_space(de);
6691
6692         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6693             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6694                 return true;
6695
6696         return false;
6697 }
6698
6699 static int
6700 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6701                     handle_t *jh, struct dentry *dentry,
6702                     const struct lu_fid *fid, struct buffer_head *bh,
6703                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6704                     bool dotdot)
6705 {
6706         struct inode *dir = dentry->d_parent->d_inode;
6707         struct inode *inode = dentry->d_inode;
6708         struct osd_fid_pack *rec;
6709         struct ldiskfs_dentry_param *ldp;
6710         int namelen = dentry->d_name.len;
6711         int rc;
6712         struct osd_thread_info *info = osd_oti_get(env);
6713
6714         ENTRY;
6715
6716         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6717                 RETURN(0);
6718
6719         /* There is enough space to hold the FID-in-dirent. */
6720         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6721                 rc = ldiskfs_journal_get_write_access(jh, bh);
6722                 if (rc != 0)
6723                         RETURN(rc);
6724
6725                 de->name[namelen] = 0;
6726                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6727                 rec->fp_len = sizeof(struct lu_fid) + 1;
6728                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6729                 de->file_type |= LDISKFS_DIRENT_LUFID;
6730                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6731
6732                 RETURN(rc);
6733         }
6734
6735         LASSERT(!dotdot);
6736
6737         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6738         if (rc != 0)
6739                 RETURN(rc);
6740
6741         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6742         osd_get_ldiskfs_dirent_param(ldp, fid);
6743         dentry->d_fsdata = (void *)ldp;
6744         ll_vfs_dq_init(dir);
6745         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6746         /*
6747          * It is too bad, we cannot reinsert the name entry back.
6748          * That means we lose it!
6749          */
6750         if (rc != 0)
6751                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6752                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6753                        osd_ino2name(inode),
6754                        dir->i_ino, dir->i_generation, namelen,
6755                        dentry->d_name.name, PFID(fid), rc);
6756
6757         RETURN(rc);
6758 }
6759
6760 static int
6761 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6762                         struct osd_it_ea *it, struct lu_fid *fid,
6763                         struct osd_inode_id *id, __u32 *attr)
6764 {
6765         struct osd_thread_info *info = osd_oti_get(env);
6766         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6767         struct osd_device *dev = osd_obj2dev(obj);
6768         struct super_block *sb = osd_sb(dev);
6769         const char *devname = osd_name(dev);
6770         struct osd_it_ea_dirent *ent = it->oie_dirent;
6771         struct inode *dir = obj->oo_inode;
6772         struct htree_lock *hlock = NULL;
6773         struct buffer_head *bh = NULL;
6774         handle_t *jh = NULL;
6775         struct ldiskfs_dir_entry_2 *de;
6776         struct dentry *dentry;
6777         struct inode *inode;
6778         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6779         int credits;
6780         int rc;
6781         bool dotdot = false;
6782         bool dirty = false;
6783
6784         ENTRY;
6785
6786         if (ent->oied_name[0] == '.') {
6787                 if (ent->oied_namelen == 1)
6788                         RETURN(0);
6789
6790                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6791                         dotdot = true;
6792         }
6793
6794         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6795         inode = osd_iget(info, dev, id);
6796         if (IS_ERR(inode)) {
6797                 rc = PTR_ERR(inode);
6798                 if (rc == -ENOENT || rc == -ESTALE) {
6799                         /*
6800                          * Maybe dangling name entry, or
6801                          * corrupted directory entry.
6802                          */
6803                         *attr |= LUDA_UNKNOWN;
6804                         rc = 0;
6805                 } else {
6806                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6807                                "check_repair, dir = %lu/%u, name = %.*s, "
6808                                "ino = %llu, rc = %d\n",
6809                                devname, dir->i_ino, dir->i_generation,
6810                                ent->oied_namelen, ent->oied_name,
6811                                ent->oied_ino, rc);
6812                 }
6813
6814                 RETURN(rc);
6815         }
6816
6817         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6818                                            ent->oied_namelen);
6819         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6820         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6821                 lma = NULL;
6822         else if (rc != 0)
6823                 GOTO(out, rc);
6824
6825         /*
6826          * We need to ensure that the name entry is still valid.
6827          * Because it may be removed or renamed by other already.
6828          *
6829          * The unlink or rename operation will start journal before PDO lock,
6830          * so to avoid deadlock, here we need to start journal handle before
6831          * related PDO lock also. But because we do not know whether there
6832          * will be something to be repaired before PDO lock, we just start
6833          * journal without conditions.
6834          *
6835          * We may need to remove the name entry firstly, then insert back.
6836          * One credit is for user quota file update.
6837          * One credit is for group quota file update.
6838          * Two credits are for dirty inode.
6839          */
6840         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6841                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6842
6843         if (dev->od_dirent_journal != 0) {
6844
6845 again:
6846                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6847                 if (IS_ERR(jh)) {
6848                         rc = PTR_ERR(jh);
6849                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6850                                "check_repair, dir = %lu/%u, credits = %d, "
6851                                "name = %.*s, ino = %llu: rc = %d\n",
6852                                devname, dir->i_ino, dir->i_generation, credits,
6853                                ent->oied_namelen, ent->oied_name,
6854                                ent->oied_ino, rc);
6855
6856                         GOTO(out_inode, rc);
6857                 }
6858
6859                 if (obj->oo_hl_head != NULL) {
6860                         hlock = osd_oti_get(env)->oti_hlock;
6861                         /*
6862                          * "0" means exclusive lock for the whole directory.
6863                          * We need to prevent others access such name entry
6864                          * during the delete + insert. Neither HLOCK_ADD nor
6865                          * HLOCK_DEL cannot guarantee the atomicity.
6866                          */
6867                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6868                 } else {
6869                         down_write(&obj->oo_ext_idx_sem);
6870                 }
6871         } else {
6872                 if (obj->oo_hl_head != NULL) {
6873                         hlock = osd_oti_get(env)->oti_hlock;
6874                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6875                                            LDISKFS_HLOCK_LOOKUP);
6876                 } else {
6877                         down_read(&obj->oo_ext_idx_sem);
6878                 }
6879         }
6880
6881         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6882         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6883                 *attr |= LUDA_IGNORE;
6884
6885                 GOTO(out, rc = 0);
6886         }
6887
6888         /*
6889          * For dotdot entry, if there is not enough space to hold the
6890          * FID-in-dirent, just keep them there. It only happens when the
6891          * device upgraded from 1.8 or restored from MDT file-level backup.
6892          * For the whole directory, only dotdot entry have no FID-in-dirent
6893          * and needs to get FID from LMA when readdir, it will not affect the
6894          * performance much.
6895          */
6896         if (dotdot && !osd_dotdot_has_space(de)) {
6897                 *attr |= LUDA_UNKNOWN;
6898
6899                 GOTO(out, rc = 0);
6900         }
6901
6902         if (lma != NULL) {
6903                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6904                         GOTO(out, rc = 0);
6905
6906                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6907                         struct lu_fid *tfid = &lma->lma_self_fid;
6908
6909                         if (likely(dotdot &&
6910                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6911                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6912                                 /*
6913                                  * It must be REMOTE_PARENT_DIR and as the
6914                                  * 'dotdot' entry of remote directory
6915                                  */
6916                                 *attr |= LUDA_IGNORE;
6917                         } else {
6918                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6919                                        "parent directory, but got %.*s under "
6920                                        "dir = %lu/%u with the FID "DFID"\n",
6921                                        devname, ent->oied_namelen,
6922                                        ent->oied_name, dir->i_ino,
6923                                        dir->i_generation, PFID(tfid));
6924
6925                                 *attr |= LUDA_UNKNOWN;
6926                         }
6927
6928                         GOTO(out, rc = 0);
6929                 }
6930         }
6931
6932         if (!fid_is_zero(fid)) {
6933                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6934                                               ent->oied_namelen);
6935                 if (rc == -ENOENT ||
6936                     (rc == -ENODATA &&
6937                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6938                         /*
6939                          * linkEA does not recognize the dirent entry,
6940                          * it may because the dirent entry corruption
6941                          * and points to other's inode.
6942                          */
6943                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6944                                "recognize the dirent, dir = %lu/%u, "
6945                                " name = %.*s, ino = %llu, "
6946                                DFID": rc = %d\n", devname, dir->i_ino,
6947                                dir->i_generation, ent->oied_namelen,
6948                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6949                         *attr |= LUDA_UNKNOWN;
6950
6951                         GOTO(out, rc = 0);
6952                 }
6953
6954                 if (rc && rc != -ENODATA) {
6955                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6956                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6957                                DFID": rc = %d\n", devname, dir->i_ino,
6958                                dir->i_generation, ent->oied_namelen,
6959                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6960                         *attr |= LUDA_UNKNOWN;
6961
6962                         GOTO(out, rc = 0);
6963                 }
6964         }
6965
6966         if (lma != NULL) {
6967                 /*
6968                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6969                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6970                  */
6971                 if (*attr & LUDA_VERIFY_DRYRUN) {
6972                         *fid = lma->lma_self_fid;
6973                         *attr |= LUDA_REPAIR;
6974
6975                         GOTO(out, rc = 0);
6976                 }
6977
6978                 if (jh == NULL) {
6979                         brelse(bh);
6980                         dev->od_dirent_journal = 1;
6981                         if (hlock != NULL) {
6982                                 ldiskfs_htree_unlock(hlock);
6983                                 hlock = NULL;
6984                         } else {
6985                                 up_read(&obj->oo_ext_idx_sem);
6986                         }
6987
6988                         goto again;
6989                 }
6990
6991                 *fid = lma->lma_self_fid;
6992                 dirty = true;
6993                 /* Update or append the FID-in-dirent. */
6994                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6995                                          bh, de, hlock, dotdot);
6996                 if (rc == 0)
6997                         *attr |= LUDA_REPAIR;
6998                 else
6999                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7000                                "the dirent, dir = %lu/%u, name = %.*s, "
7001                                "ino = %llu, "DFID": rc = %d\n",
7002                                devname, dir->i_ino, dir->i_generation,
7003                                ent->oied_namelen, ent->oied_name,
7004                                ent->oied_ino, PFID(fid), rc);
7005         } else {
7006                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7007                 if (*attr & LUDA_VERIFY_DRYRUN) {
7008                         if (fid_is_sane(fid)) {
7009                                 *attr |= LUDA_REPAIR;
7010                         } else if (dev->od_index == 0) {
7011                                 lu_igif_build(fid, inode->i_ino,
7012                                               inode->i_generation);
7013                                 *attr |= LUDA_UPGRADE;
7014                         }
7015
7016                         GOTO(out, rc = 0);
7017                 }
7018
7019                 if (jh == NULL) {
7020                         brelse(bh);
7021                         dev->od_dirent_journal = 1;
7022                         if (hlock != NULL) {
7023                                 ldiskfs_htree_unlock(hlock);
7024                                 hlock = NULL;
7025                         } else {
7026                                 up_read(&obj->oo_ext_idx_sem);
7027                         }
7028
7029                         goto again;
7030                 }
7031
7032                 dirty = true;
7033                 if (unlikely(fid_is_sane(fid))) {
7034                         /*
7035                          * FID-in-dirent exists, but FID-in-LMA is lost.
7036                          * Trust the FID-in-dirent, and add FID-in-LMA.
7037                          */
7038                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7039                         if (rc == 0)
7040                                 *attr |= LUDA_REPAIR;
7041                         else
7042                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7043                                        "update dirent, dir = %lu/%u, "
7044                                        "name = %.*s, ino = %llu, "
7045                                        DFID": rc = %d\n",
7046                                        devname, dir->i_ino, dir->i_generation,
7047                                        ent->oied_namelen, ent->oied_name,
7048                                        ent->oied_ino, PFID(fid), rc);
7049                 } else if (dev->od_index == 0) {
7050                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7051                         /*
7052                          * It is probably IGIF object. Only aappend the
7053                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7054                          */
7055                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7056                                                  bh, de, hlock, dotdot);
7057                         if (rc == 0)
7058                                 *attr |= LUDA_UPGRADE;
7059                         else
7060                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7061                                        "after the dirent, dir = %lu/%u, "
7062                                        "name = %.*s, ino = %llu, "
7063                                        DFID": rc = %d\n",
7064                                        devname, dir->i_ino, dir->i_generation,
7065                                        ent->oied_namelen, ent->oied_name,
7066                                        ent->oied_ino, PFID(fid), rc);
7067                 }
7068         }
7069
7070         GOTO(out, rc);
7071
7072 out:
7073         if (!IS_ERR(bh))
7074                 brelse(bh);
7075         if (hlock != NULL) {
7076                 ldiskfs_htree_unlock(hlock);
7077         } else {
7078                 if (dev->od_dirent_journal != 0)
7079                         up_write(&obj->oo_ext_idx_sem);
7080                 else
7081                         up_read(&obj->oo_ext_idx_sem);
7082         }
7083
7084         if (jh != NULL)
7085                 ldiskfs_journal_stop(jh);
7086
7087 out_inode:
7088         iput(inode);
7089         if (rc >= 0 && !dirty)
7090                 dev->od_dirent_journal = 0;
7091
7092         return rc;
7093 }
7094
7095 /**
7096  * Returns the value at current position from iterator's in memory structure.
7097  *
7098  * \param di struct osd_it_ea, iterator's in memory structure
7099  * \param attr attr requested for dirent.
7100  * \param lde lustre dirent
7101  *
7102  * \retval   0 no error and \param lde has correct lustre dirent.
7103  * \retval -ve on error
7104  */
7105 static inline int osd_it_ea_rec(const struct lu_env *env,
7106                                 const struct dt_it *di,
7107                                 struct dt_rec *dtrec, __u32 attr)
7108 {
7109         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7110         struct osd_object      *obj   = it->oie_obj;
7111         struct osd_device      *dev   = osd_obj2dev(obj);
7112         struct osd_thread_info *oti   = osd_oti_get(env);
7113         struct osd_inode_id    *id    = &oti->oti_id;
7114         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7115         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7116         __u32 ino = it->oie_dirent->oied_ino;
7117         int rc = 0;
7118
7119         ENTRY;
7120
7121         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7122
7123         if (attr & LUDA_VERIFY) {
7124                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7125                         attr |= LUDA_IGNORE;
7126                         /*
7127                          * If the parent is on remote MDT, and there
7128                          * is no FID-in-dirent, then we have to get
7129                          * the parent FID from the linkEA.
7130                          */
7131                         if (!fid_is_sane(fid) &&
7132                             it->oie_dirent->oied_namelen == 2 &&
7133                             it->oie_dirent->oied_name[0] == '.' &&
7134                             it->oie_dirent->oied_name[1] == '.')
7135                                 osd_get_pfid_from_linkea(env, obj, fid);
7136                 } else {
7137                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7138                                                      &attr);
7139                 }
7140
7141                 if (!fid_is_sane(fid))
7142                         attr |= LUDA_UNKNOWN;
7143         } else {
7144                 attr &= ~LU_DIRENT_ATTRS_MASK;
7145                 if (!fid_is_sane(fid)) {
7146                         bool is_dotdot = false;
7147
7148                         if (it->oie_dirent->oied_namelen == 2 &&
7149                             it->oie_dirent->oied_name[0] == '.' &&
7150                             it->oie_dirent->oied_name[1] == '.')
7151                                 is_dotdot = true;
7152                         /*
7153                          * If the parent is on remote MDT, and there
7154                          * is no FID-in-dirent, then we have to get
7155                          * the parent FID from the linkEA.
7156                          */
7157                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7158                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7159                         } else {
7160                                 if (is_dotdot == false &&
7161                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7162                                         RETURN(-ENOENT);
7163
7164                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7165                         }
7166                 } else {
7167                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7168                 }
7169         }
7170
7171         /* Pack the entry anyway, at least the offset is right. */
7172         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7173                            it->oie_dirent->oied_name,
7174                            it->oie_dirent->oied_namelen,
7175                            it->oie_dirent->oied_type, attr);
7176
7177         if (rc < 0)
7178                 RETURN(rc);
7179
7180         if (osd_remote_fid(env, dev, fid))
7181                 RETURN(0);
7182
7183         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7184                 osd_add_oi_cache(oti, dev, id, fid);
7185
7186         RETURN(rc > 0 ? 0 : rc);
7187 }
7188
7189 /**
7190  * Returns the record size size at current position.
7191  *
7192  * This function will return record(lu_dirent) size in bytes.
7193  *
7194  * \param[in] env       execution environment
7195  * \param[in] di        iterator's in memory structure
7196  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7197  *                      calculate the lu_dirent size.
7198  *
7199  * \retval      record size(in bytes & in memory) of the current lu_dirent
7200  *              entry.
7201  */
7202 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7203                               __u32 attr)
7204 {
7205         struct osd_it_ea *it = (struct osd_it_ea *)di;
7206
7207         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7208 }
7209
7210 /**
7211  * Returns a cookie for current position of the iterator head, so that
7212  * user can use this cookie to load/start the iterator next time.
7213  *
7214  * \param di iterator's in memory structure
7215  *
7216  * \retval cookie for current position, on success
7217  */
7218 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7219 {
7220         struct osd_it_ea *it = (struct osd_it_ea *)di;
7221
7222         return it->oie_dirent->oied_off;
7223 }
7224
7225 /**
7226  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7227  * to load a directory entry at a time and stored it i inn,
7228  * in iterator's in-memory data structure.
7229  *
7230  * \param di struct osd_it_ea, iterator's in memory structure
7231  *
7232  * \retval +ve on success
7233  * \retval -ve on error
7234  */
7235 static int osd_it_ea_load(const struct lu_env *env,
7236                           const struct dt_it *di, __u64 hash)
7237 {
7238         struct osd_it_ea *it = (struct osd_it_ea *)di;
7239         int rc;
7240
7241         ENTRY;
7242         it->oie_file.f_pos = hash;
7243
7244         rc =  osd_ldiskfs_it_fill(env, di);
7245         if (rc > 0)
7246                 rc = -ENODATA;
7247
7248         if (rc == 0)
7249                 rc = 1;
7250
7251         RETURN(rc);
7252 }
7253
7254 /**
7255  * Index lookup function for interoperability mode (b11826).
7256  *
7257  * \param key,  key i.e. file name to be searched
7258  *
7259  * \retval +ve, on success
7260  * \retval -ve, on error
7261  */
7262 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7263                                struct dt_rec *rec, const struct dt_key *key)
7264 {
7265         struct osd_object *obj = osd_dt_obj(dt);
7266         int rc = 0;
7267
7268         ENTRY;
7269
7270         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7271         LINVRNT(osd_invariant(obj));
7272
7273         rc = osd_ea_lookup_rec(env, obj, rec, key);
7274         if (rc == 0)
7275                 rc = 1;
7276         RETURN(rc);
7277 }
7278
7279 /**
7280  * Index and Iterator operations for interoperability
7281  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7282  */
7283 static const struct dt_index_operations osd_index_ea_ops = {
7284         .dio_lookup         = osd_index_ea_lookup,
7285         .dio_declare_insert = osd_index_declare_ea_insert,
7286         .dio_insert         = osd_index_ea_insert,
7287         .dio_declare_delete = osd_index_declare_ea_delete,
7288         .dio_delete         = osd_index_ea_delete,
7289         .dio_it     = {
7290                 .init     = osd_it_ea_init,
7291                 .fini     = osd_it_ea_fini,
7292                 .get      = osd_it_ea_get,
7293                 .put      = osd_it_ea_put,
7294                 .next     = osd_it_ea_next,
7295                 .key      = osd_it_ea_key,
7296                 .key_size = osd_it_ea_key_size,
7297                 .rec      = osd_it_ea_rec,
7298                 .rec_size = osd_it_ea_rec_size,
7299                 .store    = osd_it_ea_store,
7300                 .load     = osd_it_ea_load
7301         }
7302 };
7303
7304 static void *osd_key_init(const struct lu_context *ctx,
7305                           struct lu_context_key *key)
7306 {
7307         struct osd_thread_info *info;
7308
7309         OBD_ALLOC_PTR(info);
7310         if (info == NULL)
7311                 return ERR_PTR(-ENOMEM);
7312
7313         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7314         if (info->oti_it_ea_buf == NULL)
7315                 goto out_free_info;
7316
7317         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7318
7319         info->oti_hlock = ldiskfs_htree_lock_alloc();
7320         if (info->oti_hlock == NULL)
7321                 goto out_free_ea;
7322
7323         return info;
7324
7325 out_free_ea:
7326         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7327 out_free_info:
7328         OBD_FREE_PTR(info);
7329         return ERR_PTR(-ENOMEM);
7330 }
7331
7332 static void osd_key_fini(const struct lu_context *ctx,
7333                          struct lu_context_key *key, void *data)
7334 {
7335         struct osd_thread_info *info = data;
7336         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7337         struct osd_idmap_cache *idc = info->oti_ins_cache;
7338
7339         if (info->oti_dio_pages) {
7340                 int i;
7341                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7342                         if (info->oti_dio_pages[i])
7343                                 __free_page(info->oti_dio_pages[i]);
7344                 }
7345                 OBD_FREE(info->oti_dio_pages,
7346                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7347         }
7348
7349         if (info->oti_inode != NULL)
7350                 OBD_FREE_PTR(lli);
7351         if (info->oti_hlock != NULL)
7352                 ldiskfs_htree_lock_free(info->oti_hlock);
7353         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7354         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7355         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7356         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7357         lu_buf_free(&info->oti_big_buf);
7358         if (idc != NULL) {
7359                 LASSERT(info->oti_ins_cache_size > 0);
7360                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7361                 info->oti_ins_cache = NULL;
7362                 info->oti_ins_cache_size = 0;
7363         }
7364         OBD_FREE_PTR(info);
7365 }
7366
7367 static void osd_key_exit(const struct lu_context *ctx,
7368                          struct lu_context_key *key, void *data)
7369 {
7370         struct osd_thread_info *info = data;
7371
7372         LASSERT(info->oti_r_locks == 0);
7373         LASSERT(info->oti_w_locks == 0);
7374         LASSERT(info->oti_txns    == 0);
7375 }
7376
7377 /* type constructor/destructor: osd_type_init, osd_type_fini */
7378 LU_TYPE_INIT_FINI(osd, &osd_key);
7379
7380 struct lu_context_key osd_key = {
7381         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7382         .lct_init = osd_key_init,
7383         .lct_fini = osd_key_fini,
7384         .lct_exit = osd_key_exit
7385 };
7386
7387
7388 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7389                            const char *name, struct lu_device *next)
7390 {
7391         struct osd_device *osd = osd_dev(d);
7392
7393         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7394             sizeof(osd->od_svname))
7395                 return -E2BIG;
7396         return osd_procfs_init(osd, name);
7397 }
7398
7399 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7400 {
7401         struct seq_server_site *ss = osd_seq_site(osd);
7402         int rc;
7403
7404         ENTRY;
7405
7406         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7407                 RETURN(0);
7408
7409         if (unlikely(ss == NULL))
7410                 RETURN(-ENODEV);
7411
7412         OBD_ALLOC_PTR(osd->od_cl_seq);
7413         if (osd->od_cl_seq == NULL)
7414                 RETURN(-ENOMEM);
7415
7416         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7417                              osd->od_svname, ss->ss_server_seq);
7418         if (rc != 0) {
7419                 OBD_FREE_PTR(osd->od_cl_seq);
7420                 osd->od_cl_seq = NULL;
7421                 RETURN(rc);
7422         }
7423
7424         if (ss->ss_node_id == 0) {
7425                 /*
7426                  * If the OSD on the sequence controller(MDT0), then allocate
7427                  * sequence here, otherwise allocate sequence after connected
7428                  * to MDT0 (see mdt_register_lwp_callback()).
7429                  */
7430                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7431                                    &osd->od_cl_seq->lcs_space, env);
7432         }
7433
7434         RETURN(rc);
7435 }
7436
7437 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7438 {
7439         if (osd->od_cl_seq == NULL)
7440                 return;
7441
7442         seq_client_fini(osd->od_cl_seq);
7443         OBD_FREE_PTR(osd->od_cl_seq);
7444         osd->od_cl_seq = NULL;
7445 }
7446
7447 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7448 {
7449         ENTRY;
7450
7451         /* shutdown quota slave instance associated with the device */
7452         if (o->od_quota_slave_md != NULL) {
7453                 struct qsd_instance *qsd = o->od_quota_slave_md;
7454
7455                 o->od_quota_slave_md = NULL;
7456                 qsd_fini(env, qsd);
7457         }
7458
7459         if (o->od_quota_slave_dt != NULL) {
7460                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7461
7462                 o->od_quota_slave_dt = NULL;
7463                 qsd_fini(env, qsd);
7464         }
7465
7466         osd_fid_fini(env, o);
7467         osd_scrub_cleanup(env, o);
7468
7469         RETURN(0);
7470 }
7471
7472 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7473 {
7474         ENTRY;
7475
7476         if (o->od_mnt != NULL) {
7477                 shrink_dcache_sb(osd_sb(o));
7478                 osd_sync(env, &o->od_dt_dev);
7479
7480                 mntput(o->od_mnt);
7481                 o->od_mnt = NULL;
7482         }
7483
7484         EXIT;
7485 }
7486
7487 static int osd_mount(const struct lu_env *env,
7488                      struct osd_device *o, struct lustre_cfg *cfg)
7489 {
7490         const char *name = lustre_cfg_string(cfg, 0);
7491         const char *dev = lustre_cfg_string(cfg, 1);
7492         const char *opts;
7493         unsigned long page, s_flags, lmd_flags = 0;
7494         struct page *__page;
7495         struct file_system_type *type;
7496         char *options = NULL;
7497         char *str;
7498         struct osd_thread_info *info = osd_oti_get(env);
7499         struct lu_fid *fid = &info->oti_fid;
7500         struct inode *inode;
7501         int rc = 0, force_over_512tb = 0;
7502
7503         ENTRY;
7504
7505         if (o->od_mnt != NULL)
7506                 RETURN(0);
7507
7508         if (strlen(dev) >= sizeof(o->od_mntdev))
7509                 RETURN(-E2BIG);
7510         strcpy(o->od_mntdev, dev);
7511
7512         str = lustre_cfg_string(cfg, 2);
7513         s_flags = simple_strtoul(str, NULL, 0);
7514         str = strstr(str, ":");
7515         if (str)
7516                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7517         opts = lustre_cfg_string(cfg, 3);
7518 #ifdef __BIG_ENDIAN
7519         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7520                 CERROR("%s: device %s extents feature is not guaranteed to "
7521                        "work on big-endian systems. Use \"bigendian_extents\" "
7522                        "mount option to override.\n", name, dev);
7523                 RETURN(-EINVAL);
7524         }
7525 #endif
7526 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7527         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7528                 CWARN("force_over_128tb option is deprecated. "
7529                       "Filesystems less than 512TB can be created without any "
7530                       "force options. Use force_over_512tb option for "
7531                       "filesystems greater than 512TB.\n");
7532         }
7533 #endif
7534 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7535         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7536                 CWARN("force_over_256tb option is deprecated. "
7537                       "Filesystems less than 512TB can be created without any "
7538                       "force options. Use force_over_512tb option for "
7539                       "filesystems greater than 512TB.\n");
7540         }
7541 #endif
7542
7543         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7544                 force_over_512tb = 1;
7545
7546         __page = alloc_page(GFP_KERNEL);
7547         if (__page == NULL)
7548                 GOTO(out, rc = -ENOMEM);
7549         page = (unsigned long)page_address(__page);
7550         options = (char *)page;
7551         *options = '\0';
7552         if (opts != NULL) {
7553                 /* strip out the options for back compatiblity */
7554                 static char *sout[] = {
7555                         "mballoc",
7556                         "iopen",
7557                         "noiopen",
7558                         "iopen_nopriv",
7559                         "extents",
7560                         "noextents",
7561                         /* strip out option we processed in osd */
7562                         "bigendian_extents",
7563                         "force_over_128tb",
7564                         "force_over_256tb",
7565                         "force_over_512tb",
7566                         NULL
7567                 };
7568                 strcat(options, opts);
7569                 for (rc = 0, str = options; sout[rc]; ) {
7570                         char *op = strstr(str, sout[rc]);
7571
7572                         if (op == NULL) {
7573                                 rc++;
7574                                 str = options;
7575                                 continue;
7576                         }
7577                         if (op == options || *(op - 1) == ',') {
7578                                 str = op + strlen(sout[rc]);
7579                                 if (*str == ',' || *str == '\0') {
7580                                         *str == ',' ? str++ : str;
7581                                         memmove(op, str, strlen(str) + 1);
7582                                 }
7583                         }
7584                         for (str = op; *str != ',' && *str != '\0'; str++)
7585                                 ;
7586                 }
7587         } else {
7588                 strncat(options, "user_xattr,acl", 14);
7589         }
7590
7591         /* Glom up mount options */
7592         if (*options != '\0')
7593                 strcat(options, ",");
7594         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7595
7596         type = get_fs_type("ldiskfs");
7597         if (!type) {
7598                 CERROR("%s: cannot find ldiskfs module\n", name);
7599                 GOTO(out, rc = -ENODEV);
7600         }
7601
7602         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7603         module_put(type->owner);
7604
7605         if (IS_ERR(o->od_mnt)) {
7606                 rc = PTR_ERR(o->od_mnt);
7607                 o->od_mnt = NULL;
7608                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7609                 GOTO(out, rc);
7610         }
7611
7612         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7613                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7614                                  force_over_512tb == 0) {
7615                 CERROR("%s: device %s LDISKFS does not support filesystems "
7616                        "greater than 512TB and can cause data corruption. "
7617                        "Use \"force_over_512tb\" mount option to override.\n",
7618                        name, dev);
7619                 GOTO(out_mnt, rc = -EINVAL);
7620         }
7621
7622         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7623                 if (priv_dev_set_rdonly) {
7624                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7625                         o->od_dt_dev.dd_rdonly = 1;
7626                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7627                                       name);
7628                 } else {
7629                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7630                                       name);
7631
7632                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7633                 }
7634         } else if (priv_dev_check_rdonly &&
7635                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7636                 CERROR("%s: underlying device %s is marked as "
7637                        "read-only. Setup failed\n", name, dev);
7638
7639                 GOTO(out_mnt, rc = -EROFS);
7640         }
7641
7642         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7643                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7644                 GOTO(out_mnt, rc = -EINVAL);
7645         }
7646
7647 #ifdef LDISKFS_MOUNT_DIRDATA
7648         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7649                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7650         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7651                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7652                       "enabling the dirdata feature. If you do not want to "
7653                       "downgrade to Lustre-1.x again, you can enable it via "
7654                       "'tune2fs -O dirdata device'\n", name, dev);
7655 #endif
7656         inode = osd_sb(o)->s_root->d_inode;
7657         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7658         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7659         if (rc != 0) {
7660                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7661                 GOTO(out_mnt, rc);
7662         }
7663
7664         if (lmd_flags & LMD_FLG_NOSCRUB)
7665                 o->od_auto_scrub_interval = AS_NEVER;
7666
7667         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7668                 /* do not use pagecache with flash-backed storage */
7669                 o->od_writethrough_cache = 0;
7670                 o->od_read_cache = 0;
7671         }
7672
7673         GOTO(out, rc = 0);
7674
7675 out_mnt:
7676         mntput(o->od_mnt);
7677         o->od_mnt = NULL;
7678
7679 out:
7680         if (__page)
7681                 __free_page(__page);
7682
7683         return rc;
7684 }
7685
7686 static struct lu_device *osd_device_fini(const struct lu_env *env,
7687                                          struct lu_device *d)
7688 {
7689         struct osd_device *o = osd_dev(d);
7690
7691         ENTRY;
7692
7693         osd_index_backup(env, o, false);
7694         osd_shutdown(env, o);
7695         osd_procfs_fini(o);
7696         osd_obj_map_fini(o);
7697         osd_umount(env, o);
7698
7699         RETURN(NULL);
7700 }
7701
7702 static int osd_device_init0(const struct lu_env *env,
7703                             struct osd_device *o,
7704                             struct lustre_cfg *cfg)
7705 {
7706         struct lu_device *l = osd2lu_dev(o);
7707         struct osd_thread_info *info;
7708         int rc;
7709         int cplen = 0;
7710
7711         /* if the module was re-loaded, env can loose its keys */
7712         rc = lu_env_refill((struct lu_env *)env);
7713         if (rc)
7714                 GOTO(out, rc);
7715         info = osd_oti_get(env);
7716         LASSERT(info);
7717
7718         l->ld_ops = &osd_lu_ops;
7719         o->od_dt_dev.dd_ops = &osd_dt_ops;
7720
7721         spin_lock_init(&o->od_osfs_lock);
7722         mutex_init(&o->od_otable_mutex);
7723         INIT_LIST_HEAD(&o->od_orphan_list);
7724         INIT_LIST_HEAD(&o->od_index_backup_list);
7725         INIT_LIST_HEAD(&o->od_index_restore_list);
7726         spin_lock_init(&o->od_lock);
7727         o->od_index_backup_policy = LIBP_NONE;
7728         o->od_t10_type = 0;
7729
7730         o->od_read_cache = 1;
7731         o->od_writethrough_cache = 1;
7732         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7733         o->od_auto_scrub_interval = AS_DEFAULT;
7734
7735         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7736                         sizeof(o->od_svname));
7737         if (cplen >= sizeof(o->od_svname)) {
7738                 rc = -E2BIG;
7739                 GOTO(out, rc);
7740         }
7741
7742         o->od_index_backup_stop = 0;
7743         o->od_index = -1; /* -1 means index is invalid */
7744         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7745         if (rc == LDD_F_SV_TYPE_OST)
7746                 o->od_is_ost = 1;
7747
7748         o->od_full_scrub_ratio = OFSR_DEFAULT;
7749         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7750         rc = osd_mount(env, o, cfg);
7751         if (rc != 0)
7752                 GOTO(out, rc);
7753
7754         rc = osd_obj_map_init(env, o);
7755         if (rc != 0)
7756                 GOTO(out_mnt, rc);
7757
7758         rc = lu_site_init(&o->od_site, l);
7759         if (rc != 0)
7760                 GOTO(out_compat, rc);
7761         o->od_site.ls_bottom_dev = l;
7762
7763         rc = lu_site_init_finish(&o->od_site);
7764         if (rc != 0)
7765                 GOTO(out_site, rc);
7766
7767         INIT_LIST_HEAD(&o->od_ios_list);
7768         /* setup scrub, including OI files initialization */
7769         o->od_in_init = 1;
7770         rc = osd_scrub_setup(env, o);
7771         o->od_in_init = 0;
7772         if (rc < 0)
7773                 GOTO(out_site, rc);
7774
7775         rc = osd_procfs_init(o, o->od_svname);
7776         if (rc != 0) {
7777                 CERROR("%s: can't initialize procfs: rc = %d\n",
7778                        o->od_svname, rc);
7779                 GOTO(out_scrub, rc);
7780         }
7781
7782         LASSERT(l->ld_site->ls_linkage.next != NULL);
7783         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7784
7785         /* initialize quota slave instance */
7786         /* currently it's no need to prepare qsd_instance_md for OST */
7787         if (!o->od_is_ost) {
7788                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7789                                                 &o->od_dt_dev,
7790                                                 o->od_proc_entry, true);
7791                 if (IS_ERR(o->od_quota_slave_md)) {
7792                         rc = PTR_ERR(o->od_quota_slave_md);
7793                         o->od_quota_slave_md = NULL;
7794                         GOTO(out_procfs, rc);
7795                 }
7796         }
7797
7798         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7799                                         o->od_proc_entry, false);
7800
7801         if (IS_ERR(o->od_quota_slave_dt)) {
7802                 if (o->od_quota_slave_md != NULL) {
7803                         qsd_fini(env, o->od_quota_slave_md);
7804                         o->od_quota_slave_md = NULL;
7805                 }
7806
7807                 rc = PTR_ERR(o->od_quota_slave_dt);
7808                 o->od_quota_slave_dt = NULL;
7809                 GOTO(out_procfs, rc);
7810         }
7811
7812         RETURN(0);
7813
7814 out_procfs:
7815         osd_procfs_fini(o);
7816 out_scrub:
7817         osd_scrub_cleanup(env, o);
7818 out_site:
7819         lu_site_fini(&o->od_site);
7820 out_compat:
7821         osd_obj_map_fini(o);
7822 out_mnt:
7823         osd_umount(env, o);
7824 out:
7825         return rc;
7826 }
7827
7828 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7829                                           struct lu_device_type *t,
7830                                           struct lustre_cfg *cfg)
7831 {
7832         struct osd_device *o;
7833         int rc;
7834
7835         OBD_ALLOC_PTR(o);
7836         if (o == NULL)
7837                 return ERR_PTR(-ENOMEM);
7838
7839         rc = dt_device_init(&o->od_dt_dev, t);
7840         if (rc == 0) {
7841                 /*
7842                  * Because the ctx might be revived in dt_device_init,
7843                  * refill the env here
7844                  */
7845                 lu_env_refill((struct lu_env *)env);
7846                 rc = osd_device_init0(env, o, cfg);
7847                 if (rc)
7848                         dt_device_fini(&o->od_dt_dev);
7849         }
7850
7851         if (unlikely(rc != 0))
7852                 OBD_FREE_PTR(o);
7853
7854         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7855 }
7856
7857 static struct lu_device *osd_device_free(const struct lu_env *env,
7858                                          struct lu_device *d)
7859 {
7860         struct osd_device *o = osd_dev(d);
7861
7862         ENTRY;
7863
7864         /* XXX: make osd top device in order to release reference */
7865         d->ld_site->ls_top_dev = d;
7866         lu_site_purge(env, d->ld_site, -1);
7867         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7868                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7869                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7870         }
7871         lu_site_fini(&o->od_site);
7872         dt_device_fini(&o->od_dt_dev);
7873         OBD_FREE_PTR(o);
7874         RETURN(NULL);
7875 }
7876
7877 static int osd_process_config(const struct lu_env *env,
7878                               struct lu_device *d, struct lustre_cfg *cfg)
7879 {
7880         struct osd_device *o = osd_dev(d);
7881         int rc;
7882
7883         ENTRY;
7884
7885         switch (cfg->lcfg_command) {
7886         case LCFG_SETUP:
7887                 rc = osd_mount(env, o, cfg);
7888                 break;
7889         case LCFG_CLEANUP:
7890                 /*
7891                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7892                  * that may happend if hit failure during mount process.
7893                  */
7894                 osd_index_backup(env, o, false);
7895                 lu_dev_del_linkage(d->ld_site, d);
7896                 rc = osd_shutdown(env, o);
7897                 break;
7898         case LCFG_PARAM:
7899                 LASSERT(&o->od_dt_dev);
7900                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7901                                               cfg, &o->od_dt_dev);
7902                 if (rc > 0 || rc == -ENOSYS) {
7903                         rc = class_process_proc_param(PARAM_OST,
7904                                                       lprocfs_osd_obd_vars,
7905                                                       cfg, &o->od_dt_dev);
7906                         if (rc > 0)
7907                                 rc = 0;
7908                 }
7909                 break;
7910         case LCFG_PRE_CLEANUP:
7911                 osd_scrub_stop(o);
7912                 osd_index_backup(env, o,
7913                                  o->od_index_backup_policy != LIBP_NONE);
7914                 rc = 0;
7915                 break;
7916         default:
7917                 rc = -ENOSYS;
7918         }
7919
7920         RETURN(rc);
7921 }
7922
7923 static int osd_recovery_complete(const struct lu_env *env,
7924                                  struct lu_device *d)
7925 {
7926         struct osd_device *osd = osd_dev(d);
7927         int rc = 0;
7928
7929         ENTRY;
7930
7931         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7932                 RETURN(0);
7933
7934         /*
7935          * start qsd instance on recovery completion, this notifies the quota
7936          * slave code that we are about to process new requests now
7937          */
7938         rc = qsd_start(env, osd->od_quota_slave_dt);
7939         if (rc == 0 && osd->od_quota_slave_md != NULL)
7940                 rc = qsd_start(env, osd->od_quota_slave_md);
7941
7942         RETURN(rc);
7943 }
7944
7945 /*
7946  * we use exports to track all osd users
7947  */
7948 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7949                            struct obd_device *obd, struct obd_uuid *cluuid,
7950                            struct obd_connect_data *data, void *localdata)
7951 {
7952         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7953         struct lustre_handle conn;
7954         int rc;
7955
7956         ENTRY;
7957
7958         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7959
7960         rc = class_connect(&conn, obd, cluuid);
7961         if (rc)
7962                 RETURN(rc);
7963
7964         *exp = class_conn2export(&conn);
7965
7966         spin_lock(&osd->od_osfs_lock);
7967         osd->od_connects++;
7968         spin_unlock(&osd->od_osfs_lock);
7969
7970         RETURN(0);
7971 }
7972
7973 /*
7974  * once last export (we don't count self-export) disappeared
7975  * osd can be released
7976  */
7977 static int osd_obd_disconnect(struct obd_export *exp)
7978 {
7979         struct obd_device *obd = exp->exp_obd;
7980         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7981         int rc, release = 0;
7982
7983         ENTRY;
7984
7985         /* Only disconnect the underlying layers on the final disconnect. */
7986         spin_lock(&osd->od_osfs_lock);
7987         osd->od_connects--;
7988         if (osd->od_connects == 0)
7989                 release = 1;
7990         spin_unlock(&osd->od_osfs_lock);
7991
7992         rc = class_disconnect(exp); /* bz 9811 */
7993
7994         if (rc == 0 && release)
7995                 class_manual_cleanup(obd);
7996         RETURN(rc);
7997 }
7998
7999 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8000                        struct lu_device *dev)
8001 {
8002         struct osd_device *osd = osd_dev(dev);
8003         struct lr_server_data *lsd =
8004                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8005         int result = 0;
8006
8007         ENTRY;
8008
8009         if (osd->od_quota_slave_md != NULL) {
8010                 /* set up quota slave objects for inode */
8011                 result = qsd_prepare(env, osd->od_quota_slave_md);
8012                 if (result != 0)
8013                         RETURN(result);
8014         }
8015
8016         if (osd->od_quota_slave_dt != NULL) {
8017                 /* set up quota slave objects for block */
8018                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8019                 if (result != 0)
8020                         RETURN(result);
8021         }
8022
8023
8024         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8025 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8026                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8027                         osd->od_index_in_idif = 1;
8028                 } else {
8029                         osd->od_index_in_idif = 0;
8030                         result = osd_register_proc_index_in_idif(osd);
8031                         if (result != 0)
8032                                 RETURN(result);
8033                 }
8034 #else
8035                 osd->od_index_in_idif = 1;
8036 #endif
8037         }
8038
8039         result = osd_fid_init(env, osd);
8040
8041         RETURN(result);
8042 }
8043
8044 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8045                          struct lu_fid *fid, struct md_op_data *op_data)
8046 {
8047         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8048
8049         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8050 }
8051
8052 static const struct lu_object_operations osd_lu_obj_ops = {
8053         .loo_object_init      = osd_object_init,
8054         .loo_object_delete    = osd_object_delete,
8055         .loo_object_release   = osd_object_release,
8056         .loo_object_free      = osd_object_free,
8057         .loo_object_print     = osd_object_print,
8058         .loo_object_invariant = osd_object_invariant
8059 };
8060
8061 const struct lu_device_operations osd_lu_ops = {
8062         .ldo_object_alloc      = osd_object_alloc,
8063         .ldo_process_config    = osd_process_config,
8064         .ldo_recovery_complete = osd_recovery_complete,
8065         .ldo_prepare           = osd_prepare,
8066 };
8067
8068 static const struct lu_device_type_operations osd_device_type_ops = {
8069         .ldto_init = osd_type_init,
8070         .ldto_fini = osd_type_fini,
8071
8072         .ldto_start = osd_type_start,
8073         .ldto_stop  = osd_type_stop,
8074
8075         .ldto_device_alloc = osd_device_alloc,
8076         .ldto_device_free  = osd_device_free,
8077
8078         .ldto_device_init = osd_device_init,
8079         .ldto_device_fini = osd_device_fini
8080 };
8081
8082 static struct lu_device_type osd_device_type = {
8083         .ldt_tags     = LU_DEVICE_DT,
8084         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8085         .ldt_ops      = &osd_device_type_ops,
8086         .ldt_ctx_tags = LCT_LOCAL,
8087 };
8088
8089 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8090 {
8091         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8092         struct super_block *sb = osd_sb(osd);
8093
8094         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8095 }
8096
8097 /*
8098  * lprocfs legacy support.
8099  */
8100 static struct obd_ops osd_obd_device_ops = {
8101         .o_owner = THIS_MODULE,
8102         .o_connect      = osd_obd_connect,
8103         .o_disconnect   = osd_obd_disconnect,
8104         .o_fid_alloc    = osd_fid_alloc,
8105         .o_health_check = osd_health_check,
8106 };
8107
8108 static int __init osd_init(void)
8109 {
8110         int rc;
8111
8112         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8113 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8114         /* please, try to keep osd_thread_info smaller than a page */
8115         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8116 #endif
8117
8118         osd_oi_mod_init();
8119
8120         rc = lu_kmem_init(ldiskfs_caches);
8121         if (rc)
8122                 return rc;
8123
8124 #ifdef CONFIG_KALLSYMS
8125         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8126         priv_dev_check_rdonly =
8127                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8128 #endif
8129
8130         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8131                                  lprocfs_osd_module_vars,
8132                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8133         if (rc)
8134                 lu_kmem_fini(ldiskfs_caches);
8135         return rc;
8136 }
8137
8138 static void __exit osd_exit(void)
8139 {
8140         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8141         lu_kmem_fini(ldiskfs_caches);
8142 }
8143
8144 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8145 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8146 MODULE_VERSION(LUSTRE_VERSION_STRING);
8147 MODULE_LICENSE("GPL");
8148
8149 module_init(osd_init);
8150 module_exit(osd_exit);